Abstract

Magnetic resonance imaging (MRI) plays a critical role in diagnosis and staging of various pathologies. Traditional MRI techniques rely on a set of contrast-weighted images, which limit their capability to provide a more comprehensive assessment for clinical diagnosis and treatment monitoring. Magnetic Resonance Fingerprinting (MRF) is an advanced quantitative MRI method, which can provide rapid and accurate quantification of multiple tissue properties. My research focuses on developing and translating the MRF technique for body imaging. The developed methods have demonstrated superior performance in both repeatability and reproducibility compared to existing quantitative MRI methods. In this presentation, I will introduce recent technical advances in MRF, followed by findings from multiple clinical studies.

Abstract

This presentation explores the growing role of AI in biomedical research, focusing on two innovative advancements. First, GPT-RadPlan, an AI-driven framework, fully automates radiotherapy treatment planning, consistently outperforming traditional methods by leveraging large multimodal language models like GPT-4V. Second, we introduce TextGrad, a framework that optimizes AI agentic systems using natural language feedback from large language models, applicable across various domains, including biomedical research. Together, these tools demonstrate the potential of AI to revolutionize clinical decision-making and advance the field of biomedical research.

Abstract

There is a regulatory code written in DNA and RNA sequences that controls gene expression and isoform processing. Developing accurate models of this code is crucial for advancing human health – such models allow us to interpret harmful genetic mutations and design improved regulatory sequences. In this talk, I will first give an overview of existing machine learning methods for sequence-based prediction of regulatory functions (transcription, RNA splicing, etc.). I will then present a unified model of gene regulation that directly learns to predict raw RNA expression profiles from DNA sequence alone. In the second part of the talk, I will discuss methods for designing improved regulatory sequences and highlight their potential for molecular therapies. I will conclude by discussing future opportunities and challenges in developing even better models of gene expression.

Abstract

Acoustic noise is well-acknowledged as a major source of patient discomfort during MRI scans, and quiet scans have been shown to improve exam success rates. In this work, we demonstrate the feasibility of quiet DCE-MRI using zero echo-time (ZTE) imaging, which offers a unique combination of high scan efficiency and low acoustic noise. Specifically, we address two main challenges of using ZTE for DCE applications: (1) dominant PD-weighting and (2) poor temporal k-space sampling. To overcome these limitations, we first increase T1-weighting using phase-modulated RF pulses and a corresponding inverse problem-based reconstruction, which enables quiet post-contrast imaging. Next, we improve temporal k-space sampling using Arc-ZTE, a method that continuously slews the gradients to achieve improved k-space coverage with minimal gradient refocusing. Arc-ZTE allows for flexible and quiet dynamic imaging, which we demonstrate by visualizing contrast uptake in a custom-built phantom. Overall, these advancements enable significantly improved patient cooperation and comfort, especially for neonates and young pediatric patients.

Abstract

Cortical folding is a fundamental process in human brain development, driven by the complex interplay between mechanical forces, tissue growth, and connectivity establishment. As the cerebral cortex expands significantly in volume and surface area, it undergoes extensive gyrification, which is essential for efficient neural wiring and cognitive function. Disruptions in this process have been implicated in neurodevelopmental disorders, including autism spectrum disorder (ASD), epilepsy, and schizophrenia, where abnormal folding patterns are often accompanied by structural connectivity impairments. Despite growing evidence linking cortical morphology to brain connectivity, the underlying mechanisms governing this relationship remain largely unknown.

In this talk, I will present our latest findings on the multiscale mechanics of brain growth, folding, and connectivity development. We will explore how mechanical forces shape cortical morphology and contribute to the organization of neural circuits, as well as how deviations from these processes may lead to pathological conditions. Using an integrated approach that combines neuroimaging, multiscale computational modeling, and deep learning techniques, we aim to uncover the fundamental principles governing cortical folding and its role in healthy and disordered brain development. Furthermore, I will discuss our recent insights into the structure-property relationships of brain tissue from the microscale to the macroscale, shedding light on how mechanical properties influence cortical architecture and connectivity. These findings provide a critical foundation for understanding the origins of abnormal development associated with neurodevelopmental disorders.

Abstract

Dr. Joshua Siegel’s work has focused on developing precision functional mapping (PFM) and applying it to understand the mechanism of psychoactive drugs. Part 1 will present a recently published study showing that psilocybin, the psychedelic found in magic mushrooms, drives widespread desynchronization of brain activity across spatial scales (local, areal, global). Part 2 will discuss implementation of advanced fMRI and PFM tools at NYU Langone, including some sequence testing and pilot data.

Abstract

The incidence of loneliness has increased over the past several decades worldwide and is particularly common among people with serious mental illnesses. However, this major public health problem has been difficult to address, in part because the neural and cognitive mechanisms underlying loneliness are poorly understood. This talk will discuss research on the neurobiological correlates of loneliness and isolation in those with psychotic spectrum disorders using ASL and fMRI techniques alongside peripheral biomarkers in two independent samples.

Abstract

This talk will provide an overview of the Free.Max 0.55T scanner that was recently installed at the 22nd Street Gramercy imaging facility. Details of the system with a focus on available hardware and software features will be presented. This will include information on product pulse sequences and supported WIP packages along with examples of various imaging applications.

Abstract

Hepatic encephalopathy is a severe complication of chronic or acute liver disease causing toxin accumulation in the brain, but its structural and neurometabolic consequences are still poorly understood. Following a general introduction about MR spectroscopy (MRS) and the current clinical trends of the field, I will present the two main results of my PhD work. In a rat model of hepatic encephalopathy, we measured: (1) altered cerebellar microstructure using diffusion-weighted MRS and diffusion MRI and (2) down-regulated glucose metabolism using FDG-PET and downfield 1H MRS at 14.1 T. Methodological developments regarding MR sequence design and PET quantification that enabled these findings will be described alongside. Finally, our first implementation of diffusion MRS at CBI will be presented, comparing the “stick compartment” in human brain as derived from diffusion MRS and diffusion MRI.

Abstract

We consider the diffusion-weighted signal at strong magnetic field gradients of a finite duration. In this so-called localization regime, the magnetization is dephased everywhere except near the boundaries, making the signal sensitive to the microstructure such as cell walls. We review conventional approaches to diffusion MRI, including the cumulant expansion and diffusion-diffraction, discuss why they fail in the localization regime, and develop the physical intuition for this unconventional case that can be reached on high-performance human and preclinical scanners. Technically, we reveal how the characteristic stretched-exponential signal decay emerges at strong gradients from spectral properties of the underlying Bloch-Torrey operator. The recent theoretical and experimental advances will be discussed from the point of revising some common beliefs and practices in the field.

Abstract

Computed tomography (CT) enables noninvasive in vivo visualization of tissues’ anatomy and functionality. X-rays with a specific energy interact with tissue and generate a particular image contrast relative to that of water. Thus, X-ray energy spectrum plays an important role in CT imaging. Because the lower-energy X-rays are absorbed more readily by body tissues and do not contribute to image formation, CT imaging uses the higher-energy portion of the spectrum. Thus, CT image reconstruction faces an intrinsically contrast-limited challenge. While injection of contrast media could enhance CT images, it is not the topic of this talk. This presentation will focus on contrast enhancement (equivalent to energy-resolved CT at low energy levels) via spectral CT (dual-energy or photon-counting CT) reconstruction, which is essentially low-dose CT reconstruction. Bayesian inference for low-dose/spectral CT reconstruction aims to incorporate prior knowledge into fully data-driven maximum-likelihood solution, where prior knowledge about the body tissues can be learned from available diagnostic CT images. This presentation will further explore the relationship between the enhanced image contrasts and in vivo tissue biological characteristics for machine learning diagnosis and early detection of cancers.

Abstract

The NIH K99/R00 Pathway to Independence Award provides a unique opportunity for postdoctoral researchers to secure funding and transition to independent faculty positions. However, the application process can be challenging, with specific requirements, competitive standards, and strategies that can make or break your application. This talk will cover the basics of the K99/R00 program, and how it supports the transition from postdoc to faculty, as well as key components of the application and tips for writing. Whether you’re actively preparing a K99/R00 application or just exploring your funding options, this talk will provide valuable insights and practical tips to demystify this career award.

Abstract

Gene-expression reporter systems, such as green fluorescent protein, have been instrumental to understanding biological processes in living organisms at organ system, tissue, cell, and molecular scales. More than 30 years of work on developing MRI-visible gene-expression reporter systems has resulted in a variety of clever application-specific methods. However, these techniques have not yet been widely adopted, so a general-purpose expression reporter is still required. In this talk, we will demonstrate that the manganese ion transporter Zip14 is an in vivo MRI-visible, flexible, and robust gene-expression reporter to meet this need.

Abstract

Functional MRI can provide in vivo biomarkers that facilitate the differential diagnosis of breast cancer and the assessment of response to therapy. Non-invasive imaging techniques such as diffusion MRI can provide useful information about tumour cellularity, however image quality is limited due to image acquisition methods. Advanced acquisition strategies (such as multiplexed sensitivity encoding, or MUSE) can overcome these limitations. Dynamic contrast-enhanced MRI (DCE-MRI) can provide information about tumour vascularity and has been used in the prediction of pathological complete response to neoadjuvant chemotherapy, particularly through the use of radiomics and deep learning approaches.

While lung function tests are the gold standard for the assessment of progression in lung diseases, they suffer from measurement variability. CT-derived imaging biomarkers can provide objective structural measurements of disease that when used in combination with functional measurements can predict progression and measure response to treatment. Developments in deep learning segmentation algorithms have allowed for fast and accurate automated segmentation of tissues and structures on lung CT scans.

About the Speaker

Gabrielle Baxter completed her PhD and first post-doctorate in the department of radiology at the University of Cambridge, investigating the use of diffusion MRI, dynamic contrast-enhanced MRI, and ²³Na-MRI in breast cancer. Most recently, she was a research fellow at the Centre for Medical Image Computing in the department of computer science at University College London, where her work focused on the development of CT image-derived biomarkers of lung disease progression and the use of deep learning algorithms for the segmentation of airways, vessels, fat, and muscle.

Abstract

The hippocampus is a widely studied yet enigmatic archicortical region which serves multiple cognitive functions. Part of its mystery arises from the difficulty in studying its structure non-invasively. Irrespective of any hypothesis, past and present research has generally focused on the macro level of hippocampal structure, including volume and thickness. While sufficient for some questions, such measures are coarse and generally immutable to different properties of the intrahippocampal gray matter. This includes components such as glial cells, neurites, soma and other micron-scale structures (i.e. microstructure). Such microstructure is responsible for the computations which engender hippocampal function and are of critical importance in both health and disease. Diffusion MRI (dMRI) is one technique which can provide sensitivity to micron-scale structures non-invasively. This talk will present the current landscape, difficulties, and future perspectives of dMRI applied to the hippocampus to understand its microstructure in health, disease, and development. Overall, it looks to answer three questions: why care, what we currently know, and what we hope to know in the future.

About the Speaker

Bradley Karat is a PhD candidate in neuroscience at the University of Western Ontario. He currently studies all things MRI and hippocampus, with a particular emphasis on understanding the micron-scale structures of the hippocampus non-invasively including glia, neurites, and soma. In particular, his work focuses on applying/developing novel diffusion MRI methods to improve characterization of hippocampal microstructure in health, disease, and development. With such microstructure characterization, Bradley hopes to gain insight into the eclectic function of the hippocampus and its substructures.

Abstract

Myelin is one of the key constitutes of the central nervous system and is involved in numerous developmental and neuropathological processes. Noninvasive assessment of myelin content and integrity, however, is highly challenging with no gold standard available to date. Multi-compartment (mc) analysis of MRI signals, and specifically T2 relaxation times (mcT2), is the most common and efficient approach for quantifying myelin in vivo. The approach is based on separating the signal within each voxel into a series of signals, each originating from a distinct cellular compartment. The fast-relaxing T2 component is, in this case, associated with water residing between myelin sheaths, and provide an indirect measure of myelin content. Notwithstanding its popularity, this approach is highly ill-posed due to large ambiguities in the multi-T2 space and the low SNR that characterizes MRI signals. In this talk I will present a new data-driven approach to mcT2 analysis, which harnesses information from the entire white matter and the power of statistics to identify tissue-specific mcT2 motifs, prior to deconvolving the local signal at each voxel. This stabilizes the process of myelin quantification, and can improve the analysis of microstructural tissue compartmentation in general. Validations will be presented using computer simulations, a unique multicomponent phantom design, mice models of demyelination, as well as healthy subjects and people with multiple sclerosis.

Abstract

Microstructure imaging faces several challenges in neuroscientific and clinical applications, including long acquisition and computation times, degeneracies, and biases in parameter estimation. In this talk, I will present some of our efforts aimed at addressing these challenges. First, I will present how we use patch-based convolutional neural networks to accelerate data acquisition, focusing on the estimation of fiber orientation and microstructural properties. Next, I will demonstrate combined diffusion-relaxometry methods for more specific quantification of brain tissue microstructure and composition. I will introduce Microstructure.jl, an open-source toolbox I am actively developing in the Julia language, which unifies these methods into a cohesive framework for parameter estimation and uncertainty quantification across various biophysical models. Finally, I will showcase how these advanced techniques can capture the dynamic microstructural changes during brain development and enhance our understanding of the developing brain.

Abstract

Diffusion MRI is an important imaging tool for studying brain microstructure by detecting the random Brownian motion of water molecules. However, current diffusion MRI methods are confounded by several technical challenges for precise detection of brain microstructure. For example, the widely used diffusion MRI acquisition method, single-shot echo planar imaging (ss-EPI), is known to have a typically low resolution, relatively low SNR, and geometric distortions. The basic diffusion MRI sequence, pulsed gradient spin echo (PGSE), may only provide limited specificity for brain microstructure detection. In this presentation, I will first review how diffusion MRI can be used to detect brain microstructure. Then I will introduce our recent work on advanced diffusion encoding and acquisition and how they affect brain microstructure mapping. Finally, I will discuss further technical development and potential clinical translation of new diffusion MRI techniques.

Abstract

This presentation reports preliminary results from an on-going NIH R01 research project related to sodium MRI and Alzheimer’s disease (AD). Cerebrospinal fluid (CSF) clearance pathway (e.g., glymphatic system) might be disrupted in AD. Recent studies on mice showed impairment of CSF clearance pathway that led to a 70% decrease in amyloid beta (Aβ) clearance but sleep enhanced CSF flow and increased A𝛽 clearance by 100%. However, it is unknown whether these negative and positive impacts exist in humans due to lack of adequate non-invasive techniques for the study. In this project we proposed two unique techniques, i.e., dynamic sodium MRI and ultrashort echo time (UTE) proton MRI, to determine whether CSF clearance is enhanced during sleep, degenerated in aging, and disrupted in AD. Instead of studying perivascular space—a popular research-targeting region, this project was able to investigate full life cycle of CSF simultaneously, including the production at choroid plexus, the bulk flow in brain parenchyma, and the drainage at arachnoid villi. The overarching goal is to understand the changes of CSF clearance in aging and in AD. The outcomes will generate highly-desired knowledge about degeneration of CSF clearance and help develop effective interventions to AD.

Abstract

STAGE is a rapid multi-contrast imaging technique that can provide not only standard T1 and T2 like contrasts but also proton spin density weighted images, T2* weighted images, SWI, R2* maps, QSM maps, water content maps, circle of Willis MRA, auto segmentation of white matter, gray matter and cerebrospinal fluid and simulated FLAIR in 3 to 6 minutes. Coupled with a new denoising algorithm called CROWN, STAGE can be used to create a new series of homogeneous images that have been corrected for both radiofrequency transmit and receive inhomogeneities. During the last few years, we have focused on measuring iron content and neuromelanin (NM) in the substantia nigra (SN) for comparing idiopathic Parkinson’s disease (PD) with healthy controls and patients with other movement disorders. We have found that the volume of NM, the iron content of the SN, volume of the SN and the N1 sign can together provide an area under the curve (AUC) of 95% in distinguishing PD from healthy controls. We have developed a template of the midbrain to allow for automatic detection and quantification of these properties. During the process, we used tSWI to enhance the N1 sign visibility. Our data have been acquired using STAGE which is a rapid, quantitative, multi-contrast data collection and processing that is vendor agnostic. As such we have created a protocol that can be used for PD studies globally.

More recently, we have been using fast low flip angle multi-echo (FLAME) spin density weighted imaging to map the white matter fiber tracts from the brainstem up to the thalamus in vivo. Although tractography from diffusion tensor imaging has been commonly used to map the WM fiber tracts connectivity, it is difficult to differentiate the complex WM tracts anatomically. With a clear delineation of major fiber tracts, it may be possible to use structurally constrained DTI fiber tracking to improve their visualization. Using the FLAME approach, we have been able to map out all the major fiber tracts in a reasonably short scan time of 10 minutes with 0.67×0.67×1.34 mm³ resolution at 3T.

Abstract

Recognition is a fundamental cognitive function during which the brain projects perceptual templates learnt from past experiences onto the current sensory input. A key factor in this process is how much the brain weights prior knowledge versus the sensory input. This weighting is known to vary between individuals and may contribute to hallucinations in mental illnesses when prior knowledge is weighted to a pathologically high degree. By manipulating human subjects’ ability to use prior knowledge to recognize stimuli during fMRI and intracranial EEG neural recording, we can examine the underlying neural mechanisms of prior knowledge deployment from distinct sources of prior knowledge. By training artificial neural networks to perform similar image recognition tasks, we can identify plausible computations the human brain may use to achieve the same. Specifically, this project examines visual recognition aided by prior knowledge derived from lifelong learning as well as one-shot learning. This talk will present early results and plans to integrate data from multiple experiments.

About the Speaker

Jonathan Shor is a sixth-year PhD student advised by Biyu He, associate professor of neurology, neuroscience and physiology, and radiology. After receiving his BS in computer science, Jonathan spent 10 years in the private sector before returning to academia, earning an MS in computer science at Columbia University before joining the Vilcek Institute. He is interested in using neuroimaging and computational techniques to identify the neural correlates of conscious perception.

Abstract

The visual system develops abnormally when visual input is absent or degraded early in life. Restoration of the visual input past 7 or 8 years of age is generally thought to have limited benefit because the visual system will lack sufficient plasticity to adapt to and utilize the information from the eyes. Recent evidence, however, shows that congenitally blind adolescents can recover both low-level and higher-level visual function following surgery. In this talk, I will discuss recent work in our lab that links changes in behavioral performance to longitudinal changes in white matter integrity in congenitally blind patients with dense bilateral cataracts. Our results suggest that sufficient plasticity remains in adolescence to partially overcome abnormal visual development and help localize the sites of neural change underlying sight recovery.

Abstract

Multiple sclerosis (MS) is a neuroimmune disorder characterized by demyelination and neurodegeneration, leading to cognitive and motor impairments. This study investigates the effects of two MS-modifying medications, interferon beta and dimethyl fumarate, on functional brain connectivity using graph theoretical network analysis. We recruited 107 relapsing-remitting MS (RRMS) patients and 62 healthy controls, collecting resting-state functional MRI data to construct brain networks based on regional blood-oxygen-level-dependent signals. The study consisted of two parts: region-based and connection-based analysis. In the region-based analysis, we assessed graph theoretical measures, including degree centrality, nodal efficiency, and betweenness centrality, to evaluate alterations in brain networks and to explore hubs as highly connected regions in healthy controls and patient groups under different medications. The connection-based analysis employed an edge-centric approach to examine metrics such as entropy, co-fluctuation, and edge strength. Results from the region-based analysis showed differences in network metrics, particularly in the default mode, dorsal attention, and salience networks, among patients on different medications. The connection-based analysis revealed changes in motor, sensory, visual, and attention networks, alongside variations in the brain network strength. Entropy and changes in edge functional connectivity (eFC) revealed distinct effects of each medication. Interferon beta caused hemisphere-wide alterations in the somatomotor network, with opposite changes in each hemisphere. Conversely, dimethyl fumarate’s effect was marked by decreased entropy in the left visual and right dorsal attention networks, indicating reduced diversity in network connectivity. The study highlights potential biomarkers for the treatment effect and demonstrates the nuanced effects of interferon beta and dimethyl fumarate on brain function in patients with MS.

About the Speaker

Sara Hejazi is a brain network scientist with a diverse background in engineering and neuroscience. She received her PhD from the engineering college at the University of Central Florida in May 2024. Her doctoral research utilized brain network analysis to investigate changes in brain connectivity in patients with multiple sclerosis (MS) using fMRI data. During her internship at the Mayo Clinic’s department of artificial intelligence and informatics from summer 2022 to spring 2023, she contributed to a project involving network analysis aimed at studying the development of mild cognitive impairment (MCI) and its progression to dementia. In the summer of 2023, she joined the University of Florida College of Medicine to further expand her expertise in brain network analysis in preclinical research, specifically focusing on traumatic brain injury (TBI).

Abstract

Epileptogenic foci often organize into networks, making the diagnosis of focal epilepsy challenging due to the spontaneous nature of the disease. While triggering seizures with intra-cranial stimulation aids in diagnosis, concerns about invasiveness drive the research for non-invasive alternatives. In this talk, Chloé will highlight the potential of temporal interference (TI) stimulation in diagnosing refractory epilepsy by identifying epileptogenic zone networks (EZN) in silico. Crucial model parameters, linked to seizure onset, are inferred from patient-specific brain models to quantify brain regions’ epileptogenicity and diagnose the EZN.

About the Speaker

Chloé Duprat is a PhD candidate at the Institut des Neurosciences des Systèmes at Aix Marseille Université. She holds a master’s degree in mechanical and biomedical engineering and in computational neuroscience and neuro-engineering. Her work primarily focuses on dynamic modeling and neural stimulation modeling in patient-specific whole-brain models.

Abstract

T1ρ mapping is emerging as a promising contrast-free alternative to LGE imaging for myocardial tissue characterization. However, the high sensitivity of conventional spin-lock preparations to B0 and B1+ inhomogeneities renders it ineffective at high field strengths. Adiabatic spin-lock preparations, consisting of trains of hyperbolic-secant pulses, can overcome this limitation. During this talk I will discuss how we deigned and optimized adiabatic T1ρ preparations for in vivo mapping at 3T. I will compare the results of adiabatic T1ρ mapping with conventional non-adiabatic techniques and show how adiabatic spin-locks yielded improved precision and reproducibility. Results from a small cohort of patients indicated that adiabatic T1ρ mapping could potentially be used to discriminate between scar and healthy myocardium. Finally, I will discuss how we used a combination of selective and non-selective adiabatic RF pulses to achieve robust adiabatic T1ρ maps with dark-blood contrast.

About the Speaker

Chiara holds a BSc in biomedical engineering from Politecnico di Milano and a double MSc in biomedical engineering from Politecnico di Milano and Politecnico di Torino, with honors. She completed her master’s studies with a thesis project on image filtering and compression for super high frame rate electron microscopy at Lawrence Berkeley National Laboratory. She is currently a PhD candidate at the Magnetic Resonance Systems Lab in the department of imaging physics at TU Delft and expects to graduate in September. During her PhD Chiara has worked on designing novel RF preparations for robust spin-lock imaging at 3T, with a special focus on cardiac MR applications.

Abstract

Accurate interpretation of medical images is crucial for disease diagnosis and treatment, and AI has the potential to minimize errors, reduce delays, and improve accessibility. The focal point of this presentation lies in a grand ambition: the development of ‘Generalist Medical AI’ systems that can closely resemble doctors in their ability to reason through a wide range of medical tasks, incorporate multiple data modalities, and communicate in natural language. Starting with pioneering algorithms that have already demonstrated their potential in diagnosing diseases from chest X-rays or electrocardiograms, matching the proficiency of expert radiologists and cardiologists, I will delve into the core challenges and advancements in the field. The discussion will navigate towards the topic of label-efficient AI models: with a scarcity of meticulously annotated data in healthcare, the development of AI systems capable of learning effectively from limited labels has become a key concern. In this vein, I’ll delve into how the innovative use of self-supervision and pre-training methods has led to algorithmic advancements that can perform high-level diagnostic tasks using significantly less annotated data. Additionally, I will talk about initiatives in data curation, human-AI collaboration, and the creation of open benchmarks to evaluate the generalizability of medical AI algorithms. In sum, this talk aims to deliver a comprehensive picture of the state of ‘Generalist Medical AI,’ the advancements made, the challenges faced, and the prospects lying ahead.

Abstract

Contrast-enhanced T1-weighted imaging is an important component of many clinical protocols, such as for the detection of brain tumors and metastases. A widely used acquisition scheme is the magnetization-prepared rapid gradient-echo (MP-RAGE) sequence, which depicts both enhancing lesions and blood vessels with bright signal intensity. As an alternative, a novel family of sequences has been recently proposed, named T1 Relaxation-Enhanced Steady-State (T1RESS), which suppress the contrast of background tissue and signal of flowing blood, resulting in improved conspicuity of enhancing lesions. The improved sensitivity is especially valuable for detecting small lesions, which can have a major impact on the prognosis of a patient. In this work, radial stack-of-stars T1RESS sequences were developed, which offer improved motion robustness and enable advanced reconstructions such as GRASP for dynamic imaging. Three sequence variants were introduced, including (a) balanced T1RESS, which can potentially be used at low field (0.55T) for high-SNR T1-weighted imaging; (b) unbalanced T1RESS-FISP, which suppresses the signal of blood vessels, and (c) unbalanced T1RESS-PSIF, which provides dark-blood imaging. To broaden the clinical applicability, the combination with GRASP reconstruction, DIXON fat/water separation, and 1D GRAPPA acceleration in the Cartesian direction of the stack-of-stars geometry were implemented. The radial T1RESS sequences have been tested for brain imaging and applications in the body in healthy volunteers and patients.

About the Speaker

Ruoxun Zi is a fourth-year graduate student in the Vilcek Institute’s Biomedical Imaging and Technology training program working with Kai Tobias Block, PhD, and Riccardo Lattanzi, PhD. She holds an MS in biomedical engineering from Johns Hopkins University.

Abstract

The study of hyperuniform states of matter is an emerging multidisciplinary field, influencing and linking developments across the physical sciences, mathematics and biology [1,2]. Hyperuniform many-particle systems in d-dimensional Euclidean space are characterized by an unusual suppression of density fluctuations at large lengths scales.

The hyperuniformity concept generalizes the traditional notion of long-range order, and provides a unified theoretical framework to categorize crystals, quasicrystals and exotic disordered systems. Disordered hyperuniform many-particle systems can be regarded to be new states of disordered matter in that they behave more like crystals or quasicrystals in the manner in which they suppress large-scale density fluctuations, and yet are also like liquids and glasses because they are statistically isotropic structures with no Bragg peaks.

I will provide an overview of the hyperuniformity concept and its generalizations. Subsequently, I will discuss the diffusion spreadability S(t), which is a measure of the spreadability of diffusion information as a function of time.

Exact formulas for the spreadability in any Euclidean space dimension are derived in terms of two-point statistics that characterize the microstructure.

Further, closed-form general formulas are derived for the short- , intermediate- and long-time behaviors of S(t) in terms of crucial small-, intermediate- and large-scale structural information, respectively. The long-time behavior of S(t) enables one to distinguish the entire spectrum of translationally invariant microstructures that span from hyperuniform to nonhyperuniform media.

For hyperuniform media, disordered or not, the “excess spreadability”, S(∞) – S(t), decays to its long-time behavior exponentially faster than that of any nonhyperuniform medium, the slowest being “antihyperuniform media”. The stealthy hyperuniform class is described by an excess spreadability with the fastest decay rate among all translationally invariant structures.

We establish remarkable connections between the spreadability and problems in discrete geometry, NMR pulsed field gradient spin-echo amplitude as well as diffusion MRI measurements.

References

1. Torquato S and Stillinger FH. Local density fluctuations, hyperuniformity, and order metrics. Phys Rev E Stat Nonlin Soft Matter Phys. 2003 Oct;68(4 Pt 1):041113. doi: 10.1103/PhysRevE.68.041113.
2. Torquato S. Hyperuniform States of Matter. Phys Reports. 2018 Jun;745:1-96. doi:
10.1016/j.physrep.2018.03.0012.
3. Torquato S. Diffusion Spreadability as a Probe of the Microstructure of Complex Media Across Length Scales. Phys Rev E. 2021 Nov;104(5-1):054102. doi: 10.1103/PhysRevE.104.054102.

Abstract

Breast cancer screening, primarily conducted through mammography, is often supplemented with ultrasound for women with dense breast tissue. However, existing deep learning models analyze each modality independently, missing opportunities to integrate information across imaging modalities and time. In this study, we present Multi-modal Transformer (MMT), a neural network that utilizes mammography and ultrasound synergistically, to identify patients who currently have cancer and estimate the risk of future cancer for patients who are currently cancer-free. MMT aggregates multi-modal data through self-attention and tracks temporal tissue changes by comparing current exams to prior imaging. Trained on 1.3 million exams, MMT achieves an AUROC of 0.943 in detecting existing cancers, surpassing strong uni-modal baselines. For 5-year risk prediction, MMT attains an AUROC of 0.826, outperforming prior mammography-based risk models. Our research highlights the value of multi-modal and longitudinal imaging in cancer diagnosis and risk stratification.

Abstract

Inflammation of the central nervous system (CNS), driven by the innate immune system, plays a crucial role in the pathophysiology of multiple sclerosis (MS). Critical cell types involved in this process include CNS resident microglia and blood-derived macrophages. Imaging methods such as MRI and PET can be used to assess CNS inflammation in MS, and we will discuss our work focusing on these two approaches.

In measuring innate immune activity via MRI, we observe that chronic CNS inflammation in MS lesions is maintained in part by iron-laden pro-inflammatory microglia/macrophages at the rim of chronic active lesions (CALs). These paramagnetic rim lesions (PRLs) are believed to contribute to a more aggressive phenotype of the disease and thus represent a novel target for treatment to reduce disease progression in MS. Utilizing quantitative susceptibility mapping (QSM) to measure PRLs, our group has concentrated on creating tools to identify and quantify lesion-based chronic innate immune activity. We have been investigating the impact of chronic lesion-based inflammation on the disease course. Moreover, we have expanded our work to propose a novel application of QSM to quantify the inflammatory trajectory within PRLs, providing a new target for treatment in MS with either current or novel immune modulators.

Regarding the measurement of chronic inflammation via PET, [11C]-PK11195-PET (PK-PET) is a first-generation ligand that binds to the 18 kDa translocator protein (TSPO), expressed on the outer mitochondrial membrane of activated myeloid cells, and it has been used to demonstrate increased innate immune activity throughout the brains of MS patients. We have employed PK-PET to validate in vivo that PRLs on QSM exhibit higher inflammation than rimless lesions. Newer generation TSPO ligands, with enhanced specificity and brain penetration compared to PK-PET, which may enhance PET’s ability to identify and compare inflammatory activity across individual chronic lesions. We are currently focusing our research on a second-generation TSPO ligand, [11C]-DPA 713. To circumvent the challenges of arterial sampling, we have developed a supervised clustering algorithm (SVCA). This algorithm serves as a reliable non-invasive method for quantifying this ligand’s presence. Additionally, we are applying this innovative technique to investigate the dynamics of individual MS lesions.

Abstract

Cerebral cortical neural architecture is an important biomarker for prognosis and diagnosis of a variety of brain disorders. At present, assessing soma and neurite (axons and dendrites projecting from cell body) architecture in the cerebral cortex requires neuropathology. However, neuropathology is invasive and provides only local neuronal architecture information. Diffusion MRI (dMRI) has been the method of choice for noninvasively measuring brain microstructure. Cerebral cortical architecture is complicated consisting of somas of neurons and glia as well as neurites. In this talk, I will introduce our efforts towards developing dMRI-based techniques to accurately and reproducibly measure both soma and neurite densities in complicated cerebral cortical architecture by harnessing both a “top-down” and a “bottom-up” approach. Using the “top-down” approach, we modeled the complex cellular architecture of the cerebral cortex by establishing the non-Gaussian dMRI signal modeling for measuring the cortical micro-architecture based on theoretical modelling and diffusion simulations. In the “bottom-up” approach, we quantified the cortical micro-architecture using deep learning by mapping dMRI signal to ground-truth.

Abstract

Early detection and treatment of Alzheimer’s disease (AD) and related disorders is paramount in the prevention of neuronal degeneration and dementia. AD is characterized by a long preclinical stage noted by the deposition of amyloid plaques and tau tangles, followed by a destructive cascade of neurodegeneration typically beginning in the medial temporal lobe, leading to mild cognitive impairment and progression to dementia. However, only 40% of patients with preclinical AD progress to mild cognitive impairment, and only 21% of patients with mild cognitive impairment progress to dementia. Further, current procedures for detecting amyloid and tau deposition are either biofluid (CSF and blood) biomarkers that lack spatial information thus limiting their utility in prognosis, or PET scans which are prohibitively expensive and involve IV injection of radiotracers. Thus, there is a need to develop more accessible, spatially specific biomarkers and to differentiate patients with amyloid and tau depositions who will progress to dementia from those who will not. A promising avenue for early detection of neurodegeneration is subjective cognitive decline (SCD). SCD describes patients who complain of memory difficulties but score normally on cognitive testing. Multicenter studies and meta-analyses show increased rates of dementia in participants with SCD. Many previous studies in SCD have focused on volume and cortical thickness, with inconsistent results. Multimodal MRI provides an opportunity to develop biologically sensitive biomarkers that are less expensive than PET, do not involve IV injection of radiotracers, and provide in vivo spatial information of pathology deposition. While amyloid and tau deposits are necessary for the diagnosis of AD pathology, they are not sufficient for clinical prognosis. Further, their role in the biological pathway leading to the neurodegenerative cascade is unclear. Amyloid in particular correlates poorly with clinical symptoms. Scientific consensus at the 2022 Clinical Trials on Alzheimer’s Disease Conference concluded that it is important to explore other potential therapeutic targets alongside amyloid and tau, although the lack of biomarkers pose a significant barrier. While prior DTI research in SCD is promising for prognostication, studies in the medial temporal lobe and cortical gray matter have been limited, despite their importance in Alzheimer’s disease. Here, we use advanced diffusion MRI preprocessing and modeling to study the microstructure of the medial temporal lobe and cortical gray matter of this group with heightened dementia risk. We present several new potential diffusion MRI biomarkers of SCD.

Abstract

The advent of deep neural networks (DNNs) has marked considerable breakthroughs in magnetic resonance imaging (MRI) analysis. These state-of-the-art methodologies are increasingly deployed to resolve intricate predicaments in quantitative MRI such as diffusion-weighted MRI, and quantitative cardiac T1 and T2 distribution mapping, delivering superior speed and precision over conventional techniques. Nonetheless, existing challenges such as mitigation of motion artifacts and enhancement of resilience against extremely low signal-to-noise ratios still remain, restricting their clinical utility. To overcome these limitations, we introduce an innovative strategy integrating a physically-primed DNN architecture. This unique architecture embeds the signal decay model directly within the neural network, augmenting the network’s generalization capability and fostering the development of stable algorithms, which, in turn, produce refined predictions. Our advanced methodology reveals extensive potential applications including early assessment of response to neoadjuvant chemotherapy in breast cancer patients, establishing motion-robust quantitative cardiac T1 mapping, and T2 distribution mapping for evaluating inflammation in animal models. The proposed approach opens new vistas for more nuanced and clinically viable solutions in the realm of quantitative MRI analysis, paving the way for enhanced diagnostic precision and patient outcomes.

Abstract

Astrocytes are non-neuronal glial cells in the central nervous system with myriad functions: they maintain the blood brain barrier, recycle neurotransmitters, and are even thought to be a component of every neuronal synapse. We have found that astrocytes also form gap junctional networks that help neurons survive degeneration. Brain regions connected by astrocytes support one another metabolically, but these connections also make linked regions collectively vulnerable to degenerative stress. Astrocyte networks have never been mapped, a critical need considering that these networks may shape the course of degeneration. Here, we use a combination of novel biomolecular tools, tissue clearing, and light sheet imaging to reveal for the first time the shape, extent, and potential plasticity of astrocytic networks in intact brains.

Abstract

The goal of the MRI4ALL Hackathon 2023, which took place last October around the i2i Workshop, was to create a fully-fledged low-field MRI scanner in just four days and to share all developments as open-source resources. Fifty two scientists from 16 institutions participated in the event and worked in four teams on the construction of the main magnet, gradient coils, RF hardware, and software platform. In this talk, we will discuss how the idea for the hackathon came together, present the created scanner design and components, and summarize our experience from the hackathon week. We will also give a live demo of the scanner, which is currently located at our 22nd Street laboratory.

Abstract

Low-field magnetic resonance imaging (LFMRI) offers greater accessibility to MR scanners by reduced manufacturing and maintenance costs. However, the signal-to-noise ratio of the acquired images is inherently diminished by the use of low field strengths. The standard technique of averaging multiple acquisitions (to increase SNR) reduces LFMRI accessibility for patients by increasing scan time and cost. Hence, one of the recent trends in LFMRI research focuses on employing advanced image processing techniques to enhance few-average, and even single-average, LFMR image quality and enable greater scanner accessibility.

In this talk, we will cover several strategies for self-supervised learning of deep neural networks (DNNs) for MRI denoising, given only unlabeled noisy data. We begin by reviewing standard observation models for parallel (multi-coil) MRI contaminated with additive white Gaussian noise. We will then cover state-of-the-art loss functions for training DNNs when a single, or potentially multiple, noisy observations of each subject are available for training (ex. SURE, Noise2Noise, Coil2Coil). We use labeled MRI datasets with synthetic degradation to allow for quantitative comparison of the different methods. Finally, we will show applications of these techniques to unlabeled noisy MRI datasets of the lung and the prostate acquired at 0.55T. Complications arising in the translation from synthetic to real-world experiments, such as coil-correlation and noise-level estimation, will be highlighted.

Abstract

In this talk we highlight the importance of biophysical modeling of the diffusion MRI (dMRI) signal in order to obtain metrics that are sensitive and specific to brain microstructure. We will focus on the standard model of diffusion in white matter (WM), its validation, parameter estimation, and clinical applications. First, Ricardo will present a comprehensive histological validation of standard-model parameters by characterizing WM microstructure in sham and injured rat brains using 3D electron microscopy and ex vivo dMRI. This validation reveals that the standard model is sensitive and specific to microscopic properties. Next, Ying will discuss several different approaches aimed at resolving the degeneracy issue commonly encountered in parameter estimation using clinically feasible dMRI protocols. He will compare the model constraints of existing methods and an unconstrained, data-driven approach. He will present in vivo data regarding early development, multiple sclerosis, and stroke for validation of these diffusion model parameter estimators.

Abstract

Magnetic Resonance Imaging (MRI) below 1 Tesla is a recently emerging area, driven by the integration of modern MRI infrastructures (gradient system, RF coils, software, etc.) and the unique physical prosperities of the low magnetic field. These modern low-field systems provide lower susceptibility, improved B0 homogeneities, lower specific absorption rate (SAR), lower acoustic noise, and favorably scaled relaxivities (lower T1 and longer T2/2*). These advances unlock new clinical possibilities, particularly in lung imaging, dynamic imaging, and imaging near metallic implants. In this talk, I will give an overview of research conducted using a prototype whole-body 0.55 T MRI (MAGNETOM Aera, Siemens Healthineers) at the Dynamic Imaging Science Center (DISC) of USC since 2021. Covering applications from head to toe, this talk highlights the promising outcomes and potential clinical implications of this innovative 0.55 T MRI approach in cardiac, lung, body, musculoskeletal imaging, and so on.

Abstract

First, Dr. Asslaender will discuss the connection between longitudinal relaxation and magnetization transfer. He will explain recent advances in the biophysical modeling of these effects, outline implications for our understanding of longitudinal relaxation in biological tissue, such as brain white matter, and touch on new avenues for diagnostic imaging of neurodegenerative diseases such as multiple sclerosis. Advanced biophysical models commonly involve a plethora of parameters, which hampers time-efficient mapping of these parameters. In the second part of the talk, Dr. Flassbeck will discuss approaches to overcome these challenges. He will discuss pulse sequence design and optimization in the realm of MR fingerprinting, discuss the concept of the hybrid state, and touch on image reconstruction of highly undersampled dynamic imaging.

Abstract

Cerebral palsy (CP) is a neuromusculoskeletal condition arising from a neural lesion of the central nervous system which occurs before, during or soon after birth. The neural lesion leads to progressive muscular degeneration making the condition the most common cause of physical disability in paediatric populations. MRI and Diffusion Tensor Imaging (DTI) provides great potential for an in vivo assessment of muscle structural and architectural properties of lower-limb muscles. This allows for an understanding of the structure-function relationship in lower-limb muscles. Previous research has investigated structural and architectural properties of lower-limb muscles including muscle volume and pennation angle in paediatric populations with CP. In this research, anatomical MRI was used to extract volumes, i.e. structures, of lower-limb muscles. Additionally, DTI was used to generate voxel-based fibre orientations based on the diffusion properties of water molecules within lower-limb muscles, and a deterministic tracking algorithm was used to reconstruct 3D muscle architecture. Structure and architecture datasets were combined to develop longitudinal, statistical models of lower-limb muscles morphology and 3D fibre orientations. The developed models characterised the dominant variations of lower-limb muscles over a specific period of time in paediatric populations with CP and healthy counterparts. The developed models offered a longitudinal, quantitative analysis of muscle structure and architecture providing insights into the progression of muscle degeneration in paediatric populations with CP, which may assist in the design of targeted clinical interventions for motor dysfunctions associated with this condition.

Abstract

Recent development of time-dependent diffusion MRI (TDDMRI) has demonstrated unique advantages in depicting cellular microstructures by characterizing the time-dependence of restricted water diffusion. Previous simulation and animal studies have shown that TDDMRI is sensitive to microscopic pathology in tumors, yet clinical translation of this technique has been challenging due to the hardware requirement for high gradient strength. Nevertheless, initial tests in human studies are emerging. This talk will first introduce TDDMRI, from theory to animal and human studies, and then describe our recent endeavor in developing new pulse sequences for clinical translation of TDDMRI. I will then talk about our clinical studies in prostate cancer, glioblastoma, and breast cancer, which showed promising clinical value. Last, I will also touch upon our recent work on developing an ultra-high-gradient system for better potential of TDDMRI.

Abstract

The choroid plexus (ChP), a highly vascularized structure in the brain ventricular system, is increasingly recognized for its vital role in cerebrospinal fluid (CSF) hemostasis and waste clearance, particularly in relation to aging and dementia. While some imaging studies have investigated its volumetric changes, the majority of ChP studies have relied on histopathological approach. Limited MRI research has been conducted on age-related vascular degeneration, primarily due to resolution and tissue contrast constraints. This presentation will focus on the in vivo exploration of vascular aging in the ChP utilizing multimodal MRI data, including USPIO-enhanced susceptibility-sensitive imaging on 7T as well as dynamic-contrast enhanced (DCE) and arterial spin labeling (ASL) perfusion MRI on 3T; These modalities allow for detailed characterization of both vascular anatomical and blood flow alterations in normal aging process. We observed significant age-related vascular degenerative changes in three adult lifespan study cohorts. These changes include reduced vascular enhancement and blood flow, accompanied by structural alterations, such as increased ChP stromal volume, cysts formation and changes in water mean diffusivity on diffusion MRI. Combined, the observed vascular aging in the ChP through MRI may play a role in compromised ChP functions, including CS secretion, filtration, and waste clearance. These in vivo insights are anticipated to enhance our understanding of the ChP’s involvement in cognitive dysfunction.

Abstract

Quantitative MRI techniques spatially resolve biophysical parameters generally by leveraging two steps in tandem: image reconstruction followed by parameter fitting. In this work, we propose improvements to both aspects of this traditional pipeline. Using the Cramér-Rao bound as a figure of merit, we design a computationally efficient low-rank reconstruction method that significantly improves the accuracy and precision of the downstream biophysical parameters. This is combined with a neural network designed to replicate the properties of a minimum variance unbiased estimator. We test our methods for single-compartment T1/T2 MR-fingerprinting in addition to two-pool magnetization transfer imaging in the hybrid state. Combined, our proposed methods facilitate the development, validation, and translation of novel quantitative MRI biomarkers.

Abstract

The brain represents two percent of the body weight but consumes 20 percent of the energy budget. Thus the brain’s oxygen supply and metabolism are carefully controlled to maintain its function and health. Aberrant brain oxygen homeostasis is implicated in many brain disorders such as Alzheimer’s disease and cerebrovascular diseases. It is challenging to measure brain oxygen metabolism in vivo, often requiring the injection of radioactive tracers with very short half-life. In this seminar, I will introduce an MRI based technique to measure brain oxygenation and metabolism in rodents and humans across the lifespan, from one day to 90 years old. I will also show potential applications of the technique in brain aging and diseases.

Abstract

Lengthy data acquisition remains a major bottleneck in magnetic resonance imaging (MRI), often necessitating tradeoffs in resolution and signal-to-noise ratio. Thus, reconstruction and acquisition techniques for rapid imaging, noise reduction and improved data acquisition have received great interest. Each of these directions correspond to a specific inverse problem with its own distinct forward operator dictated by the underlying imaging physics.

In this talk, we will describe recent advances that link these inverse problems in MRI through the lens of intelligent physics-driven technologies. We will first focus on physics-driven deep learning (DL) methods for accelerated MRI. In this context, we will overview our pioneering work on self-supervised learning strategies for training such reconstruction algorithms when ground-truth data is not available, which is a common problem in MRI. We will also show how these can be extended to a subject-specific zero-shot setting when a training database cannot be curated. We will then explore state-of-the-art methods for denoising MRI series that utilize random matrix theory based approaches. We will discuss how this strategy can be combined with physics-driven DL reconstruction to provide a synergistic improvement. Finally, we will overview emerging developments for improving radiofrequency pulse design with a focus on improving field inhomogeneity at ultrahigh field strengths.

Abstract

The talk will focus on the high b-value analysis of brain diffusion MRI data to achieve a robust quantification of white matter axonal properties such as the apparent diffusivities, radius, T2, and signal fraction. However, a common restrictive assumption behind this kind of quantification is the absence of cellular compartments. We will see that by using zonal modeling (a generalization of the spherical variance approach) of the high b-value signal it is possible to account for cell-like contributions and achieve an unbiased quantification of axonal properties. Although not free from its challenges, this unbiased quantification provides the collateral advantage of a better resilience to partial volume effects with gray matter and cerebrospinal fluid. These findings extend beyond current methodologies based on powder averaging, highlighting the importance of unbiased quantification in diffusion MRI.

Abstract

Motion in MRI scans causes image corruption and remains a barrier to clinical imaging. We propose Beat Pilot Tone (BPT), a simple yet powerful serendipitous system exploitation that turns any MRI receiver chain into a radio frequency (RF) motion sensing system that can operate at arbitrary frequencies (up to several GHz). Our contact-free system can be implemented on any MRI scanner regardless of field strength. Through electromagnetic field simulations and experiments, we explain BPT’s novel mechanism: two or more transmitted RF tones form motion-modulated standing wave patterns that are sensed by the same receiver coil arrays used for MR imaging. These waves are incidentally mixed by intermodulation and digitized simultaneously with the MRI data. BPT achieves an order of magnitude greater sensitivity to motion than other RF methods in detecting and separating common motion types (respiratory, bulk, cardiac, and head motion) in volunteers. Moreover, BPT offers tunable sensitivity to motion based on the transmit frequencies; at microwave frequencies, BPT can detect millimeter-scale vibrations (ballistocardiograms). With multiple antennas and frequency-multiplexing, BPT can operate as a multiple-input multiple-output (MIMO) system. Preliminary experiments have demonstrated the utility of BPT for retrospective head motion correction. BPT significantly expands the capability of any MRI system, paving the way toward multi-modality, motion-robust, and simultaneous RF and MR imaging.

Abstract

Lymph node metastases significantly influence patient prognosis and treatment decisions. To improve lymph node staging, 3D mGRE imaging is combined with a superparamagnetic iron oxide contrast agent (USPIO) while addressing challenges associated with respiratory motion, ultrahigh field strengths, and T2*-weighted imaging.

Abstract

Deep generative models use deep neural networks to model complex distributions. Score-based generative models (SGMs) have recently become the de facto method for modeling image distributions. As such, there has been a concerted effort to leverage SGMs to solve inverse problems in medical imaging. In this work we develop methods for the application of SGMs to PET image reconstruction, specifically considering the nuances of the modality. We propose adaptations to sampling methods, MR image guided reconstruction using classifier-free guidance, and acceleration for scaling to fully 3D reconstruction.

Abstract

The talk will describe imaging modalities that can be used to assess bone microstructure and its contribution to bone strength. Studies in anorexia nervosa and obesity as well as studies following caloric restriction and weight gain will be discussed.

No abstract was provided for this talk.

Abstract

Magnetic resonance fingerprinting (MRF) is a rapid quantitative imaging technique. We compare T1 and T2 measurements from MRF to T1 variable flip angle mapping and T2 multi-echo spin echo mapping in the NIST phantom, 10 healthy volunteers and 17 glioma patients. Our results suggest magnetisation transfer (MT) effects in healthy brain tissue and glioma cause biases between MRF and other mapping methods. MRF’s sensitivity to MT presents an opportunity to characterise brain tumours in more detail.

Abstract

This presentation will delve into new tractography methods used for presurgical mapping, specifically focusing on multi-level fiber tractography (MLFT) and constrained spherical deconvolution (CSD) probabilistic tractography. We will outline the unique advantages these techniques offer in a presurgical environment, as demonstrated by two different research studies. This talk will start with an overview of neurosurgery, highlighting the critical role of neuroimaging. We will clarify basic concepts relating to diffusion MRI, fiber tractography (FT), and brain mapping with transcranial magnetic stimulation (TMS) and intraoperative direct electrical brain stimulation (DES). Additionally, we will provide a brief overview of two techniques developed in our lab, which have been integral to the success of the fiber tracking studies under discussion. These innovative methods were essential for the two studies investigating the importance and potential of advanced tractography methods in the field of neurosurgery.

Abstract

Aging and neurodegeneration are associated with changes in brain tissue at the molecular level, affecting its organization, density, and composition. These changes can be detected using quantitative MRI (qMRI), which provides physical measures that are sensitive to structural alterations. However, a major challenge in brain research is to relate physical estimates to their underlying biological sources.

In this talk, I will highlight approaches for differentiating between changes in the concentration and composition of lipid and iron in the brain. I will first present a biophysical model that stems from the notion of relaxivity, the ability of a certain compound to increase the MR relaxation of the surrounding water proton spins. I will then suggest a phantom system of lipid and iron forms to test the relaxivity approach. Next, I will describe the intrinsic relaxivity of brain tissue in vivo during changes in the aging human brain. Finally, I will compare the in vivo approach to histological characterization of lipids and iron compositions of the brain.

Abstract

MRI at lower field strengths (e.g., 0.55 T or below) has emerged as an exciting area of research in recent years, preseting both new opportunities and unique challenges. One notable advantage of low-field MRI is reduced cost, which is crucial to increasing the accessibility of MRI in healthcare facilities with limited resources. However, the benefits of low-field MRI go beyond cost considerations, and it offers distinct opportunities for certain MRI applications such as shorter T1, reduced metal artifacts, longer T2* times, and higher gadolinium relaxivity. In the meantime, it is also essential to acknowledge that low-field MRI does have certain important limitations, such as a lower signal-to-noise ratio and reduced fat suppression performance. Also, to reduce cost, current 0.55 T scanners have had limited gradient performance.

Last year, UCSF successfully installed a state-of-the-art Siemens 0.55T Free.Max MRI scanner. Since then, our team has been actively investigating its capabilities in body imaging. In this talk, I will provide an overview of our firsthand experience with the Free.Max scanner for body MRI and will also discuss the opportunities and challenges for body imaging at this field strength.

Abstract

The use of MR to plan, deliver, monitor, and assess the efficacy of radiotherapy is an area of increasing interest and exploration. In the context of personalized treatment, a principal benefit of this modality is the ability to guide adaptive radiation therapy with improved target definition and without the exposure of healthy tissue to ionizing radiation. Recent technical advances in hybrid devices that combine a medical linear accelerator with a diagnostic MRI scanner have enabled such methods of increased treatment precision under real-time guidance and adaptation. These hybrid systems enable daily planning to account for interfraction as well as intrafraction motion using high-resolution, volumetric, real-time monitoring of tumor and organs-at-risk during radiotherapy delivery. They have the potential to account for both complex and systematic changes whether they are due to random changes in organ shape or more systematic changes in tumor volume using both online and offline adaptive radiotherapy. This talk will focus on the current status of such hybrid MR-guided radiotherapy systems and discuss some challenges and potential solutions.

Abstract

Deep Learning has shown success across many problem domains, however harnessing its full potential in the life sciences requires understanding both the causal processes generating the biomedical data and how to develop new analytical models that inherit the power of deep learning while respecting the complexities in the data. We will illustrate this principle in several ways. First, when building predictive models, the core assumption for the past several decades has been that samples are independent and identically distributed. Yet, in biomedical sciences, they can be clustered by study site, subject, or batch, leading to poor model fitting. We propose Adversarially-Regularized Mixed Effects Deep learning (ARMED) to separate cluster-invariant from cluster-specific features through additions to existing neural nets. These include an adversary, a Bayesian random effects subnetwork, and an approach to handle unseen clusters. When applied across vastly different types of biomedical data and prediction tasks, it learns more biologically plausible features and improves performance over conventional networks, including on data from clusters unseen during training. Second, deep learning models consisting of parallel subnetworks are increasingly used to solve complex prediction tasks, e.g., from multi-modal inputs. Hyperparameter optimization (HPO) algorithms can tune models to achieve better performance, but existing approaches fail to consider network structure. We propose Module Adaptive Bayesian Optimization (MABO). With a separate surrogate Bayesian model for each subnetwork and judicious transfer learning, MABO requires 2x-16x fewer computational resources than standard HPO approaches while significantly improving accuracy. Join us to learn how MABO is democratizing access to performant HPO. We will conclude with an application to predict patient-specific outcomes. In depression, the lack of biomarkers to inform antidepressant selection is a key hurdle to personalize treatment. While measures of brain activity from fMRI have shown promise, their successful application to this task hinges not only on the choice of model but also on the data preparation, including fMRI data augmentation, artifact suppression, and HPO. We build predictors of individual treatment outcomes by identifying reward processing measures from appropriately prepared fMRI and clinical assessment, and show how the model attains an excellent number-needed-to-treat performance for the physician.

Abstract

Type C hepatic encephalopathy (HE) is a severe neuropsychiatric disease occurring as a consequence of chronic liver disease, for which the prognosis is poor in the absence of liver transplantation. The understanding of biochemical mechanisms underpinning neurological and cognitive dysfunctions is still incomplete. In the first part of this presentation, we will show that diffusion-weighted MR spectroscopy (dMRS) and imaging (dMRI) at 14.1T probed cell-specific changes in metabolite diffusivities in the cerebellum of a rat model of type C HE compared to control rats, as well as an increased intra-neurite and intra-axon water diffusivity after matter-specific biophysical modeling of the diffusion signal in white and gray matter. These results confirm an alteration of cell density and/or of neurite network complexity observed ex vivo by histology and render dMRS a highly valuable tool to probe cell-specific microstructure in vivo. In the second part, a new preclinical 18F-FDG PET methodology to compute quantitative 3D maps of the regional cerebral metabolic rate of glucose (CMRglc) from a labeling steady-state PET image of the brain and an image-derived input function will be presented. This quantitative approach showed its strength in comparisons of groups of animals with divergent physiologies. In vivo, a 50 percent lower brain glucose uptake, concomitant with an increase in brain glutamine and a decrease in the main osmolytes measured by 1H MR spectroscopy, was observed in the hippocampus and in the cerebellum of HE rats, confirming the hypothesis of energy metabolism alteration in HE.

Abstract

Using 3D images to train AI models is computationally challenging because they consist of tens or hundreds of times more pixels than their 2D counterparts. To be trained with high-resolution 3D images, convolutional neural networks resort to downsampling them or projecting them to two dimensions. We developed an effective alternative, 3D Globally-Aware Multiple Instance Classifier (3D-GMIC), a neural network that enables efficient classification of full-resolution 3D medical images. Our method substantially reduces the GPU memory and computation requirements, and generalizes well to an external dataset. Moreover, it explains its predictions by providing pixel-level saliency maps, despite being trained solely on image-level labels.

Abstract

Multiple pathological processes interact with brain regions that support episodic memory, resulting in subtle performance differences well before overt clinical impairment. One common pathological pathway is Alzheimer’s disease, and it is established that both hallmark features of this disease are detectable in many older clinically unimpaired adults (amyloid plaques and tau tangles). We have demonstrated heterogeneity in spatial patterns of tau PET signal among clinically unimpaired adults in key cortical nodes relevant to visual associative memory processes (ventral temporal lobe, precuneus, inferior parietal). Along these lines, we find that fMRI metrics of cortical reinstatement during retrieval of associative pairs is reduced in clinically unimpaired adults with genetic risk of Alzheimer’s (APOE4+), and also shows associations with global measures of abnormal tau in the cerebrospinal fluid and focal signal in the ventral temporal lobe. Interestingly, both abnormal tau and cortical reinstatement strength independently predict individual differences in overall memory performance, emphasizing the presence of multiple age-related pathways that influence cortical mechanisms of memory.

Abstract

MRI is a powerful imaging tool. However, technical challenges such as long setup and scan time, sensitivity to physiological and bulk motion, and difficulties with horizontal and longitudinal comparison due to its qualitative nature, have significantly hindered MRI from realizing its full potential. In recent years, with the blooming advances in acquisition, reconstruction, and deep learning, push-button MRI has become possible: it will be a short, continuous acquisition producing bulk-motion-robust, physiological-motion-resolved, and quantitative images containing structural and functional information.

In this seminar, I will introduce my work towards push-button MRI. The first part will focus on Multitasking DCE, which enables continuous free-breathing acquisition with 3D flexible anatomical coverage, around 1-millimeter spatial resolution, 1-second temporal resolution (15-60 times higher than clinical DCE protocols), and accurate quantification of tissue microvasculature properties in the presence of physiological motion. This technique has been successfully translated into clinical practice for the differential diagnosis and disease characterization of carotid atherosclerosis, pancreatic cancer, chronic pancreatitis, and breast cancer.

The second part will introduce new advances in echo-planar time-resolved imaging (EPTI), a highly efficient technique for rapid neuroimaging. The advanced version of EPTI—spherical EPTI (sEPTI)—has achieved an additional 1.4x acceleration and improved system robustness, allowing for whole-head T2* and QSM quantification with 1-millimeter isotropic resolution in just 45 seconds. A generalizable navigator is currently being developed for motion and B0 tracking with a latency of approximately 100 milliseconds to achieve B0 and bulk-motion corrected imaging.

In conclusion, the seminar will provide an outlook into the future of MRI. By leveraging the advances in acquisition, reconstruction, and deep learning, MRI will achieve further acceleration, faster reconstruction, richer information in tissue structure and function, and improved scientific and clinical outcomes.

Abstract

Mercure is a flexible open-source DICOM orchestration platform developed by the Center for Advanced Imaging Innovation and Research. It offers an intuitive web-based user interface and extensive monitoring options. It can be used for dispatching DICOM studies to different targets based on easily definable routing rules and for processing DICOM series with custom-developed algorithms, such as inference of AI models for medical imaging.

The Center for Advanced Imaging Innovation and Research has deployed several models via Mercure, making them directly available for researchers via Visage. This presentation will cover an overview of Mercure, examples of deployed models currently used by NYU Langone researchers in their PACS workflow, and instructions for deploying your own models via Mercure. There will be a brief introduction to the MONAI Deploy framework and how it can be utilized with Mercure for the rapid deployment of pre-trained open source models. Examples of deployed models will include glioma segmentation, anatomical brain segmentation, and whole body CT image segmentation.

Abstract

Thin skin-wearable devices, sometimes referred to as epidermal electronics, found in many research articles fall short as concept-only representation due to the fundamental discrepancy in the mechanics of thin-film materials and rigid essential components. This talk introduces a set of engineering solutions to overcoming the challenges in manufacture and assembly of epidermal electronics and the soft wearable bioelectronics platform in general. Strategic integration of thin-film electronics with soft elastomers allows the stretchable biopotential electrodes to maintain the conformal contact with the skin while the integrated circuit deforms naturally with the body. The stretchable electrodes with optimized design and structure for intimate skin integration are capable to perform high-fidelity electrophysiology and accurate analysis of the skin’s electrical properties, such as the galvanic skin responses. Moreover, direct integration of small, off-the-shelf chip sensors (e.g., accelerometer, pulse oximeter, and microphone) with a stretchable electronic platform opens the possibility for concurrent monitoring of multiple physiological parameters, while providing researchers with freedom of device placement on the body. Implementation of smartphone applications embedded with real-time classification algorithms demonstrates the feasibility of multifaceted analysis with a high clinical relevance. Finally, results from multiple human studies of various scenarios reveal the true potential of the soft bioelectronics as both a powerful research tool and a game-changer for wearables-enabled digital health.

Abstract

Diffusion-weighted MRI is a powerful medical imaging technology that allows for noninvasive assessment of degenerative spinal cord. In particular, diffusion tensor imaging (DTI) has been widely used in evaluation of mouse spinal cord affected by amyotrophic lateral sclerosis (ALS). However, no significant changes in diffusivity, which can provide environmental information, were found between wild type and mutant groups. To address this issue, a novel ultra-high b-value diffusion-weighted MRI technique was utilized to identify early-stage diffusivity changes caused by ALS in the spinal cord’s degraded microenvironment. This presentation will cover the imaging protocols for both ex vivo and in vivo studies, as well as post-processing techniques using continuous time random walk and multicomponent analysis. The speaker will also briefly discuss representative studies that she assisted with during her graduate research assistantship.

Abstract

In my talk, I will introduce the motivation behind performing fMRI in the spinal cord at 7 T and the technical challenges that are magnified by this shift. These include the relative paucity of RF coils available for spinal cord MRI at 7 T, increased spatial and temporally-varying B0 inhomogeneity due to proximity to the vertebral column and lungs, and other situations where established methods in brain fMRI need to be re-evaluated and re-optimized for use in the spinal cord. I will then review several technical developments that facilitate spinal cord fMRI at 7 T, including hardware design, dynamic B0 shimming, and collection of GRAPPA autocalibration signal data, and finish with an overview of a thermal pain stimulus task-fMRI study where we applied these methods.

Abstract

Magnetic Resonance Imaging (MRI) is a superb imaging modality, which offers rich information about the human body. However, its clinical use is hindered by the long scan duration and high sensitivity to motion artifacts. Moreover, due to the high complexity of the MRI system, MRI components are commonly designed separately, which leads to sub-optimal performance. Although machine learning (ML) techniques have shown great promise for addressing these limitations, their development is hindered by the scarcity of suitable training data. This seminar will introduce new strategies for developing ML-powered computational frameworks for fast, motion-robust MRI. First, data-related challenges will be discussed; it will be demonstrated that naïve use of open-access medical databases could lead to biased, overly optimistic results. Then, new strategies for rethinking the entire imaging pipeline will be discussed. Two computational frameworks for rapid dynamic (temporal) imaging will be introduced, with end-to-end acquisition-reconstruction design. The first, BladeNet, combines a motion-informative “PROPELLER” sampling technique with a unique ML-based reconstruction network. This framework enables fast, motion-robust, free-breathing imaging, which is highly suitable for pediatric body imaging. The second framework, K-band, addresses challenges in 4D (dynamic-volumetric) MRI by introducing an end-to-end pipeline design, with fast data acquisition and self-supervised reconstruction. This pipeline enables training model-based reconstruction networks using only limited-resolution data, with real-time generalization to high-resolution reconstruction during inference. The seminar will conclude with an outlook to the future of computational medical imaging, focusing on low-coast portable scanners and personalized longitudinal healthcare.

Abstract

Producing accurate input-output mapping of the human connectome is a great scientific challenge our time and technologies. My team has been pursuing MRI diffusion approaches to making an accurate connectome for 14 years. I have been disappointed at the low accuracy of current methods to be able to reproduce established anatomical connectivity. The diffusion MRI community has not sufficiently recognized that known anatomy shows substantial within-tract-fasciculus migration of small scale axon bundles (0.3 mm diameter) passing at small angles (under 10 degrees) and making sharp turns (270 degrees in 0.1 mm) with punctate areas of crossings within tract. This anatomy poses what I view to be insurmountable challenges for in vivo diffusion human imaging to produce a reasonably accurate (over 90 percent) input-output connectome map. We are developing a new method we call Fasciculus Axon Connective Tissue Multiscale Imaging (FACTMI) using multiscale imaging including MRI at 20-micron resolution and selective 0.1-micron optical Magnify imaging. I believe an accurate connectome can be provided at viable cost and throughput within five years.

I will describe our team effort to achieve the “Pittsburgh early visual system connectome challenge.” Can an accurate (greater than 90 percent hits minus false alarms) map each of 4 fasciculus coding inputs (each eye lateral and nasal visual field) to the 12 (6 layers of lateral geniculate nucleus left and right) outputs matching the established anatomically correct answer published over 50 years ago (Wiesel & Hubel 1966). I will review embryonic development, fasciculation, and neuronal migration, in the context of local guidance molecules and changing topology that we expect observe in the 8 cm from eye to lateral geniculate nucleus (LGN).

With our Max Planck/Stuttgart colleagues, we are a developing harvested tissue slice imaging to deliver high-resolution (20 micron voxel) MRI structural tensor fasciculus following to map the pig and human early optical system. It includes developing CMOS MRI-on-a-chip technology for large parallel-optimized coil arrays within 2.5 mm of the tissue for imaging at 14 T. The 20-micron imaging enables following the walls of fasciculi that are 20-100 microns thick with 4,000 MRI slices from the eye to LGN. We use 0.1-micron Magnify histology to provide counts of axons in each fasciculus along the path. We will follow fasciculus walls to predict the path of fasciculated axons in each of more than 150 fasciculi (0.2 to 0.8 mm diameter sets of ~8,000 axons) per optic nerve. We are testing MRI methods with phantoms that provide ground truth of the paths of millions of taxons (textile axon size tubes; 0.9 micron in diameter) routed in paths to match histology data. We test biological validation in harvested pig tissue at viable scanning time (4-day scans of the optic nerve system at 20-micron voxels). We test the accuracy of the connectome mapping quantification by scoring the ability to predict the number and diameter of the axons at each fasciculus at points along the path from the eye, through the optic chiasm, to layers of LGN based on 20-micron MRI data. We will use deep learning synthetic histology to predict observed histology from low (1 mm) and high-resolution (20 μm) MRI data.

Abstract

Optoacoustic tomography is a mesoscale volumetric imaging modality which combines the advantages of optical imaging contrast and low acoustic scattering in tissue. In my work, I apply optoacoustic tomography for minimally invasive functional neuroimaging, taking advantage of exogenous contrasts such as near-infrared dyes and genetically encoded calcium indicators as well as endogenous hemodynamic contrast. Each experiment utilizes wavelengths within the near-infrared optical tissue window, at which low light absorption allows optoacoustic imaging at centimeter-scale depths in tissue. In addition to experiments with anesthetized mice, I also developed and implemented the first setup for optoacoustic neuro-tomography in awake, behaving animals.

Abstract

Radial MRI sequences have shown significantly higher robustness to motion, enabling patient exams during ongoing motion such as free breathing. In addition, radial sequences offer powerful undersampling properties, especially when combined with compressed sensing, which can be exploited for flexible image contrast manipulation and for highly accelerated DCE-MRI using advanced reconstruction algorithms such as GRASP. In this talk, I will present two novel applications of radial MRI that benefit from its unique sampling properties.

The first application addresses fat suppression at 0.55 T. Low-field MRI systems have gained strong interest due to lower cost. However, the reduction in field strength leads to significant challenges for fat suppression due to the small spectral fat/water distance, which makes conventional fat suppression techniques ineffective. To address this limitation, I will describe an alternative approach for fat suppression using continuous radial acquisition during frequency sweep of an RF saturation pulse, combined with frequency-resolved compressed-sensing reconstruction.

The second application aims at volumetric dynamic MR imaging for functional kinematic assessment of the wrist, which can be useful for evaluating wrist instability. Existing real-time MRI methods are typically either limited to 2D imaging or provide only low temporal resolution and insufficient image quality. To address this challenge, I will present a novel approach for volumetric dynamic wrist examination that assembles 2D real-time data into 3D snapshots using MRI-visible markers on a 3D-printed platform to guide continuous ulnar-radial deviation.

Abstract

Prostate cancer is the most diagnosed malignancy and the fourth leading cause of death in men, with more than 80 percent of men developing the condition by the age of 80. MRI has become an increasingly important tool for prostate cancer diagnosis and management, including biopsy guidance. Faster imaging and automated diagnostics may enable more cost-effective workflows and make prostate MRI more widely accessible. To achieve faster imaging, there has been a surge in machine learning based MR reconstruction research. Supervised machine learning based methods for image reconstruction require vast amounts of raw k-space data for model training. The limited availability of raw k-space datasets motivated NYU Langone Health and Meta AI Research (formerly Facebook AI Research) to release the fastMRI dataset in 2020. To our knowledge, fastMRI is the largest public dataset of raw k-space, including knee and brain MRIs, acquired from a clinical population. This resource encourages exploration of multiple fast, pathology informed reconstruction methods by the MR and AI community. To further advance this goal, this study aims to add a biparametric prostate dataset to fastMRI to facilitate the development of machine learning tools to increase the utility of prostate MRI.

Abstract

During MR acquisitions, magnetic susceptibility causes field distortion, a slight Larmor frequency shift and further induces a faster signal decay. Hence, its effects are often treated as an undesired artifact. The phase of a gradient recalled echo signal captures spatial variations of magnetic susceptibilities. By solving a magnetic dipole model using the B0 inhomogeneity field map, we are able to recover the tissue’s local magnetic susceptibility namely quantitative susceptibility mapping (QSM). QSM reflects the local molecular contents and tissue architecture and has shown great potential in research on brain development and neurodegenerative diseases. Common magnetic susceptibility sources in the brain are paramagnetic species such as iron deposition, hemorrhage, and diamagnetic species such as myelin, calcification plaques, beta-amyloid plaques, and tau-protein tangles.

However, what if the paramagnetic and diamagnetic sources co-localize in one voxel? Using the simple dipole model, the phase effect from positive and negative magnetic susceptibilities will cancel out and appear zero value in QSM. This talk will introduce a recently proposed 3-pool signal model DECOMPOSE-QSM that can resolve the susceptibility mixture situation. The multi-echo gradient echo MR signal of one voxel is represented as a summation of signals from 3 voxel pools. The parameters of the signal model are used to construct paramagnetic and diamagnetic component susceptibility maps. The application of this method shows its potential to emphasize the iron overload in basal ganglia for Parkinson’s disease. It also potentially can detect demyelination in an Alzheimer’s disease study. Further, with multi-orientation acquisitions, more coherent susceptibility-based fiber structures are revealed with DECOMPOSE-QSM processed magnetic susceptibility tensor imaging (STI) and high angular resolution susceptibility imaging (HARSI).

Abstract

Interpretability is a crucial aspect of deep learning models in medical imaging applications. However, achieving interpretability through fully supervised methods requires a significant amount of human-provided training labels, which can be costly and difficult to scale in large datasets. Furthermore, in many tasks, human-provided training labels are not available, as even experts may not possess the necessary knowledge. In these situations, full supervision is not practical. In this talk, I will present two lines of research on building interpretable deep learning models that can learn from imperfect labels. The first series focuses on a model that can localize cancer without the need for localization labels. We will explore its application in breast cancer screening and the prognosis of COVID-19 patients. The second series centers on learning from imprecise labels. We will examine how this approach enables the model to interpret breast ultrasound and histopathology images.

Abstract

Magnetic resonance is unique in its ability to image a wide range of physiological processes noninvasively, using a rich palette of different imaging contrasts—often without the injection of an external agent. At the heart of this ability lies the concept of a pulse sequence: our ability to exquisitely control the quantum-mechanical evolution of nuclear spins using external electromagnetic fields. I will talk about recent work in my group, focusing on the development of new pulse sequences for imaging completely new neurophysiological contrast types, ranging from intracellular viscosity to neurotransmitter dynamics, with far-reaching implications for neuroimaging and medical diagnostics.

Abstract

Alzheimer’s disease (AD) studies using established imaging (MRI and PET)- and cerebrospinal fluid (CSF)-based biomarkers have shown that molecular changes begin years before symptom onset, but mechanisms underlying the hallmark pathological aggregation of amyloid and tau are still unknown. To address this from a ¹H-MRS standpoint, studies have used conventional PRESS sequences in cognitively unimpaired populations to examine metabolic dysfunction associated with AD development and progression in otherwise normal-appearing tissue in the posterior cingulate. While there are ex vivo histopathological and in vivo imaging evidence that AD pathological hallmarks are not confined to the posterior cingulate, ¹H-MRS interrogation of other regions in cognitively unimpaired elderly has not been done.

The hippocampus, one of the earliest affected regions in AD, is rarely interrogated by ¹H-MRS because it is situated in a location with poor B0 homogeneity and large susceptibility effects. In the first part of this presentation, I will show how we applied a recently validated long-TE sLASER sequence, which, with its high bandwidth adiabatic pulses, reduces chemical shift displacement errors (CSDE) inherent to PRESS; and, with its long-TE, minimizes macromolecular contribution to the background signal, improving quantification of metabolites measured from the hippocampus.

Furthermore, other cortical (posterior and isthmus cingulate, precuneus) and subcortical (caudate, putamen, globus pallidus, thalamus) gray matter structures have shown vulnerability to AD neurodegeneration. In the second part of this presentation, I will show how we applied ¹H-MRSI to examine spatial characteristics of metabolic dysfunction in these seven gray matter regions (plus one control region, the lateral occipital).

We then tested whether metabolites measured from both ¹H-MRS and ¹H-MRSI were associated with (1) APOE4 genotype, a risk factor for amyloid burden, (2) CSF p-tau181, an indicator of tau burden, and (3) morphometry metrics (volume, cortical thickness) indicative of neurodegeneration.

Abstract

The research goal of Dr. Cai’s lab is to build new paradigms of nanotechnology for biological and biomedical applications. At the Tech4Health Institute, his team is developing novel miniaturized optical and mechanical probes to measure and manipulate biological input and output signals. By borrowing nanolithographic technology from the semiconductor industry, his lab creates optical metasurfaces, planar optics, and cell nano-interfaces with unprecedented precision and functionality. Integrating these nanoengineered 2D surfaces on a series of functional platforms (e.g., 3D structures, soft materials, active micro/nano systems), innovations for broad applications become feasible in a dynamic and tunable fashion. In this talk, Dr. Cai will present his recent work on (1) Optical metasurfaces and nanophotonics for bioimaging and biosensing; (2) Nanoengineered cell interfaces for mechanobiology study. In particular, he will highlight the adoption of dielectric materials in optical metasurfaces and cell interfaces. Based on Mie resonance instead of plasmonics, dielectric metasurfaces eliminate the ohmic loss, thermal effects and interactions with fluorescence, which not only provide better optical efficiency, repeatability, stability, and bio-compatibility for imaging and sensing, but also enable super-resolution fluorescence microscopy for nanopattern-based cell mechanobiology study.

Abstract

Deep learning (DL) models trained on large, labeled datasets are currently the state-of-the-art in several classification and segmentation tasks in medical imaging applications. However, obtaining expert manual annotations for data hungry DL models is time-consuming. Additionally, the performance of supervised DL models can be limited by the choice of data representation. This talk explores techniques for transforming images to their new representations that can improve the performance of DL models. The first work presents MR contrast synthesis as a mechanism to transform images to a better representation by utilizing domain knowledge regarding the task of interest. Motivated by the abundance of unlabeled MR imaging datasets compared to labeled ones, the next work explores representational learning techniques to provide suitable initialization for DL models for subsequent tasks. The improvements in segmentation performance from a novel contrastive learning approach with representational constraints derived from multi-contrast MR images is investigated.

Abstract

Positron emission tomography (PET) is a diagnostic tool which utilizes a radiopharmaceutical (compound tagged with a positron emitter) and its detection of the emission of ɣ-rays (511 keV) that results from the annihilation of those positrons. It is an ever-growing field with newly-developed PET radiotracers being approved for clinical use and novel radiopharmaceuticals being reported in the literature. PET is a multi-field tool in which researchers in the areas of oncology, radiology, and psychology are able to benefit from its precise, real-time analysis. The aim of this talk is to give insight into our radiochemistry lab and to showcase what is possible with the use of PET radiotracers.

Abstract

With the average age of the world population increasing very rapidly, diseases that restrict mobility such as osteoarthritis and sarcopenia are on the rise. Conventional imaging methods, including MRI, fail to detect early changes in cartilage and skeletal muscle caused by these diseases, resulting in a missed opportunity for timely treatments. In this talk, I will present several MRI methods that, by providing insight into structure and function, can better characterize healthy joints and skeletal muscle, as well as detect early signs of disease.

Osteoarthritis is not only a disease of cartilage, but it involves the entire joint. In particular, the synovium plays a fundamental role in the progression of osteoarthritis. In my talk, I will illustrate how quantitative MRI can facilitate the diagnosis of synovial inflammation in the knee, and how quantitative MRI can detect the effect of non-surgical treatments on articular cartilage.

Muscle aging is characterized by many compositional and structural modifications, that strongly affect muscle strength, and can lead to sarcopenia. In the second part of the talk, I will present technical advances and applications of several MRI methods, including T2-relaxometry, phase contrast, and diffusion MRI, that can provide insight into skeletal muscle composition and structure, as well as their connection to function.

Abstract

How we think about and study the brain—both in health and disease—is enhanced by our ability to see inside a living human person. And through the power of MRI, ‘seeing’ comes in the form of detailed, multi-contrast images with exquisite sensitivities to brain microstructure. Alterations in microstructure, such changes in myelination or neuronal loss, can culminate in tissue atrophy and disruptions in neuronal connectivity that are detectable using in vivo neuroimaging. It is critical to detect such processes at the microstructural scale because, by the time damage becomes macrostructural, it is largely irreversible. While many quantitative imaging techniques have shown promise in characterizing microstructure, it remains a challenge to link specific cellular-level features to a single contrast or model parameter. In this talk, I will present a hypothesis-driven approach to pediatric imaging that integrates both prior knowledge from neurobiology with high-dimensional data sets to extract biologically relevant features from in vivo MRI. In particular, this talk presents the differences in microstructural hypoconnectivity in brain white matter of typically-developing children and children with a rare genetic disorder that impacts axonal morphology. I will then present a jointly assembled framework that fuses multimodal data across spatial scales within a translatable non-human-primate model (microscopy, ex vivo MRI, and in vivo MRI) to better inform our work on the impact of prenatal exposure to infectious disease. In summary, I will discuss the detection of neurobiological features with specificity through multimodal data sets with the goal of refining and possibly validating current standards of quantitative imaging.

Abstract

When programming MRI pulse sequences on a Siemens system, researchers may find it difficult to have access to sequence resources, including state of the art sequences and basic codes that are protected by intellectual property rights. Furthermore, they may have issues sharing sequence codes or getting inside the Siemens programming environment. In this context, we propose an open-source toolbox, developed in the IDEA toolkit, that provides a modular structure for MRI pulse sequence developments. With its object-oriented structure and its clear variable names, it allows to develop sequences such as spin echo or gradient echo, 2D or 3D, using single line or echo-planar reading. Also, it proposes some preparation tools in order to be able to program diffusion-weighted sequences using trapezoid or arbitrary-waveform diffusion encoding gradients, combined with a fat saturation pulses. GinkgoSequences has been made compatible with the Gadgetron¹ tool for reconstruction and has already been used to program diffusion-weighted spin echo and STEAM echo-planar imaging sequences, which have successfully been tested on phantoms and on human volunteers.

References

1. Hansen MS, Sørensen TS. Gadgetron: an open source framework for medical image reconstruction. Magn Reson Med. 2013 Jun;69(6):1768-76. doi: 10.1002/mrm.24389.

Abstract

As scanner technologies continue to advance, the acquisition of high resolution multivariate data increasingly facilitates more quantitate measures of physical systems. However, these more complex data require increasingly advanced computational methods for analysis and visualization. This applies to standard neuroimaging techniques such as MRI and EEG that are ubiquitous in both basic science research and in clinical settings as well as to other fields such as meteorology where imaging is performed by Doppler radar. Despite the ubiquity and importance of these methods, there remain significant challenges in extracting quantitative information from these data. In this talk I will discuss some recent work on new theoretical and computational methods for 1) The analysis of multimodal MRI data including structure from high resolution anatomical MRI, local physiology and structural connectivity from diffusion MRI, and detection of functional brain networks from functional MRI; 2) Decoding and spatially resolving the EEG signal. Time and interest permitting, I will give a brief overview of meteorological applications of our methods.

Abstract

In this talk, I will present an overview of ongoing projects in my lab to extend the GRASP (Golden-angle RAdial Sparse Parallel) MRI framework for rapid continuous imaging. In the first part, I will describe a technique for 4D GRASP MRI at sub-second temporal resolution, which can be applied for imaging motion (or motion plus contrast changes) in real time. I will also describe the extension of this imaging method to a new framework, called MR motion fingerprinting, towards volumetric imaging with sub-second imaging latency that will be required in MRI-guided radiotherapy or interventions. In the second part, I will describe magnetization-prepared GRASP MRI (MP-GRASP), a framework for rapid motion-robust quantitative imaging and multiparametric imaging. Specific examples of MP-GRASP will include GraspT1, GraspT1-Dixon, GraspCEST etc. Implementation of these techniques using deep learning (e.g., DeepGraspT1) will also be described. In the third part, I will briefly summarize our recent effort to extend GRASP-MRI to other sampling trajectories and applications, including golden-angle Cartesian (for imaging non-moving organs), golden-angle spiral (for imaging the lung) and golden-angle rotated PROPELLER/BLADE/EPI (for diffusion imaging).

Abstract

In this talk I will discuss the rationale for thinking about high channel count arrays, the challenges in building them, and potential future directions.

Abstract

“Magnetic resonance imaging is a key diagnostic tool in modern healthcare, yet it can be cost-prohibitive given the high installation, maintenance and operation costs of the machinery. There are approximately seven scanners per million inhabitants and over 90% are concentrated in high-income countries. We describe an ultra-low-field brain MRI scanner that operates using a standard AC power outlet and is low cost to build. Using a permanent 0.055 Tesla Samarium-cobalt magnet and deep learning for cancellation of electromagnetic interference, it requires neither magnetic nor radiofrequency shielding cages. The scanner is compact, mobile, and acoustically quiet during scanning. We implement four standard clinical neuroimaging protocols (T1- and T2-weighted, fluid-attenuated inversion recovery like, and diffusion-weighted imaging) on this system, and demonstrate preliminary feasibility in diagnosing brain tumor and stroke. Such technology has the potential to meet clinical needs at point of care or in low and middle income countries.”

Abstract quoted from
“A low-cost and shielding-free ultra-low-field brain MRI scanner”
Nature Communications
doi: 10.1038/s41467-021-27317-1.

No abstract was provided for this talk.

Abstract

This talk will provide an overview of Gregor’s research experience in MRI at Siemens Healthcare, including efforts to make an MR fingerprinting implementation robust and ready for clinical research, and explorations of further capabilities of the method. The talk will also explore recent developments in deep learning image reconstruction methods for turbo spin echo sequences and concepts for aiding radiologists by automatically detecting and diagnosing pathologies in MR images in the musculoskeletal domain.

Abstract

I will talk about a new concept to accelerate parallel imaging by using dynamic instead of static receive coil sensitivities. This method is described by Felix Glang et al. in a recent paper titled “Accelerated MRI at 9.4 T with electronically modulated time-varying receive sensitivities”, published in Magnetic Resonance in Medicine. I will also show some novel concepts of how this approach can be extended to dynamic receive dipoles.

Abstract

High-resolution MR imaging is important for the qualitative and quantitative analysis of brain structures. Unfortunately adverse subject motion during the acquisition introduces image blurring, lowers the image quality and the effective spatial resolution. The acquisition of fat-navigators have enabled the correction of motion-induced blur. Alongside developments in data acquisition, camera-guided 3D tracking to enable markerless motion-correction has been commercialized. In this study, we aimed to investigate the impact of different types of head motion on brain MR images and compare the retrospective motion correction using fat navigators and markerless tracking devices.

Abstract

Alterations in the default mode network (DMN) are known to be associated with aging and with neurological and psychiatric diseases. We assessed age-dependent changes in interactions within and between the DMN and other brain areas and correlations of these interactions with a battery of neuropsychological tests to formulate a macroscopic model of aging.

Using a novel multivariate analysis method on resting-state functional MRI data from 50 young and 31 old healthy individuals, we identified directed intra- and inter-DMN pathways that differed between the groups and used correlations with neuropsychological tests to infer behavioral meaning.

We observed that visual and limbic inter-DMN pathways in old subjects engaged at low frequency, involved the dorsal posterior singulate cortex (PCC), and correlated with reduced attention and working memory. In contrast, in young subjects they were at high frequency and involved the ventral PCC. Sensorimotor-DMN pathways were efferent in young subjects and afferent in old subjects, with the latter correlated with reduced attention and working memory.

We suggest a macroscopic model of aging centered in the DMN. The model implies that the reduced sensorimotor efferent brought about from reduced physical activity and the increased need to control such activities by the medial prefrontal cortex (mPFC) causes a higher dependency on external than on internal cues. This results in a shift from ventral to dorsal PCC of inter-DMN pathways. Consequently, one way to slow these processes would be by increasing sensorimotor activity, therefore stressing the critical importance of physical activity and suggesting how it might slow aging.

Abstract

Alzheimer’s disease (AD), the most common cause of dementia in the elderly, is clinically characterized by impaired cognitive function and memory loss. While plasma, cerebrospinal fluid and PET imaging can provide biomarkers of the following AD pathological hallmarks: extracellular beta-amyloid (Aß) deposits (“A”), intracellular tau protein tangles (“T”), and atrophy due to synaptic and neuronal loss [“(N)”], pathophysiological processes at the earliest stages of the AD continuum are still largely unknown. Furthermore, AD is recognized as a heterogeneous disorder, whereby patients who fit the clinical criteria for AD may differ in levels of AT(N). To this end, proton magnetic resonance spectroscopy (¹H-MRS) can shed light on neurochemical changes that may precede AT(N), and yield complementary biomarkers of AD pathology. In this talk, I will present preliminary findings from a 1H MRS study in cognitively normal subjects. Using MRSI EPSI for whole-brain coverage, we measured metabolite values from cortical gray matter regions implicated in the tau Braak pathway (which has a specific spatio-temporal spread): posterior cingulate (Braak IV), precuneus (Braak V), cuneus (Braak V); and a negative control region, the lateral occipital gyrus. Using sLASER for localization and improved quantification in the temporal lobe, we measured metabolite values from the left hippocampus (Braak II). For all metabolites in all regions, I examined (1) correlations between metabolite levels and atrophy, using morphometry metrics from structural MRI, (2) correlations between metabolite levels and CSF p-tau181, and (3) metabolite differences between APOE4 carriers and non-carriers.

About the Speaker

Anna Chen is a third-year graduate student in Vilcek Institute’s biomedical imaging and technology PhD training program. Anna is advised by Ivan Kirov, PhD. She has a background in cognitive neuroscience, and is interested in using MRS techniques to better understand brain metabolism in disease.

Abstract

Accelerated MRI has made MRI exams faster and more affordable, making it possible to investigate new diseases and physiological processes in the human body. Artificial intelligence and machine learning tools are now common tools for image reconstruction and analysis. However, just recently these tools have been used to learn how to improve the MRI acquisition.

In this talk, we briefly review the evolution of undersampled acquisitions and the new machine learning tools for this task (such as LOUPE, BJORK, and), emphasizing the new tools that have been developed at NYU to improve the MRI acquisition for accelerated MRI. Our recently developed Bias-Accelerated Subset Selection (BASS) algorithm has improved the learning speed of the sampling pattern in compressed sensing and deep learning image reconstruction, allowing us to push even further the limits of acceleration in MRI.

Abstract

Biophysical modeling of the diffusion MRI signal offers the exciting potential of bridging the gap between the macroscopic MRI resolution and the cellular level tissue microstructure, effectively turning our MRI scanner into a noninvasive in vivo microscope. In the brain white matter (WM), the standard model (SM) of the diffusion signal was proposed as a general framework unifying many previous WM models. However, careful histological validation is required. Previous efforts used histology from several modalities of microscopy to quantify tissue metrics, and used them to evaluate parameters obtained from diffusion MRI. Yet, a comprehensive histological validation of the SM so far has been lacking. We used segmented 3D electron microscopy and ex-vivo diffusion MRI to characterize sham and injured rat brain white matter microstructure and perform a comprehensive histological validation of the sensitivity and specificity of the SM parameters. The large number of segmented 3D axons, in the order of ten thousand per sample, allowed us to better quantify tissue properties compared with previous studies.

Abstract

The first near-infrared (NIR I) wavelength range of 650-950 nm is preferable in many optical biological imaging techniques due to reduced light absorption by hemoglobin and water. I will present in vivo functional neuroimaging findings from optical and optoacoustic imaging in the NIR I range. My work utilizes novel dyes and genetically encoded calcium indicators as well as endogenous hemodynamic signals.

About the Speaker

Sarah is a doctoral candidate in the laboratory of Shy Shoham, PhD, at NYU Langone’s Tech4Health Institute. Her research is currently focused on functional neuroimaging with optoacoustic tomography.

Abstract

Recognition is a fundamental cognitive function during which the brain projects perceptual templates learnt from past experiences onto the current sensory input. A key factor in this process is how much the brain weights prior knowledge versus the sensory input. This weighting is known to vary between individuals and may contribute to hallucinations in mental illnesses when prior knowledge is weighted to a pathologically high degree. By manipulating human subjects’ ability to use prior knowledge to recognize stimuli during fMRI and electrocorticography neural recording, this project examines the underlying neural mechanisms of prior knowledge deployment across sensory modalities and distinct sources of prior knowledge. Specifically, this project examines visual and auditory recognition, and prior knowledge derived from lifelong learning as well as one-shot learning. This talk will present initial results and plans to integrate data from multiple experiments.

About the Speaker

Jonathan Shor is a fourth-year graduate student in the Vilcek Institute’s Biomedical Imaging and Technology PhD Training Program working with Dr. Biyu He. Shor has a background in computer science and is interested in developing computational models of cognitive functions, such as conscious perception. His current focus in the Perception and Brain Dynamics Lab is establishing the neural mechanisms driving prior knowledge deployment during recognition.

Abstract

White matter bundles underline structural connectivity in the human brain: they link functional cortical and subcortical gray matter areas and thus, they allow complex interactions between these regions. Awake brain surgery is a unique opportunity to study these white matter fibers, to which we have physical access inside the post-resection cavity. In the present protocol, we study the structural connectivity using diffusion MRI – tractography and the effective connectivity using electrocorticography – subcortico-cortical evoked potentials. We attempt to correlate biomarkers provided by these two methods in order to better understand the propagation of electrical activity in the brain.

About the Speaker

Petru Isan is currently undertaking a Master’s degree in Neurosciences at the Paris Est-Creteil University. He has graduated from the Tours Faculty of Medicine and is a second year Neurosurgery resident at the Pasteur 2 University Hospital in Nice, France. At the moment, Petru is doing an internship at INRIA Sophia-Antipolis, where he is investigating how different brain regions communicate with each other.

Abstract

Digital breast tomosynthesis (DBT) is a new imaging technique in mammography. Even though it is proven to be even more accurate than full-field digital mammography, false-positive recalls are still a subject of concern in the breast cancer screening setting. We developed an AI system using the screening mammography exams at NYU Langone Health which could save 29% of unnecessary recalls and potentially reduce radiologist workload by 40% while missing no malignancies. Specifically, our system consists of deep neural networks trained on both breast-wise pathology labels and a limited amount of pixel-level segmentation labels. The system can also highlight the location of suspicious findings on 2D and 3D mammography images for AI decision support.

Abstract

Biophysical models provide specificity to the tissue microstructure with diffusion MRI. In the brain, the Standard Model (SM) of diffusion in white matter (WM) was proposed as an overarching framework unifying many previous WM models. To stabilize the parameter estimation in clinical datasets with limited information, different constraints have been adopted to the WM models, resulting in different outcomes. Meanwhile, a machine learning (ML) model has shown promise in estimating SM parameters without constraints. To evaluate these WM models, we first compare the accuracy and specificity of them in simulation. Then we apply these models to early brain development, multiple sclerosis and stroke. Through this extensive comparison both in simulation and in several pathologies or processes, our goal is to determine the most reliable WM model for clinical datasets to extract tissue microstructure parameters.

About the Speaker

Ying Liao is a doctoral candidate in Vilcek Institute’s Biomedical Imaging and Technology training program working with Els Fieremans, PhD, and Dmitry Novikov, PhD. He has a background in biomedical engineering and is interested in developing and employing machine learning (ML) methods to characterize tissue microstsructure. His focus in the MRI biophysics lab is the ML-based estimation of white-matter parameters in the standard model of diffusion MRI.

Abstract

Structural neuroimaging is central to MRI’s role in both clinical practice and neuroscience. In addition to its role as a diagnostic modality, structural MRI provides the basis for cortical morphometric analyses that are widely used to study both developmental and degenerative processes. However, many populations of interest are often unable to remain still enough to produce the high-quality structural MRI needed for either clinical interpretation or quantitative morphometry. Moreover, while group differences in MRI-based morphometry have been presented across a variety of populations, it is generally difficult to extend these results to single-subject diagnoses.

I will present my ongoing work towards making structural neuroimaging methods both more robust, and more sensitive to disease processes. First, I will show how we have improved morphometric accuracy by studying motion as a source of bias and developing methods to help ameliorate these errors. As an example of this work, I will present refinements to our motion correction system for application in the upcoming HEALthy Brain and Child Development study. Second, I will show how we are developing intra-cortical measures by studying Frontotemporal Lobar Degeneration (FTLD) using joint ex vivo MRI and histopathology. I will present our recent findings of iron-rich pathology within the cortical laminae in FTLD, and discuss our plans for translating these results to in vivo imaging studies with the goal of single-subject diagnosis.

Abstract

Aging is a major risk factor of neuronal loss and cognitive impairment (e.g., Alzheimer’s disease). Neurovascular abnormalities and brain atrophy are proved to be pathophysiological biomarkers in normal- and abnormal-aging brains. In vivo detection of microvascular changes underlying neurodegeneration plays a crucial role in early diagnosis and better understanding disease mechanism in age-related dementia. Susceptibility weighted imaging (SWI) and quantitative susceptibility mapping (QSM) are sensitive to the deoxygenated hemoglobin in the veins, which can be used to map the venous vasculature and characterize venous (de)oxygenation level in the brain. This presentation will demonstrate two studies that use high resolution SWI and QSM 1) to characterize the venous oxygenation/utilization changes related to neurodegeneration in the elderly; 2) to map the in vivo venous vasculature of hippocampus at 7T. These studies aim to gather more evidence on the role of SWI/QSM as an early imaging marker of age-related neurodegenerative diseases.

Abstract

White matter changes are a frequent observations in the aging human brain and are considered a reliable indicator for cognitive impairment and long-term functional decline. On T2-weighted fluid attenuated inversion recovery magnetic resonance images, these lesions appear as white matter hyperintensities (WMH) and are commonly associated with vascular degeneration. From a physics perspective, however, the persistent (onset) locations of periventricular WMHs along the edges of the lateral ventricles suggests involvement of mechanical (over)loading of the ependymal cells forming the functional brain-fluid barrier. We use computational modeling to systematically explore the relationship between brain aging, white matter changes, and WMH formation. To that end, we build anatomically accurate brain models and predict the mechanical loading of periventricular tissues. We observe that maximum ependymal cell stretch consistently localizes in the anterior and posterior horns irrespective of ventricular volume or shape. More importantly, these locations coincide with periventricular WMH locations observed in our patient scans. From these results, we pose that further analysis of white matter pathology in the periventricular zone that includes a mechanics-driven deterioration model for the ventricular wall.

Abstract

Next month NYU will be the 6th facility in the US to receive an FDA approved CT scanner with photon counting capabilities. Photon counting CT represents a new paradigm in CT scanning. Compared to conventional CT where the energies of the photons incident on the detector are reported as a total sum, photon counting CT measures the energies of individual photons (i.e., “counts” them). In this talk I will describe the fundamental physics of photon counting CT, the implications of this approach to scanning, and its clinical benefits.

Abstract

Iron is critical for healthy brain biochemistry and function. While deficient peripheral iron was found to increase psychiatric morbidity risk, in vivo examination of non-heme brain iron in psychotic spectrum disorders (PSD) are lacking. The current study employed quantitative MRI to examine iron content in several iron-rich subcortical structures in a young adult PSD group. Using a modified cross-relaxation imaging method, we fitted R1 and macromolecular proton fraction maps and estimated region-wise R2* values using a linear regression model. Our findings suggest that subcortical non-heme iron deficiencies play a role in PSD risk and symptomatology and may precede both structural and myelin alterations.

Abstract

DCE-MRI is a critical imaging method used in cancer diagnosis. However, it is limited by long reconstruction time, and has inherent trade-off between temporal and spatial resolution. While machine learning has been shown to improve MR reconstruction quality in several studies, these methods require ground truth image data for training. For DCE-MRI reconstruction, we do not have access to a simultaneously high temporal and spatial resolution ground truth image. In this talk, I’m going to introduce a novel pipeline to simulate realistic ground truth training images based on pharmacokinetic models and anatomical structure. We trained the machine learning model to reconstruct images by using the simulated k-space and ground truth images. The result shows, with different models, machine learning models could reconstruct images with high image quality in less time. This simulation pipeline is available online, suitable for future development and exploration.

Abstract

Biophysical modeling of diffusion MRI data is appealing due to its potential to provide specificity to pathological processes. However, robust parameter estimation of the Standard Model (SM) of diffusion in white matter has been elusive due to intrinsic model degeneracies. Machine learning approaches improve parameter estimates but at low SNR these are determined by the training data. We develop a theory to analyze this behavior as function of SNR. Finally, we use these results to explore the design of optimal scanner-specific protocols to enable SM estimates in 15-minute acquisitions where we show reproducible results. Combining protocol optimization and robust parameter estimation may enable quantitative microstructure mapping in clinical settings.

Abstract

Visualizations of white matter fibers are reconstructed in vivo from diffusion MRI through tractography. To this end, dedicated reconstruction techniques need to identify spatial orientations of fibers, typically by seeking maxima of orientation distribution functions (ODFs). However, commonly used methods often fail to reconstruct fibers crossing at shallow angles below 40 degrees. We aim to break this barrier with our proposed approach called ODF-fingerprinting (ODF-FP).

In this talk, I will introduce the concept of ODF-FP and the process of generation of ODF dictionaries that covers biologically plausible microstructure parameters. I will present the accuracy of crossing fibers reconstruction with ODF-FP in numerical simulations, diffusion phantoms, and the mouse model. In the latter, I will show the ODF-FP recontruction of the optic pathways from in vivo diffusion MRI acquisition validated with the manganese chloride enhancement of the reconstructed tracts.

Abstract

Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) has a very high sensitivity in detecting breast cancer, but it often leads to unnecessary biopsies and patient workup. In this project, we developed and used an artificial intelligence (AI) system to improve the overall accuracy of breast cancer diagnosis and personalize management of patients undergoing DCE-MRI. In a clinical validation study, the AI system was found to be statistically equivalent to 5 board-certified breast radiologists. Radiologists’ performance improved when their predictions were averaged with AI’s predictions. We demonstrated the generalizability of the AI system using multiple data sets from Poland and the U.S. In subgroup analysis, we observed consistent results across different cancer subtypes and patient demographics. We showed that the AI system can reduce unnecessary biopsies in the range of clinically relevant risk thresholds. This would lead to avoiding benign biopsies in up to 20 percent of all BI-RADS category 4 patients. This exploratory work creates a foundation for deployment and prospective analysis of AI-based models for breast MRI.

Abstract

MRI technology is being continuously developed and new MRI scanners are installed every year. However, most of them are used clinically, without being available to researchers. Furthermore, the majority is in western countries. As a result, access to MRI research and training has been limited.

Several open-source software tools are available to simulate different aspects of the MRI experiment, from hardware design to signal encoding and image reconstruction. However, their dissemination has been limited because they are not general enough, often lack documentation, and require extensive background knowledge. Nevertheless, they have enormous potential.

The goal of this project is to integrate, generalize and extend this existing software to develop a comprehensive open-source software platform to simulate the complete lifecycle of an MRI experiment. By relying on a web-based graphic user interface and cloud computing, Cloud MR will enable anyone with an internet connection to perform MRI research anywhere in the world. Furthermore, it will be a unique tool for MRI training, which is increasingly needed due to the clinical widespread use of MRI.

Abstract

Basic understanding of brain metabolite diffusion as measured using diffusion-weighted magnetic resonance spectroscopy (DW-MRS) in vivo has progressed over the last years, allowing relevant interpretation of DW-MRS in terms of cellular microstructure. When combined with adequate modeling, DW-MRS may even allow cell-specific (neuron versus astrocytes) quantification of some microstructural parameters. We will see how recent results suggest that DW-MRS may actually open perspectives beyond cell structure determination, namely characterizing diffusion in the extracellular space, and assessing the distribution of brain lactate between neurons, astrocytes and the extracellular space, which is a highly relevant neuroscience question related to energy metabolism and the astrocyte-to-neuron lactate shuttle.

Abstract

Spiral MRI has been known since 1983. Spiral trajectory offer an extremely fast and efficient way to cover two-dimensional k-space with an intrinsically one-dimensional trajectory – much more efficeint than the line-by-line scanning of commonly used Cartesian sampling. Spirals have additional advantages like intrinsic motion compensation. They still haven’t made it into clinical routine due to their extreme sensitivity against deviations of the actual from the nominal trajectory and against off-resonance effects, where even slight inhomogeneities due to susceptibility effects lead to strong image artifacts.

The presentation will discuss principles and implementation of single shot spiral TSE at 1.5 and 3 T. High-quality images with 1 mm in-plane resolution are acquired in < 200 ms allowing extremely fast screening e.g. in non-cooperative patients.

Abstract

Hybrid-state Free precession (HSFP) is a quantitative transient state technique that allows the rapid mapping of relaxation times, in part, due to the high signal which results from a fully balanced sequence design. However, this design makes the transient magnetization of HSFP highly susceptible to disruptions caused by eddy-current induced phases. These eddy current artifacts in balanced sequences result from large jumps in k-space. The 3D kooshball HSFP sequence samples the spin dynamics repeatedly while acquiring different parts of k-space. We swap individual k-space lines between different repetitions in order to minimize jumps within each repetition. This reordering can be formulated as a traveling salesman problem, and we tackle the discrete optimization with a simulated annealing algorithm.

In the second part, HSFP based quantitative MRI is applied at 0.55 T. Here, the SNR-efficient design of HSFP allows quantitative maps to be obtained in 12 mins with 1 mm isotropic resolution. Further, a theoretical analysis of the Cramér–Rao bound is performed and the expected SNR is compared to 3T.

Abstract

The disruption of blood-brain barrier (BBB) is associated with various pathologies in the brain. Dynamic contrast-enhanced MRI has been widely used as a tool for quantitatively measuring the changes in the microstructural environment in brain. In recent year, there has been increasing interest in measuring the BBB disruption in Alzheimer’s disease (AD) and normal aging. However, unlike diseases that exhibits substantial changes in brain such as brain tumor, the BBB disruption in AD or aging is suggested to be very subtle.

We utilize Golden-angle RAdial Sparse Parallel (GRASP) sequence to effectively measure this subtle disruption and to overcome the following 3 challenges: (a) To reduce the long scan time to observe small extravasation of the contrast agent, (b) To obtain the arterial input function using the help of AI, and (c) To validate and measure the sensitivity of current approach in different levels of BBB disruption.

To address these challenges, we have (a) developed a novel pharmacokinetic model suitable for measuring BBB disruption with reduced scan time, (b) trained and implemented a deep neural network to deterministically estimate the capillary-level input function, and (c) conducted an animal study to artificially induce the different levels of BBB disruption and compared the sensitivity of measuring subtle disruption via the conventional Gadolinium contrast agent exchange rate and the water exchange rate using Ferumoxytol contrast agent.

Abstract

Active multi-disciplinary research is ongoing to discover quantitative biomarkers for early diagnosis, monitoring and assessment of joint degeneration. Medical imaging has played a substantial role in this area; for example, radiographs can detect structural alterations in bone, but these scans have low sensitivity for detecting tissues that are thought to be important in joint degeneration in OA (such as cartilage, menisci and other soft tissues) and cannot capture changes in bone marrow (that is, bone marrow lesions). Conversely, magnetic resonance imaging (MRI) has higher sensitivity than radiography in detecting soft tissue changes, bone marrow oedema and early osteophytic changes. Advanced quantitative imaging techniques, novel computerized image post-processing and more recent machine learning (ML) techniques have made possible further advances towards quantitative characterization of early joint degeneration and identification of imaging biomarkers associated with OA. Deep learning advances are revolutionizing the use of MRI in clinical research by augmenting activities ranging from image acquisition to post-processing. Automation is key to reducing the long acquisition times and processing of MRI, conducting large-scale longitudinal studies, and quantitatively defining morphometric and other important clinical features of both soft and hard tissues in various anatomical joints. Deep learning methods have been used recently for multiple applications in the musculoskeletal field. Compared with labor-intensive human efforts, DL-based methods have advantages and potential in all stages of imaging as well as post-processing steps, including aiding in diagnosis and prognosis. In this talk, I’ll explore how recent applications of DL have improved imaging-based understanding of knee OA. We illustrate how DL techniques are applied at all stages of imaging to enable automation of acquisition analysis and new imaging biomarkers discovery.

Abstract

Diffusion weighted imaging is a powerful technique sensitive to tissue microstructure. Three possible applications of this technique will be presented: (a) to early detect slow growing diffuse glioma models; (b) to evaluate the effect of ageing on mouse muscle microstructure in dystrophic and healthy mice; (c) to image the motor unit activity in the human muscles.

(a) Diffuse gliomas (WHO grade II to IV) are the most common primary brain tumours in humans. Their diffuse infiltration into the surrounding normal brain precludes complete resection and they all eventually recur, usually having progressed to a more aggressive tumour. The infiltrative part, which is “invisible” using conventional T1 and T2-weighted MRI is difficult to target with treatment. We investigated whether diffusion MRI might be a useful method to detect the microstructural changes induced in the normal brain by the slow infiltration of glioma sphere cells. Localized proton MR spectroscopy of lesions and immunohistochemical assessment were compared with imaging results.

(b) During postnatal development, muscle fibres grow enormously and the sarcolemma dynamically and constantly expands. Investigating this process in the time up to maturity may help the understanding of clinical onset of infant myopathies, such as Duchenne muscular dystrophy.The evolution of hindlimb muscle microstructure between young (development) and adult mice was investigated in dystrophic and healthy muscles using diffusion-weighted imaging protocols and histology. Multiple diffusion times (range: 25 – 350 ms) were explored, and significant differences between the diffusion properties of hindlimb muscles in healthy and diseased mice were found for long diffusion times, with increased water diffusivity in dystrophic mice. Muscle fibre size distributions showed higher variability and lower mean fibre size in dystrophic than wild-type animals. The extensive uptake of Evans Blue Dye in dystrophic muscles revealed a substantial sarcolemmal damage, suggesting diffusion measurement as more consistent with altered permeability rather than changes in muscle fibres.

(c) Neuromuscular diseases can lead to characteristic changes in the anatomy of motor units (MU). A diffusion-weighted imaging protocol sensitive to contraction was developed by synchronizing in scanner-electrical stimulation of MU with a pulsed gradient spin-echo imaging sequence. The spatial pattern of muscle fibres forming different MU was imaged in six healthy controls and subsequently in four patients with confirmed amyotrophic lateral sclerosis (ALS). Florid fasciculation in ALS patients was revealed at a frequency higher than healthy controls.

Abstract

The single frequency nature of magnetic resonance (MR) allows the design and optimization of fast and robust algorithms for computational electromagnetics, based on integral equations. Specifically, surface integral equations (SIE) can be employed to analyze radio-frequency (RF) transmit-receive coils, while volume integral equations (VIE) can model the interactions between RF fields and biological tissues with finite electrical properties. The fast and accurate estimation of the above interactions is critical for optimal RF coil design at ultra-high-field MRI and to solve inverse problems based on MR measurements. For example, rapid SIE/VIE methods can be used to estimate tissue electrical properties, as in Global Maxwell Tomography, or to estimate coil sensitivities, as in Maxwell Parallel Imaging. This talk will describe our latest developments in this areas and present results for various applications of computational electromagnetics at ultrahigh field MRI.

Abstract

More than 12 million Americans currently carry a form of orthopedic, cardiovascular, or neuromodulation device and the number grows by 100,000 annually, driving medical implant market to reach $160 billion by 20221. It is estimated that 50%-75% of patients with active electronic implants will need to undergo MRI during their life time, with some needing repeating examination. Recent advances in device engineering have led to a new generation of electronic implants that are largely immune to MRI-generated static and gradient fields. Tissue heating from radiofrequency (RF) excitation fields, however, remains a major issue.

This talk will give an overview of recent advances in development of specialized MRI hardware for implant-friendly imaging with a focus on patients with deep brain stimulation implants. We then discuss the unique role of computer modeling in MRI safety assessment and recent success stories in guiding the surgical practice toward MR-friendly device implantation, and novel insights into RF heating of implants in new MRI platforms (e.g., vertical scanners). Finally, we will talk about feasibility of applying machine learning for real-time risk assessment of MRI in patients with conductive implants.

Abstract

There exists an abundance of neural network techniques that achieve impressive performance on various visual processing tasks. Most of these methods require full supervision and large annotated training datasets. Collecting such training datasets and annotations is often expensive and time-consuming, or not possible due to limitations on data privacy. In this talk, I discuss how weakly supervised learning and synthetic data generation can be used as a substitute when full training data annotations are not available. The proposed methodologies are evaluated on various computer vision and medical imaging benchmarks.

Abstract

Magnetic Particle Imaging (MPI) is an emerging noninvasive biomedical imaging modality that shows great promise for vascular and cellular imaging. MPI uses different physics from all conventional imaging modalities. MPI offers ideal “positive” tracer contrast, because human tissues produce zero MPI signal and tissue is magnetically transparent. The signal comes only from superparamagnetic iron oxide tracers (25-nm SPIOs). MPI has very high molar sensitivity because the SPIO magnetization is 51 million times stronger than the nuclear magnetization, M0, imaged in a 3.0T MRI. Indeed, we can even detect 1 micromolar [Fe] concentrations in seconds with quantitative, linear and positive contrast. MPI technology has not reached its true physics limit; we believe MPI could soon achieve single-cell sensitivity and 100-micron resolution with optimized tracers.

MPI applications today already compete with nuclear medicine studies on dose-limited sensitivity. But MPI offers a zero-radiation option for: tracking, cellular therapies [2,3]; pulmonary embolism detection with 3D ventilation-perfusion MPI, akin to Tc99m V/Q studies [4]; capillary-level noninvasive CBV & CBF for stroke or traumatic brain injury [5], or to rule out a traumatic gut bleed (akin to RBC-Tc99m scintigraphy) [6]. Cancer MPI is more challenging but soon could provide a noninvasive screening alternative to X-ray mammography for high-risk women with radiologically dense breast tissue [7]. We recently showed that antibody labeled SPIOs can track WBCs (akin to In-111 WBC studies) [9]. WBC-MPI could emerge to be the best method to image infection, inflammation or cancer, and a powerful tool for optimizing immunotherapies. An important advantage relative to scintigraphy GI bleed and V/Q studies (which can take up to 3 hours including prep and scan) is speed: the targeted magnetic tracers can be safely injected immediately from the refrigerator providing first scans in just a few minutes. A crucial advantage of WBC-MPI is zero radioactivity of the tracers. CAR-T and CAR-NK cell therapies cannot survive the radiation dose of In-111 and so you cannot use standard nuclear medicine tools to track these exciting immunotherapies. Indeed, Immuno-MPI could soon become medicine’s most powerful tool for optimizing immunotherapies. Our lab has recently developed a potential breakthrough in MPI that already shows dramatic SNR and spatial resolution dramatically (10-fold for SNR and linear resolution) [9]. Experiments and physical models show that chained SPIOs act like ferromagnetic particles, with remanence and coercivity. This is well-modeled as a positive, regenerative feedback control system akin to Schmitt trigger comparators. Moreover, the new tracers show enormous improvements in SNR and spatial resolution, allowing for up to 1000-fold reduction in voxel volume. Our new tracers are not superparamagnetic (SPIO); they are actually superferromagnetic particles [9]. We will show that superferromagnetic tracers could remove the final obstacle to human MPI, allowing for safe 1mm resolution in a human MPI scanner with cost comparable to a human whole-body 0.5T MRI scanner.

References

1. Gleich and Weizenecker Nature 435, 2005.
2. B. Zheng et al., Nature Scientific Reports 5, 2015.
3. B. Zheng et al., Theranostics, 2016.
4. X. Zhou, et al., Phys. Med. Biology 62, 2017.
5. R. Orendorff, et al., Phys. Med. Biology 62, 2017.
6. E. Yu et al. ACS Nano, 11 (12), 2017.
7. E. Yu, et al Nano Lett., 17 (3), 2017.
8. P. Chandrashkharan, Nanotheranostics, 2021.
9. ZW Tay, Small Methods, 2021.

Abstract

Acute and chronic human diseases including liver and lung diseases, cancer, cardiovascular diseases and virus infection, share common key determinants including inflammation and fibrosis. In order to facilitate early detection, staging, and treatment responses, it is essential to develop a non-invasive imaging methodology that will allow us to longitudinally map and quantify the dynamic changes of inflammation biomarkers, such as chemokine receptors and collagen expression, during disease progression and upon treatment. Here we report our recent breakthrough in optimization, characterization, formulation, and production of a set of novel human protein-based contrast agents (ProCA®s) pioneered by our team for both preclinical and clinical applications.

We have developed a human collagen-targeted MRI contrast agent (hProCA32.collagen) with optimized binding fibrosis specificity. hProCA32.collagen exhibits 6.7-fold and 13.7-fold higher binding affinities for collagen type I over types III and IV, respectively. Our developed inflammation and fibrosis biomarker-targeted contrast agents specifically delineate activation of several types of cells and can capture the pro-metastasis niche and fibrosis associated with fatty liver and tumor microenvironment. With newly enabled dual and multi-color MR imaging methodology (precision imaging by MRI, pMRI) at multiscale, we have achieved robust longitudinal detection of early-stage liver and lung fibrosis, as well as micro-metastasis quantification of molecular biomarker changes for staging and monitoring treatment responses. We are moving rapidly toward clinical applications in early detection, monitoring progression, image-guided intervention/treatment, and patient stratification against human diseases including NASH, ASH, IPF, COPD, and metastasis from multiple cancers.

Acknowledgement: This work was supported in part by the NIH grants R01DK126080, R33CA235319, R42CA183376, R42AA025863, UT2AA028659, and S10OD027045 to Jenny Yang

References

1. Salarian M, Turaga RC, Xue S, Nezafati M, Hekmatyar K, Qiao J, Zhang Y, Tan S, Ibhagui OY, Hai Y, Li J, Mukkavilli R, Sharma M, Mittal P, Min X, Keilholz S, Yu L, Qin G, Farris AB, Liu ZR, Yang JJ. Early detection and staging of chronic liver diseases with a protein MRI contrast agent. Nat Commun. 2019;10(1):4777. Epub 2019/10/31. doi: 10.1038/s41467-019-11984-2. PubMed PMID: 31664017; PMCID: PMC6820552.
2. Tan S, Yang H, Xue S, Qiao J, Salarian M, Hekmatyar K, Meng Y, Mukkavilli R, Pu F, Odubade OY, Harris W, Hai Y, Yushak ML, Morales-Tirado VM, Mittal P, Sun PZ, Lawson D, Grossniklaus HE, Yang JJ. Chemokine receptor 4 targeted protein MRI contrast agent for early detection of liver metastases. Sci Adv. 2020;6(6):eaav7504. Epub 2020/02/23. doi: 10.1126/sciadv.aav7504. PubMed PMID: 32083172; PMCID: PMC7007242.
3. Xue S, Yang H, Qiao J, Pu F, Jiang J, Hubbard K, Hekmatyar K, Langley J, Salarian M, Long RC, Bryant RG, Hu XP, Grossniklaus HE, Liu ZR, Yang JJ. Protein MRI contrast agent with unprecedented metal selectivity and sensitivity for liver cancer imaging. Proceedings of the National Academy of Sciences of the United States of America. 2015;112(21):6607-12. doi: 10.1073/pnas.1423021112. PubMed PMID: 25971726; PMCID: 4450423.
4. Turaga RC, Yin L, Yang JJ, Lee H, Ivanov I, Yan C, Yang H, Grossniklaus HE, Wang S, Ma C, Sun L, Liu ZR. Rational design of a protein that binds integrin alphavbeta3 outside the ligand binding site. Nat Commun. 2016;7:11675. Epub 2016/06/01. doi: 10.1038/ncomms11675. PubMed PMID: 27241473; PMCID: PMC4895024.

Abstract

Breast cancer is the most common cancer in the UK and in women globally. Imaging methods like mammography, ultrasound (US) and magnetic resonance imaging (MRI) play an important role in the diagnosis and management of breast cancer; they are generally utilized to provide anatomical or structural description of tumors in the clinical setting. It is widely accepted that the tumor microenvironment influences the phenotype, progression and treatment of breast cancer. This gave the impetus to move beyond tumor visualization in images to radiomics in order to provide additional disease characterization and early biomarkers of tumor response.

Due to their ability to assess physiological processes in vivo, positron emission tomography (PET) and MRI can provide non-invasive characterization of the tumor microenvironment, including perfusion, vascular permeability, cellularity and hypoxia, which is associated with poor clinical outcome and metastasis. Clinical imaging studies in breast tumors have hitherto assessed tumor physiological parameters separately, with only few directly comparing data from these modalities. To this end, hybrid PET-MRI represents an attractive option as it can allow examination of dynamic functional processes and features of tumors simultaneously, while also conferring methodological advantages to the way imaging information is combined.

The main aim of this work is to provide a better understanding of breast cancer pathophysiology using simultaneous PET and multi-parametric MRI. In particular, this work aims to explore relationships between imaging biomarkers of tumor vascularity measured by dynamic contrast-enhanced (DCE) MRI, cellularity using diffusion-weighted imaging (DWI) and hypoxic status using 18F-fluoromisonidazole (18F-FMISO) PET. Correlations between functional PET-MRI parameters and immunohistochemical (IHC) biomarkers of hypoxia and vascularity as well as MRI morphological tumor descriptors are also presented. This study concludes with an investigation of the utility of MRI markers of perfusion to quantitatively monitor and predict pathological response in patients undergoing neo-adjuvant chemotherapy (NACT) as surrogate markers of hypoxia.

Abstract

Effective treatment of perceptual disorders such as amblyopia requires that we pinpoint the site(s) of neural impairment. Classic results indicate that the reduced visual acuity and contrast sensitivity in amblyopia are associated with smaller cell bodies in the LGN of the thalamus and a weaker drive of activity in cortex. However, it is unknown if the LGN is the first site of neural impairment, or if earlier retino-thalamic projections are affected as well.

We used diffusion MRI to quantify the white matter integrity of the retino-thalamic pathway in amblyopes and age-matched controls. We found reductions in fractional anisotropy in both the optic nerve and optic tract compartments of the retino-thalamic pathway of amblyopes. Moreover, when comparing between anisometropic and strabismic subtypes of amblyopia, we found that much of the effect was driven by anisometropic amblyopia.

These results suggest that the perceptual deficits that characterize amblyopia are due in part to impairments in the earliest segments of the brain’s visual processing pathway. Moreover, the treatment of the disorder may require different interventions and timecourses depending on the type of amblyopia. Future work should separately consider the impact of anisometropic and strabismic amblyopia, and carefully re-consider if the optic nerve impairments may be detected using retinal imaging methods.

Abstract

Accurate segmentation of substantia nigra (SN) and red nucleus (RN) is challenging, yet important for understanding brain diseases like Parkinson’s disease (PD). This work proposes improved algorithms to segment SN and RN from T2^*-weighted images and quantitative susceptibility mapping (QSM) MRI. After optimising segmentation algorithms to produce reliable SN and RN segmentations, multiple MRI (QSM, R2^*, diffusion tensor imaging, arterial spin labelling, and volume) metrics extracted from the SN and RN are compared across groups (both PD/healthy controls and across cognitive subgroups) and investigate relationships with global cognitive ability and motor function in PD employing Bayesian regression models, and interesting evidence of associations is obtained. The multi-modal MRI features are also utilized to distinguish healthy controls and PD using a linear support vector machine (SVM) classifier. Therefore, multiple imaging modalities measuring complementary tissue characteristics such as iron deposition, microstructural alterations, perfusion changes, and volume atrophy may be useful for monitoring several ongoing processes in midbrain nuclei in Parkinson’s disease and also could be helpful for the distinction between PD and controls.

Abstract

There is currently no microscopy technique that qualifies as in vivo and non-invasive… or is there? It had been a long winding road, but diffusion MRI combined with biophysical modeling may just fill that role. In this talk, we will (i) go through the most recent validation steps of the white matter microstructure model and its applications to characterize white matter pathology, (ii) start from scratch to build a relevant model for cortical gray matter, and (iii) take a fresh look at diffusion functional MRI as a promising alternative to BOLD, particularly at low field strength.

Abstract

Light field (LF) photography has properties such as refocusing, perspective change, occlusion removal that yields a breakthrough in microscopy, 3-dimensional (3D) displaying and rendering, augmented and virtual reality improvements. We can extract all of these properties by LF post-processing. However, a high-quality LF is bulky that makes these post-processing computations time-consuming and challenging for real-time deployment. To reduce the wait-time for LF data compression, we have merged conventional image compression techniques with residual convolutional neural networks for LF compression to make the task of LF storing and streaming more than two orders of magnitude faster than the state-of-the-art. Also, we developed RefNet to extract a set of refocused images from the raw LF in real-time as an example of the credibility of the machine learning techniques for LF feature extraction. While our proposed RefNet is faster in estimating the refocused images than the classical methods, it is more robust than current state-of-the-art none-learning methods in color prediction where discretizing the image will cause artifacts.

Abstract

The act of recognizing a percept involves matching the sensory input with one’s prior knowledge to arrive at a best fit. When input is ambiguous, the best fit may be so poor that recognition fails. However, this threshold varies among individuals, up to pathological cases in which the weight given to prior knowledge is so great that individuals “recognize” things that are not there as in the case of hallucinations often suffered by schizophrenia patients. Distinct sources of prior knowledge have been identified to influence recognition, including one-shot learning, lifelong learning, and expectation, but the neural mechanisms underlying this process are poorly understood. By using two paradigms that manipulate the deployment of prior knowledge while recording neural activity with 7T fMRI and electrocorticography (ECoG), I will test for a common neural process modulating the weight given to one-shot learning and lifelong prior knowledge in visual processing tasks. In this talk, I will present these paradigms along with some preliminary data I have collected, discuss further analysis plans, and possible follow-up studies.

Abstract

Three-dimensional electron microscopy (EM) techniques have enabled acquiring images of hundreds of micrometers of tissue with synaptic resolutions—images whose size can range from gigabytes to several petabytes. Applying manual or semi-automated methods for tracing and analyzing individual ultrastructures, even for a small section in such datasets, consumes hundreds of hours of experts’ time.

We developed ACSON and DeepACSON pipelines to automatically segment the entirety of neuronal processes in multi-resolution EM volumes of white matter. In ACSON, we automatically segmented white matter ultrastructures in high-resolution small field-of-view EM volumes. In DeepACSON, we emphasized low-resolution EM imaging to cover larger fields of view where severe membrane discontinuities became unavoidable. DeepACSON performed convolutional neural network (CNN)-based semantic segmentation and cylindrical shape decomposition (CSD)-based instance segmentation. CSD is a top-down instance segmentation algorithm we designed to decompose under-segmented myelinated axons into their constituent axons, accounting for the tubularity of axons as a global objective.

ACSON and DeepACSON segmented hundreds of thousands of long-span myelinated axons, thousands of cell nuclei, and millions of mitochondria with excellent evaluation scores, enabling comprehensive 3D morphometry of the white matter ultrastructures and capturing nanoscopic morphological alterations in healthy and pathological brains.

Abstract

The sensitivity and specificity of brain MRI are limited by the low image encoding efficiency, leading to long acquisition time and limited spatial resolution especially for in vivo imaging. In order to address this, this talk will present our newly developed acquisition method, Echo Planar Time-resolved Imaging (EPTI), which uses novel encoding strategies in the high-dimensional space, together with efficient data sampling schemes, to allow better use of multi-channel receiver coil arrays and shared data correlation to achieve high acceleration capability.

EPTI has been extended and applied to improve the efficiency of quantitative relaxometry, functional and diffusion imaging. We demonstrate that the significantly improved imaging efficiency enables ultra-fast multi-parametric mapping at submillimeter isotropic resolution with an order-of-magnitude faster acquisition speed, functional MRI with higher neuronal specificity as well as dMRI with higher SNR efficiency and better structural integrity. The future application of the proposed techniques should improve the diagnosis power of clinical brain MRI and allow further understanding of the structural and functional organization of the human brain.

Abstract

Intracortical myelin is a critical feature of the cortical mantle that is assumed to closely relate to high-order cognitive and behavioral functioning. Its abnormalities have also been implicated in a myriad of psychiatric and neurodegenerative disorders including schizophrenia and Alzheimer’s disease. One of the challenges in studying the cerebral cortex is the presence of non-uniformly distributed microstructural features across cortical layers. In this talk I’ll discuss how we may utilize myelin variations across the cortex and characterize cortical myeloarchitecture using cortical profiles sampled from high-resolution MRI images. Findings from applying this method in out schizophrenia dataset and the Human Connectome Project Aging dataset will be presented.

No abstract was provided for this talk.

Abstract

Genetically encoded fluorescent calcium indicators are a crucial tool for preclinical neuroimaging. Most of these indicators have fluorescent excitation and emission ranges at visible wavelengths, with few reliable indicators existing in the biologically useful near-infrared range. In the past few years, some progress has been made on developing near-infrared indicators. I will present my ongoing in vivo work with NIR-GECO2G, a state-of-the-art near-infrared calcium indicator.

Abstract

The use of advanced imaging technologies has increased significantly in the past two decades and has revolutionized patients’ treatment on a daily basis, in terms of earlier and more accurate diagnosis. Essentially, no critical medical decisions are taken without relying on some sort of imaging. In the future, these decision-making processes will rely, to an even greater extent, on molecular imaging, in which personalized imaging probes, designed for specific medical conditions, will be used for diagnosis and to assess treatment success, by allowing clinicians to monitor therapy non-invasively and over time. Dr. Gilad research is in the intersect of synthetic biology and molecular imaging. where his lab is implementing the principles of synthetic biology to develop cutting edge technologies for better understanding the central nervous system and cancer. We bioengineer genetically encoded gene circuits and novel fusion protein based on unique properties adopted from a variety of organisms. The Gilad lab has been focusing on bioengineering of genetically encoded reporters for MRI mostly based on chemical exchange saturation transfer (CEST). Using protein engineering tools and machine learning algorithms, we have improved the sensitivity and expended the arsenal of reporters. These reporters were implemented in array of in vivo models with an emphasis on neuroimaging and cancer. We complement our reporters with genetically encoded optical sensors that allow detecting neurotransmitters.

Abstract

The GRASP project, started from 2011, is 10 years old today! GRASP MRI represents years of innovation and efforts by a research team consisting of MRI physicists, clinician scientists and industry partners. To date, GRASP MRI has been successfully demonstrated in many clinical applications; its overall performance has been greatly improved after years of optimization; and it has also been extended to a number of new variants. In this talk, Li will take this opportunity to summarize the GRASP developments over the past decade and to discuss future directions that GRASP MRI could potentially be heading to. Of course, in the era of artificial intelligence, how to make a smart version of GRASP by incorporating the latest deep learning technology is an important question we have to think and plan. If you are interested in hearing the latest of this project, you won’t want to miss this story.

Abstract

Traumatic brain injury (TBI) is a global health concern, with mild TBI (mTBI) accounting for 60%-80% of cases. TBI sequelae can be histologically explained by axonal varicosities known as diffuse axonal injury, but this pathology is not detectable using conventional CT and MRI. 1H MRS is a technique sensitive to neurochemical alterations which may enable more precise evaluation of TBI severity and prognostication when macroscopic structural damage is lacking. Unfortunately, varying results in regard to which metabolite(s) are most likely to be affected and what brain region(s) should be sampled, contribute to the limited clinical use of MRS in TBI. 1H MRSI has shed light on the regional distribution of metabolite findings, but a key part of translating the new knowledge to the clinic rests on determining how reproducible are the results of any particular study. This talk will present 1) initial data from a project intending to test the reproducibility of 1H MRSI findings from previous studies with a different mTBI cohort, 2) an outlook on future directions, as well as 3) recent findings from sodium imaging.

Abstract

Neurodegenerative diseases such as Alzheimer’s disease cause changes and disruption to cortical microstructure and architecture. MRI could potentially be sensitive to such changes. There is a growing interest in modelling human cortical areas using a combination of quantitative MRI and 3D microscopy ex vivo. This presentation contains a brief review of MR modalities that could be used for this purpose in addition to a Monte Carlo simulation study of DWI in light fluorescence microscopy samples.

Abstract

Our effort at NYU School of Medicine towards building deep neural networks for Digital Breast Tomosynthesis (DBT) volumes ranked #1 at the DBTex challenge. In this international challenge, we built an AI system to find biopsy-proven lesions from the DBT volumes collected from the Duke University Hospital. In this talk, I will discuss how our team was able to reach the best performance on the external dataset by utilizing our own private datasets at NYU Langone and how the model outputs could benefit the radiologists.

No abstract was provided for this talk.

Abstract

Water exchange between compartments in the brain (e.g., the vascular, ventricular, extracellular, and intracellular spaces) is a crucial biological process for maintaining homeostasis and may serve as a biomarker for diagnosis of structural and functional deficits. FEXI and DKI(t) are promising diffusion MRI techniques for measuring apparent exchange in the brain. FEXI employs a double-diffusion-encoding scheme to filter tissue compartment based on differences in diffusivities and measures the recovery in diffusion measurements over an increasing mixing time, characterized by Apparent Exchange Rate (AXR). DKI(t), on the other hand, measure apparent exchange based on its effects on the asymptotic decay of diffusion kurtosis described by the Kärger model. In this study, we investigated the relationship between FEXI and DKI(t) based measurements of apparent exchange in post-mortem mouse brains and elucidated its confounding factors in determining the desired exchange process.

Abstract

The parallel growth of obesity and diabetes has escalated over the last four decades placing over 1.9 billion overweight and obese individuals at increased risk of developing cardiovascular disease (CVD). This risk has been attributed to the pressure of a low-grade inflammatory state, but the mechanism underlying the inflammation is unclear. An increased epicardial adipose tissue volume or thickness quantified by echocardiography, computed tomography (CT) or magnetic resonance (MR) has been shown to correlate with cardiovascular disease and diabetes independent of anthropometric measurements such as body mass index. However, in visceral obesity, epicardial adipose tissue can assume a white adipose phenotype that is hypothesized to be associated with proinflammatory markers. The white adipose tissue may precede the accumulation of fat and increase in epicardial adipose volume. The objectives for this talk are to discuss the current theories of defining cardiovascular or cardiometabolic risk, what research has been done by others and our group that could leverage future utilization of imaging as a surrogate marker of identifying patients at risk for adverse CVD outcome. We will emphasize the research performed by our group to understand the correlation between the increased epicardial edipose fat fraction quantified by water-fat imaging and coronary artery disease including tissue inflammation defined by lipidome and transcriptome profiling in patients undergoing open-heart surgery. Epicardial, extrapericardial, and subcutaneous depots expressed different imaging, lipidome and transcriptome signatures. Furthermore, increased epicardial fat fraction positively correlated with coronary artery disease, tissue ceramides, a pro-inflammatory lipids, and proinflammatory gene expressions. We will discuss research questions and future direction of utilizing epicardial fat fraction to risk stratify CVD patients and monitor therapeutic response.

Abstract

Previous research has suggested both perfusion and free water (FW) alterations in Psychotic Spectrum Disorders (PSD), assessed independently of each other. To study PSD neuropathology, we applied a three-compartment IVIM-FWI model which disentangles FW diffusion and perfusion along with an anisotropic diffusion tissue compartment. The estimation of each of these metrics may be affected when the effects of the other are not taken into consideration. Previous histological studies have suggested an array of microvascular and microstructural deficits likely to impact perfusion and FW in PSD, including increased inflammation, morphological differences in capillaries, and disruptions in the neurovascular unit cells and the blood brain barrier. The aim of this research was to evaluate, for the first time, if the three compartment IVIM-FWI model can describe microvascular and microstructural changes in PSD in both gray and white matter. Additionally, we examined the relationships between the IVIM-FWI derived measures of perfusion fraction (PF), FW, and fractional anisotropy of tissue (FAt) and psychosis duration, cognition, and MR spectroscopy metabolites.

Abstract

Pulse design and reconstruction are two important topics in MR research for enabling faster imaging. On the pulse design side, selective excitations that confine signals to be within a small ROI instead of the full imaging FOV can promote sampling sparsity in the k-space, as a direct outcome of the change of the corresponding Nyquist sampling rate.

On the reconstruction side, besides improving algorithms’ capability on restoring images from less data, another objective is to reduce the reconstruction time, particularly for dynamic imaging. This talk presents our developments on these two perspectives: The first part introduces a pulse design framework built on our efficient auto-differentiable Bloch simulator. By propagating the derivatives in an automatic way, this tool connects excitation objectives (e.g., accuracy) directly to the pulse waveforms to be designed without approximations such as the small-tip model. It enables us to address excitation losses that are previously not approachable. We apply this tool on outer volume saturated inner volume imaging, which confines imaging signals into an ROI by selectively spoiling spin magnetizations outside.

Abstract

Renal tissue hypoxia is considered to be an important factor in the development of numerous acute and chronic kidney diseases. Blood oxygenation sensitized MRI can provide quantitative information about changes in renal blood oxygenation via mapping of T2*. Simultaneous MRI and invasive physiological measurements in rat kidneys demonstrated that changes in renal T2* do not accurately reflect renal tissue oxygenation under pathophysiological conditions. Confounding factors that should be taken into account for the interpretation of renal T2* include renal blood volume fraction and tubular volume fraction. Tubuli represent a unique structural and functional component of renal parenchyma, whose volume fraction may rapidly change, e.g., due to alterations in filtration or tubular outflow.

Diffusion-weighted imaging (DWI) provides a method for in-vivo evaluation of water mobility. In the kidneys intravoxel incoherent water motion may be linked to three different sources: i) renal tissue water diffusion, ii) blood perfusion within intrarenal microvasculature and iii) fluid in the tubules. The latter provides means to probe for changes in the tubular volume fraction. Recognizing this opportunity this presentation examines the feasibility of assessing tubular volume fraction changes using the non-negative least squares (NNLS) analysis of DWI data.

No abstract was provided for this talk.

Abstract

This talk will explore new ways to use local magnetic field control besides conventional “B0 shimming”. Perturbations of the main magnetic field (“B0”) due to tissue susceptibility interfaces are a long-standing obstacle in Magnetic Resonance applications. Inhomogeneous B0 fields can lead to artifacts such as geometric distortion, signal voids, poor RF pulse performance, and spectral line broadening. This has limited the use of diffusion, functional, and spectroscopic MR imaging in many regions of the brain and body. Recently, it has been shown that multi-coil arrays of independently-driven loops placed close to the body can generate nonlinear, high spatial-order field offsets to “shim out” unwanted susceptibility fields on a subject-specific basis, benefiting field homogeneity and image quality. In this talk, we explore the potential for repurposing multi-coil shim arrays for new applications that exploit their nonlinear, rapidly-switchable local field offsets. Examples include tailored field offsets for improved lipid suppression in MR spectroscopic imaging; zoomed functional MRI of target brain anatomy; flip angle correction at ultra-high field; and supplementary spatial encoding for improved parallel imaging. We will also explore ways to add local field control capability to coil arrays originally designed for other applications, such as RF receive arrays and Transcranial Magnetic Stimulation probe arrays, so that their degrees of freedom can be brought to bear.

Abstract

The presentation will provide a broad overview of the history of myelin water imaging in humans. Myelin water imaging is based on measurement of the short T2 component of water in brain and spinal cord tissue. What began as a lengthy single slice, single center measurement has expanded to many countries on multiple continents in just over 25 years. Important work along the way has included post-mortem validation studies in human CNS tissue, comprehensive assessment of development and normal characterization in adults, as well application to many neurological diseases including multiple sclerosis, concussion, stroke and beyond. The creation of normative atlases and development of faster analysis approaches promises to help move myelin water imaging to clinic in the coming decade.

Abstract

Parkinson disease (PD) is a neurodegenerative disorder characterized pathologically by nigrostriatal dopaminergic terminal loss and the development of Lewy pathology in surviving neurons of the substantia nigra (SN). Lewy pathology extends beyond the SN, and can be found in limbic and prefrontal cortical regions associated with cognitive decline. In vivo assessment of cortical microstructure and the extent of pathological changes will be clinically useful to monitor disease progression. For this purpose, our study used two diffusion MRI models, diffusion tensor imaging and neurite orientation dispersion and density imaging, to study the microstructural changes in the cerebral cortex of PD participants (n=18) compared to healthy controls (n=8). We demonstrate that in the absence of cortical thinning, PD pathology is associated with significant abnormalities in cortical diffusion metrics. Specifically, we found that the anterior cingulate cortex and inferior temporal lobe are consistently involved in PD through reductions in the intracellular volume fraction, fractional anisotropy (FA) and increased orientation dispersion index. FA reductions were extensive and involved more limbic areas such as entorhinal cortex, parahippocampus and insula. These findings are consistent with the presence of Lewy pathology in limbic regions and might be reflecting the earliest stages of tissue involvement in PD.

Abstract

Cardiovascular Magnetic Resonance (CMR) is a valuable tool that enables non-invasive characterization of tissue and assessment of cardiac function. Parametric mapping techniques play an important role in CMR due to their sensitivity to physiological and pathological changes in the myocardium. The capability of mapping T1 and T2 simultaneously in a single scan makes the novel cardiac Magnetic Resonance Fingerprinting (cMRF) technique a promising technology to facilitate diagnosis and treatment evaluation in various cardiac diseases. Unlike conventional parametric mapping approaches which may yield different T1 or T2 values for the same subject depending on the specifics of the MRI system hardware or pulse sequence implementation, cMRF has the potential to offer reproducible measurements of tissue properties on all MRI scanners. This talk aims to introduce the basics of the cMRF technique, including pulse sequence design, dictionary generation, and pattern matching, as well as highlighting potential applications.

Abstract

Commercial magnetic resonance imaging (MRI) systems cost millions of euros to purchase, require large electromagnetically shielded spaces to house, are extremely expensive to maintain and require highly trained technicians to operate. These factors together means that their distribution is confined to centrally-located medical centres in large towns and cities. Globally over 70% of the world’s population has absolutely no access to MRI, and clinical conditions which could benefit from even very simple scans cannot be treated. In the financially developed world, although MRI is diagnostically very important, the high cost and fixed nature prohibits any type of role in widespread health screening, for example. The magnetic fields typically used are very high, which means that there are severe contraindications so that, for example, MRI cannot currently be used in the emergency room. From the considerations above it is clear that if low-field MRI could be made more portable, accessible and sustainable then it would open up new opportunities in both developed and developing countries.

Rather than designing a highly sophisticated and expensive piece of equipment that can be used for all types of scanning, we use the philosophy of tailored design, such that we can design much more inexpensive systems for specific medical applications. Thus rather than one large MRI, the model is similar to having tens of different mobile ultrasound machines in a medical facility. In order to achieve portability, we design systems that use thousands of very small low-cost permanent magnets, arranged in designs which have no fringe field and therefore very easy siting requirements. The low magnetic fields allow scanning of patients with implants, and the scanner could potentially be transported on an ambulance for differentiation of hemorrhagic or ischemic stroke, for example. This talk will cover aspects of magnet, gradient and RF coil design for low fields (~50 mT) , as well as corrections for gradient- and B0-distortions, and present the latest in vivo results as well as an outlook on future developments.

Abstract

There is a pressing need to identify deterioration amongst patients with COVID-19 in order to avoid life-threatening adverse events. Chest radiographs are frequently collected from patients presenting with COVID-19 upon arrival to the emergency department, since it is considered as a first-line triage tool and the disease primarily manifests as a respiratory illness. In this talk, I will discuss the AI prognosis system we developed using data collected at NYU Langone Health to predict in-hospital deterioration, defined as the occurrence of intubation, mortality, or ICU admission. In particular, our system consists of an ensemble of an interpretable deep learning model to learn from chest X-ray images and a gradient boosting model to learn from routinely collected clinical variables, e.g. vital signs and laboratory tests. The system also computes deterioration risk curves to summarise how the risk is expected to evolve over time. The results of retrospective validation on the held-out test set, the reader study, and silent deployment in the hospital infrastructure highlight the promise of our AI system in assisting front-line workers through real-time assessment of prognosis.

Abstract

Less is known about the structure-function relationship in the human brain than in any other organ system. The challenge of studying brain structure is that brain networks span multiple spatial scales, from individual neurons all the way to whole-brain systems. Diffusion magnetic resonance imaging (MRI) holds great promise among noninvasive imaging methods for probing cellular structure of any depth and location in the living human brain. Robust methods for in vivo mapping of tissue microstructure by diffusion MRI remain elusive due to the demand for fast and strong diffusion-encoding gradients. I will present an overview of our group’s efforts to advance MR hardware, biophysical modeling, and validation of microstructural metrics derived from diffusion MRI in order to probe the structure of the human brain across multiple scales. I will review current progress and applications of these methods to study axonal microstructure in the normal and aging human brain and assess axonal damage in multiple sclerosis.

To bridge the divide between the neuroscientific and clinical use of MRI in probing tissue microstructure, this presentation will also provide an overview of our ongoing efforts to optimize, translate and validate novel encoding and reconstruction techniques for the ultrafast acquisition of high-resolution, multi-contrast MR images in a clinical setting. These efforts are exemplified in our recent work exploring the benefits of improved speed and resolution of ultrafast susceptibility-weighted imaging to study microvascular injury in patients with severe COVID-19 using radiologic-pathologic correlative examinations.

Abstract

Recently, methods employing single- and dual-frequency saturation are gaining recognition to detect events on microstructural and molecular level. Specifically, Chemical Exchange Saturation Transfer (CEST) employs selective saturation of the exchanging protons and subsequent detection of the water signal decrease to create images that are weighted by the presence of a metabolite or pH1. Here, we will describe aspects of translating CEST to reliable clinical applications at 3Tesla and discuss its potential uses in human oncology, specifically breast cancer. Second, we will discuss a method called inhomogeneous Magnetization Transfer2 (ihMT), which employs dual-frequency saturation to create contrast originating from the residual dipolar couplings and thus specific to microstructure. We will focus on principles of ihMT, its comparison to other white matter metrics (diffusion) and the methods application to the detection of myelin in brain and spinal cord.

Abstract

Due to a substantial shortening of the RF wave length (below 15 cm at 7T), RF magnetic field at UHF has a specific transmit (Tx) excitation pattern with strongly decreased (more than 2 times) values at the periphery of a human head. This effect is seen not only in the transversal slice but also in the coronal and sagittal slices, which considerably limits the longitudinal Tx-coverage (along the magnet’s axis) of conventional surface loop head arrays. In this work, we developed a novel human head UHF array consisted of 8 transceiver folded-end dipole antennas circumscribing a head. Due to the asymmetrical shape of the dipoles (bending and folding) and the presence of an RF shield near the folded portion, the array simultaneously excites two modes, i.e. a circular polarized mode of the array itself, and the TE mode (“dielectric resonance”) of the human head. Mode mixing can be easily controlled by changing the length of the folded portion. Due to this mixing, the new dipole array improves longitudinal coverage as compared to unfolded dipoles. By optimizing the length of the folded portion, we can also minimize the peak local SAR value and decouple adjacent dipole elements.

Abstract

This lecture will provide a brief clinical overview of SARS-CoV-2 infection and COVID-19 manifestations in the lungs. Imaging findings in the chest will be defined and literature reports summarized. Our evolving clinical experience will be described, including the subacute and chronic manifestations of COVID-19 lung disease we are now seeing. Finally, completed and ongoing thoracic COVID research projects will be presented.

Abstract

Magnetic Resonance Imaging has revolutionized the field of neuroscience by providing a non-invasive means to study the brain, to understand its organization, specialization and anomalies in an unprecedented manner. Despite the rapid advances in MRI instrumentation, it is still challenging to achieve high quality data in an efficient manner for several MR imaging modalities, especially for those modalities involving multi-dimensional imaging. In this talk, I will discuss several computational approaches that we have developed to achieve high efficiency MR imaging to enable many applications. These approaches strive to achieve high resolution, high SNR and artifact-free MRI by jointly optimizing the contribution of MR acquisition, the signal modeling under investigation and the reconstruction methods to provide meaningful information in an efficient manner. In this talk, I will focus the discussion mainly on diffusion magnetic resonance imaging and our work towards improving the efficiency of this modality.

Speaker Bio

Merry Mani received her PhD in 2014 from the University of Rochester, NY. Later in 2014, she joined the Magnetic Resonance Research Facility at the University of Iowa as a post-doctoral research fellow, where she developed new imaging methods on the 7T MRI. In 2019, she became an Assistant Professor in the department of Radiology, Carver College of Medicine, University of Iowa. Her lab focuses on integrating cross-disciplinary tools such as signal modeling and signal processing with imaging physics and image analysis tools to enable high efficiency MRI. These include the development of novel pulse sequences and optimization of sampling trajectories and reconstruction methods for maximum performance.

Abstract

Like standard gradients, nonlinear gradients modulate the magnitude of Bz as a function of position; the difference is that the magnitude as a function of position is generally not linear or unidirectional. One important consequence of gradient nonlinearity is that the modulation of spins is no longer sinusoidal, so MR data do not correspond to points in k-space. Therefore, early encoding strategies focused on optimizing sequences by considering encoding in the spatial domain. However, a k-space analysis of nonlinear encoding provides significant insights on sequence design and suggests novel strategies, such as FRONSAC encoding. With FRONSAC, most of the encoding comes from a standard linear trajectory (e.g. Cartesian, radial or spiral), but nonlinear gradients are used to effectively increase the width of the k-space trajectory. For an undersampled scan, the additional width reduces gaps in k-space and improves reconstructions, but most other properties of the underlying linear method are unchanged. For example, Cartesian-FRONSAC retains features like insensitivity to off-resonance spins and timing delays, ease of changing FOV, resolution, and orientation, and relatively simple contrast behavior, while still allowing for higher undersampling factors. This versatile approach can be added to nearly any sequence, improving undersampling artifacts even for low channel arrays, as we have shown by acquiring a full FRONSAC-enhanced brain protocol in a cohort of healthy subjects.

An additional emerging application of nonlinear gradients is in generating diffusion contrast. In some sense, a linear gradient is the maximally egalitarian way to distribute a ΔB(x): it generates the same Gx (d(ΔB)/dx) everywhere, but the peak Gx across the FOV is the lowest possible. By allowing nonlinearity, Gx is different at each voxel, but it can be concentrated to certain regions of interest. Thus, for specialized applications, it may be possible to achieve massive gradients strengths and very high diffusion weightings using simple equipment. For example, for prostate DWI, we propose an inside-out nonlinear gradient, which simulations suggest will ultimately double CNR in ADC maps.

Abstract

Wouldn’t it be great to perform a surgery without incision or bleeding? “Histotripsy” is the first non-invasive, non-ionizing, and non-thermal ablation technique that is invented by Dr. Xu and her colleagues at the University of Michigan. Using ultrasound pulses applied from outside the body and focused to the target diseased tissue, histotripsy produces a cluster of energetic microbubbles at the target tissue using the endogenous gas pockets with millimeter accuracy. These microbubbles, each similar in size to individual cells, function as “mini-scalpels” to mechanically fractionate cells to acellular debris in the target tissue. The acellular debris is absorbed over time via metabolism, resulting in effective tissue removal. Off-target tissue remains undamaged and no incision is needed. Thus histotripsy can perform non-invasive surgery guided by real-time imaging. Histotripsy has potential for many clinical applications where non-invasive tissue removal is desired. Recent research in Dr. Xu’s lab also shows potent immune response and abscopal effects induced by histotripsy and its potential for immunotherapy. Dr. Xu will talk about the mechanism and instrumentation development of histotripsy as well as the latest pre-clinical and clinical studies of histotripsy for cancer, neurological, cardiovascular, and immunotherapy applications.

Abstract

Zhen Xu is a tenured Associate Professor and Associate Chair of Graduate Education at the Department of Biomedical Engineering at the University of Michigan, Ann Arbor, MI. She received the Ph.D. degree in biomedical engineering from the University of Michigan in 2005. Her research focuses on ultrasound therapy and imaging, particularly histotripsy. She received the IEEE Ultrasonics, Ferroelectrics, and Frequency Control (UFFC) Outstanding Paper Award in 2006; National Institute of Health (NIH) New Investigator Award at the First National Institute of Biomedical Imaging and Bioengineering (NIBIB) Edward C. Nagy New Investigator Symposium in 2011, The Federic Lizzi Early Career Award from The International Society of Therapeutic Ultrasound (ISTU) in 2015, the Fellow of American Institute of Medicine and Bioengineering in 2019, and The Lockhart Memorial Prize for Cancer Research in 2020. She is an associate editor for IEEE Transactions on UFFC and Frontiers in Bioengineering and Biotechnology, Deputy VP of UFFC Ultrasonics Standing Committee, and an elected board member of ISTU. She is a principal investigator of grants funded by NIH, Office of Navy Research, American Cancer Association, and Focused Ultrasound Foundation. She is also co-founder of HistoSonics, a startup company developing histotripsy for oncological applications.

Abstract

Vision researchers have been analyzing behaviors of radiologists during screening to understand how and why they miss tumors or misdiagnose. In this regard, eye-trackers have been instrumental in understanding visual search processes of radiologists. However, most relevant studies in this aspect are not compatible with realistic radiology reading rooms. In this talk, I will share our unique experience for developing a paradigm shifting computer aided diagnosis (CAD) system, called collaborative CAD (C-CAD), that unifies CAD and eye-tracking systems in realistic radiology room settings. In other words, we are creating artificial intelligence (AI) tools that get benefits from human cognition and improve over complementary powers of AI and human intelligence. We first developed an eye-tracking interface providing radiologists with a real radiology reading room experience. Second, we proposed a novel computer algorithm that unifies eye-tracking data and a CAD system. The proposed C-CAD collaborates with radiologists via eye-tracking technology and helps them to improve their diagnostic decisions. The proposed C-CAD system has been tested in a lung and prostate cancer screening experiment with multiple radiologists. More recently, we also experimented brain tumor segmentation with the proposed technology leading to promising results. In the last part of my talk, I will describe how to develop AI algorithms which are trusted by clinicians, namely “explainable AI algorithms”. By embedding explainability into black box nature of deep learning algorithms, it will be possible to deploy AI tools into clinical workflow, and leading into more intelligent and less artificial algorithms available in radiology rooms.

Abstract

This talk will provide an overview of work that our group has done on mapping connectional anatomy from diffusion MRI, and a preview of where this path might lead us next. First, I will discuss our previously developed algorithms for reconstructing white-matter pathways from diffusion MRI. These include both supervised and unsupervised methods with a common theme: like neuroanatomists, they define white-matter bundles based on relative position with respect to neighboring anatomical structures, rather than based on absolute coordinates in a template space. This makes them robust to individual variability and to the effects of disease or healthy development and aging.

Second, I will present results from recent post mortem validation studies, where we have evaluated the accuracy of diffusion MRI with respect to polarization-sensitive optical coherence tomography in human samples, or chemical tracing in non-human primates. Our results suggest that existing methods for inferring the orientation of axon bundles from diffusion MRI do not benefit substantially from very high b-values. This implies that our analysis tools have not kept up with the rapid progress of our hardware, and that new tools are needed to fully take advantage of the data that can be acquired by today’s ultra-high-gradient MRI scanners. I will end the talk by discussing how we may be able to address this, by using the post mortem data not only to evaluate existing methods but to engineer the next generation of tractography algorithms.

Abstract

Myelin abnormalities in schizophrenia spectrum disorders have been suggested by histological studies, which have shown aberrations in myelin lamellae, oligodendrocyte structure, and myelin- and oligodendrocyte-related gene expression. However, in vivo examination of myelin content, especially the intra-cortical myeloarchitecture remains limited. In our current project, we employ magnetization transfer imaging to derive macromolecular proton fraction (MPF), a quantitative estimate of myelin content. This talk will focus on data suggesting a flattening of the cortical myelin profile in patients with schizophrenia spectrum disorders and an association of cortical myelin alterations with illness progression and cognitive outcomes. Preliminary findings on whole-brain myeloarchitectural similarity changes in schizophrenia will also be presented.

Speaker Bio

Yu Veronica Sui is a second-year graduate student in Sackler Institute’s Biomedical Imaging and Technology training program working with Mariana Lazar. She has a background in cognitive psychology and is interested in developing and employing new imaging and analytics methods to characterize the neural bases of psychiatric disorders. Her focus in Lazar Lab is psychosis-related pathological changes in the brain, including both microstructural and connectivity abnormalities.

Abstract

An alluring feature of magnetic resonance imaging (MRI) is its potential to provide quantitative and specific characterizations of tissue. However, the barriers to the realization of quantitative MRI (qMRI) are many and progress has been slow. This presentation will include vignettes of technical, experimental, and translational efforts to develop and utilize qMRI, with primary applications being the characterization of white matter micro-structure/composition and bone fracture risk.

No abstract was provided for this talk.

Abstract

Lower field strength MRI systems paired with high-performance hardware and advanced imaging methods offer unique opportunities for clinical imaging. Specifically, this system configuration offers improved safety for MRI-guided invasive procedures, improved imaging in high-susceptibility regions including the lung, and advantages for efficient image acquisitions. In light of developments in MRI engineering and available computational power, and as well as the drive to reduce MRI costs, there is significant value in revisiting lower field MRI in the context of modern clinical imaging. This talk will describe the experience of the NHLBI imaging patients on a ramped down 0.55T system for 2 years.

Speaker Bio

Dr. Adrienne Campbell-Washburn is the Director of the MRI Technology Program at the National Heart, Lung, and Blood Institute (National Institutes of Health). Her research focuses on the development of MRI technology for cardiac imaging, lung imaging, and MRI-guided interventions. She works on developing advanced MRI acquisitions that leverage non-Cartesian sampling and reconstruction methods using state-of-the-art computational resources in the clinical environment. Her research aims to improve SNR-efficiency, imaging speed, interventional procedural guidance including device safety and visibility, motion robustness, quantification, and clinical integration.

Abstract

Imaging biomarkers that bridge neuronal abnormalities in vivo and behavior, and animal models and human patients, are urgently needed to quicken discovery and application of novel disease-modifying therapy, but are not yet available. I will be discussing novel MRI and OCT approaches for measuring sustained and excessive production of free radicals (i.e., oxidative stress) in neuronal laminae without a contrast agent in untreatable neurodegenerative disease. These studies set the stage for translating and managing anti-oxidant treatment in patients for the first time.

Abstract

Existing clinical infrastructures severely under-utilize modern computation resources, leading to costly errors, slow workflows and limited research opportunities. However, trends in cloud computing and machine learning are rapidly changing this landscape. Tech companies, such as Amazon, Google and Microsoft, are now racing to integrate high performance computing into clinical settings. Medical imaging stands to gain tremendously from advances in computing power, which will enable many previously unthinkable applications.

In this talk, I will focus on three directions on leveraging these emerging computing resources to improve medical imaging: 1) reconstructions of high dimensional volumetric dynamic MRI on the order of 100GBs; 2) continuous learning and image quality improvement from undersampled datasets; 3) optimizing end-to-end systems across clinical workflow.

Abstract

Dynamic MR image quality is limited by the temporal and spatial resolution trade-off. Adopting machine learning in the reconstruction network provide alternative method to reconstruct the image of better quality. This presentation will focus on our work using Recurrent Neural Networks(RNN) as regularizer in our dynamic MR reconstruction network. The regularizer is designed to take time series of kspace with flexible length and use it to reconstruct image series. I will show how we design simulation to get ground truth images for model training. Then I will show the output of the trained network on simulated breast perfusion MR data with comparison to other reconstruction methods.

Spaker Bio

Zhengnan Huang joined Florian Knoll’s lab in 2018. He has educational background in bioinformatics. His research interest is MR image reconstruction and machine learning application.

Abstract

In this talk, we discuss three deep learning techniques to improve the image quality of 3D diffusion MRI Images. We first introduce a novel low-memory method, which allows us to control the GPU memory usage during training therefore allowing us to handle the processing of 3-dimensional, high-resolution, multi-channeled medical images. Secondly we present the first multi-task learning approach in data harmonization, where we integrate information from multiple acquisitions to improve the predictive performance and learning efficiency of the training procedure. Thirdly we present an extension of the transposed convolution, where we learn both the offsets of target locations and a blur to interpolate the fractional positions. All three techniques can be applied in other image-related paradigms.

Abstract

Dysregulated immune signaling contributes to many neuropsychiatric conditions. Brain PET imaging can measure neuroimmune factors that inform treatment development for such conditions. This talk will focus on PET imaging of the 18-kDa translocator protein (TSPO). Preclinical work informing the interpretation of TSPO signal, including imaging dynamic responses to endotoxin, an acute immune stimulus, will first be presented. Next, human data imaging TSPO in tobacco smoking, alcohol use disorder, and Alzheimer’s disease will be presented to demonstrate diverse applications of these techniques. Whole body imaging of TSPO following acute alcohol administration as an immune stimulus will also be presented. Finally, work characterizing new radiotracers that complement TSPO measures in immune signaling will be presented. This work depicts ways in which PET imaging can be leveraged to study immune function in the context of neuropsychiatric disorders.

Abstract

A novel investigation of rheo-markers (proton T2* and sodium multiple quantum filtering) shows the potential for multi-nuclear MRI biomarkers in mechanically loaded joints with good evidence of a dynamic 23Na environment during compression which may be useful for early OA detection before symptoms occur.

Abstract

A majority of primary total hip arthroplasty (THA) devices function well but implant failures occur. This presentation will cover our long standing efforts to utilize MRI in identifying patients needing premature implant revision due to adverse local tissue reactions (ALTRs). The utility of advanced multi-spectral imaging to reduce metallic susceptibility artifact and visualize synovitis, osteolysis, and tendon tears near arthroplasty will be displayed. I will also show results from our on-going studies using MRI to evaluate patients with different THA bearing materials to determine which factors are predictive of abnormal synovial reaction. Finally, data will be shown regarding the longitudinal prevalence of MRI detected ALTRs in a cohort of high functioning THA patients.

Abstract

Alzheimer’s disease (AD) is a progressive neurodegenerative disorder characterized by the neuropathological accumulation of amyloid beta (Ab) plaques and neurofibrillary tangles comprised of hyperphosphorylated tau. Tau is a microtubule-associated protein involved in microtubule stability and when tau is hyperphosphorylated, microtubules become destabilized which leads to impaired axonal transport. Axonal transport is an important cellular process that shuttles vesicles, neurotransmitters, and mitochondria from the soma to the synapse. Perturbations in axonal transport disrupt neuronal activity by reducing the transport of mitochondria, increasing reactive oxygen species (ROS) and diminishing the formation of active zones at the synapse. Axonal transport deficits are thought to occur early and continue to progress in AD. Thus, there is a significant need and strong scientific premise to identify the mechanisms by which axonal transport deficits occur and also can be improved in AD.

Olfactory receptor neurons are the only part of the central nervous system (CNS) with direct access to the outside world. They lie at the beginning of a neural network which projects to the olfactory bulb followed by the piriform olfactory cortex (primary olfactory cortex), the entorhinal cortex (secondary olfactory cortex). The olfactory system is also the first system affected in AD patients and mouse models of AD before cognitive deficits develop. Indeed, using Manganese Enhanced MRI (MEMRI), we have shown that axonal transport deficits in the olfactory receptor neurons occur before the appearance of learning and memory deficits appear and are reversed when we reduce ROS levels by overexpressing superoxide dismutase 2 (SOD-2) in AD mice. Here, we describe our current efforts with reducing oxidative stress in the olfactory structures in mouse models of AD with intranasally applied nanoantioxidants.

Abstract

Functional MRI is used extensively in human brain research, enabling characterization of distributed brain activity underlying complex perceptual and cognitive processes. However, it has been limited in utility in rodents. I will present the work we have done to establish awake mouse MRI, characterize the properties of the hemodynamic response function as different from humans and how these two aspect enabled us to conduct whole-brain fMRI of the behaving animal. I will expand on recent work using whole-brain functional imaging of head-fixed mice performing odor discrimination and conclude by showing additional behavioral modalities we develop with the goal to establish this approach as a platform to be used extensively in the field.

Abstract

A novel imaging technique is presented, capable of simultaneously quantifying time-resolved blood flow velocities and the relaxation constants of static tissue. This is accomplished through the use of a Magnetic Resonance Fingerprinting (MRF) based approach. The developed technique, termed “Flow-MRF”, allows accurate mapping of velocities and relaxation constants in measurement times up to 4-fold shorter than conventional MRI-based velocimetry techniques.

Abstract

Chemical exchange saturation transfer (CEST) is a relatively new type of MRI contrast that indirectly detects low concentration labile protons through water signal with enhanced sensitivity. In this presentation, I will explain the principles of CEST imaging and its applications. I will show results from using CEST to image D-glucose (glucoCEST) in vivo, first on brain tumor mouse model at ultra-high magnetic field, then on human brain tumor patient on 7T system. Our recent effort of translating the technique to clinical field strength and the promise and challenges of glucoCEST at clinical field strength will be also be discussed.

Abstract

CSF, PET and MRI multimodal studies enable the early diagnosis of Alzheimer’s Disease. We have proposed a model in which interactions between biomarkers in preclinical AD result in a two-phase phenomenon: an initial phase of cortical thickening due to amyloid-related inflammation, followed by a cortical atrophy phase which occurs once tau biomarkers become abnormal. These results have implications in the selection of patients for clinical trials and the use of MRI as a surrogate marker of efficacy. We will also present data showing the potential of studying the cortical microstructure with DTI to assess these early changes and in the diagnosis of other neurodegenerative diseases.

Abstract

In this talk, I will introduce the combination of the advanced fMRI method with the emerging neuro-techniques to decipher the neuro-glial-vascular (NGV) coupling basis of brain state dynamics. First, we will see through the large voxel acquired from conventional fMRI to decipher the contribution from distinct vascular components to the fMRI signal. A newly developed single-vessel fMRI method allows identifying the activity-evoked hemodynamic signal propagation through the cerebrovasculature in the brain with either sensory inputs or optogenetic activation. Second, we will combine the fMRI with the optical fiber-mediated calcium recordings to decipher the cell-type-specific contribution to the fMRI signal from neurons and astrocytes. Meanwhile, we will also show how extracellular glutamate can be recorded simultaneously to mediate NGV interaction. Finally, we are going to present how the global fMRI signal fluctuation can be linked to the brain state changes. We merge the pupillometry with the multi-modal fMRI to examine the detailed arousal index by pupil dynamics and fMRI fluctuation. In summary, we hope to provide a novel perspective to under brain function with multi-modal fMRI across different scales.

Abstract

Quantitative MRI (qMRI) parameters such as T1 provide physical parametric measurements crucial for clinical and scientific studies. However, an important challenge in applying qMRI measurements is their biological specificity, as they change in response to both molecular composition and water content. I will discuss an approach that disentangles these two important biological quantities and allows for decoding of the molecular composition from the qMRI signal. I will demonstrate that this approach can reveal the molecular composition of lipid samples. Furthermore, we identify region-specific molecular signatures in the human brain that have been validated against histological measurements. Last, we exploit our method to reveal region-specific molecular changes in the aging human brain. I suggest that the ability to disentangle molecular signatures from water-related changes opens the door to a quantitative and specific characterization of the human brain.

Abstract

This talk will start with a brief introduction of what is negative BOLD response in fMRI data and what are its characteristics. It continues by categorizing different types of negative BOLD signal according to their properties and outlines the optimal techniques used to extract negative BOLD response.

The applications of negative BOLD response are unlimited, however, three on-going research projects in our lab which extensively rely on negative BOLD response will be presented. First project uses negative BOLD response to demonstrate how spontaneous activity and task-evoked activity in the brain give rise to two spatially overlapping but temporally dissociable signals which both are manifested in fMRI data. Using these findings, the second project attempts to use negative BOLD response to demonstrate evidences for the hierarchical structure in the human brain functional networks. This is done by demonstrating that task-evoked negative BOLD response in the Default mode network is modulated by switching attention whereas the functional connectivity between the same network of regions remain intact. Finally, we introduce negative BOLD response as a new brain biomarker that could potentially differentiate between normal and pathological ageing brains.

Abstract

Several multiple fiber methods have been proposed that seem to overcome the limitations of the diffusion tensor and methodologies aimed to provide information from the diffusion signal, but that are mostly suited for single fiber population regions. Although the majority of these multiple fiber methods where created with the primary purpose of improving tractography results, some of them are able to provide per-bundle dMRI derived metrics. However, biological interpretations of such metrics are limited by the lack of histological confirmation.

To this end, we developed a straightforward biological validation framework. Unilateral retinal ischemia was induced in ten rats, which resulted in axonal (Wallerian) degeneration of the corresponding optic nerve, while the contralateral was left intact; the intact and injured axonal populations meet at the optic chiasm as they cross the midline, generating a fiber crossing region in which each population has different diffusion properties. Five rats served as controls. High-resolution ex vivo dMRI was acquired five weeks after experimental procedures.

We correlated and compared histology derived information to per-bundle descriptors obtained from three multiple fiber methodologies for dMRI analysis: constrained spherical deconvolution (CSD) and two multi-tensor (MT) representations. We found a tight correlation between axonal density (as evaluated through automatic segmentation of histological sections) with per-bundle apparent fiber density (from CSD) and fractional anisotropy (derived from the MT methods). The multiple fiber methods explored were able to correctly identify the damaged fiber populations in a region of fiber crossings (chiasm). Our results provide validation of metrics that bring substantial and clinically useful information about white-matter tissue at crossing fiber regions.

Our proposed framework is useful to validate other current and future dMRI multiple fiber methods; it also can be extended for the analysis of other pathological conditions, such as inflammation and demyelination, in order to evaluate the capabilities of these dMRI methods to differentiate between.

Abstract

Elucidating the relationship between the structure and function of the brain is one of the main open questions in neuroscience. The capabilities of diffusion MRI-based (dMRI) techniques to quantify connectivity strength between brain areas, referred to as structural connectivity, in combination with modalities to quantify brain function such as electrocorticography (ECoG) have enabled advances in this field.

The aim of the project that I will talk about is to establish a relationship between dMRI-based structural connectivity and effective connectivity maps based on the propagation of Cortico-Cortical Evoked Potentials (CCEPs). To this end, we applied direct electrical stimulation of the cortex during awake surgery of brain tumor patients and recorded the induced electrophysiological activity with subdural ECoG electrodes.

I will briefly summarize our study of seven patients. For each of them, we correlated dMRI-based structural connectivity measures, including streamline counts and lengths, with delays and amplitudes of CCEPs. In addition, we used the structural information to predict the CCEP propagation with a linear regression model.

Abstract

Arterial spin labeling (ASL) is a non-invasive MRI technique for cerebral blood flow (CBF) measurement by using magnetically labeled blood spins as endogenous tracers. The recent development of ASL has promoted it as a useful imaging tool for tissue perfusion assessment in cerebrovascular disorders. For perfusion imaging, after spin tagging, images are generally acquired at a relatively long post-labeling delay time (~1.8s) when the labeled blood from labeling plane reaches capillaries/tissue. Additional physiological information can be derived during the passage of labeled blood through the cerebral arterial trees into capillaries and tissue, such as dynamic MR angiography, vascular territorial mapping, cerebral blood volume (CBV) and vascular compliance et al, all of which also provide useful information in the diagnosis and treatment of cerebrovascular disease. In this talk, I will introduce my work about these recent advances in ASL beyond CBF measurement.

Abstract

Inge Brinkmann will provide an overview of her work at Siemens Healthineers.

Abstract

In recent years, deep convolutional neural networks have overcome several challenges in the field of computer vision and image processing. In particular, pixel-level tasks such as image segmentation, restoration, generation, enhancement, and inpainting, showed significant improvements thanks to advances in the technique. In general, the supervised training of a neural network entails solving a high dimensional non-convex optimization problem whose objective is to transform the vectors of the input domain to a prescribed output. However, due to the high dimensionality of the parameter space and the presence of saddle points and large flat regions on the error surface, the process of training a neural network is extraordinarily challenging. We propose modeling new loss functions to facilitate training while improving the generalization of models for pixel-level regression and classification tasks. Our newly introduced loss functions modify the optimization landscape to achieve better results in regions which are notoriously more prone to failure. They increase the overall optimization performance and accelerate convergence. We applied our formulations to instance segmentation of cells with full and weak supervision and tested them on challenging biological images with isolated and cluttered cells. We also propose a new pixel-level regression loss function applied to the multi-focus image fusion problem resulting in the joint learning of activity level measurement and fusion rule. New pre-processing and post-processing techniques to help improve the solutions are also introduced. Our methods have shown significant improvements in the segmentation and image restoration tasks as reported by a diverse set of metrics and visual inspections.

Abstract

Recent achievement of deep learning algorithms using convolutional neural networks (CNNs) yields high performance of image classification and segmentation. The algorithms have been applied to assist doctor’s medical decision more efficiently and effectively. In this talk, I will introduce deep learning applications to rotator cuff tears, glaucoma, and intraocular pressure relations with daily diet pattern.

Abstract

Osteoarthritis (OA) is the most common form of arthritis, affecting millions of people in the US for which only palliative treatments are available until joint replacement surgery. The elusiveness of effective OA treatments is the consequence of OA being a complex disease. OA is a multifactorial disease with inflammatory, metabolic, and mechanical causes involving all tissues of the joint. Thus, we still lack understanding on OA pathogenesis, in part due to the lack of diagnostic biomarkers that can detect early pathological changes in the joint and monitor therapy. A major barrier in OA research is to see and understand the interplay between OA factors to both be able to phenotype OA and provide patient-specific treatments. At OLE! (Osteoarthritis Lab for Experimental !maging), we aim to solve this technological problem by developing advanced imaging technology that can monitor in vivo of the influence of OA factors and treat them. We have established an innovative research program for in vivo molecular imaging of the degenerative joint. We are developing imaging probes with theragnostic potential that combine the specificity of biochemical assays with anatomical and tissue-specific assessment of early degenerative changes.

Abstract

Conventional T1- and T2- weighted pulse sequences are routinely used in the clinic for the diagnosis of a variety of pathologies. Quantatative estimation of tissue relaxation times can be used to further improve the quality of diagnosis in applications including cardiac, abdominal, and musculoskeletal imaging. In this talk, I will introduce a radial Turbo Spin Echo (RADTSE) pulse sequence for simultaneous T2w imaging and T2 mapping. Specifically, I will present a RADTSE pulse sequence with very long echo train lengths and variable refocusing flip angles for improved slice coverage in abdominal breath-held imaging. I will also discuss a simultaneous multi-slice excitation technique to improve the slice and SNR efficiency of double inversion RADTSE for cardiac imaging. Finally, I will give an overview of my ongoing research on quantitative T1 mapping and the use of artificial intelligence for analysis of deep brain structures.

Abstract

Magnetic Resonance Spectroscopy (MRS) is used to non-invasively monitor the in-vivo biochemistry of tissue, by quantifying the concentrations of several prominent metabolites, including glutamate, choline, GABA and creatine, among others. Conventional MRS produces static estimates of concentrations. In this talk, I will present two recent advances in MRS methodology which provide a more dynamic information. First, I will discuss our work on multiparametric MRS, which simultaneously quantifies metabolite concentrations and relaxation times (T1, T2). Both T1 and T2 provide information about the molecular microenvironment of the metabolites via their microscopic dynamics. In the second part of the talk, I will discuss our work on functional MRS, which examines the temporal changes to several prominent metabolites in response to external stimuli, and discuss some of our interpretations to the changes measured in this unsolved, fascinating puzzle.

Abstract

In this talk, I will introduce my recent research activities from image formation to post processing using examples of whole body scatter estimation and image reconstruction for Biograph mMR and a deep learning powered lung analysis post processing application. For the Biograph mMR, we designed a new method to process step and shoot sinogram to simulate a whole body sinogram and reconstruct the whole body image directly, which increases the quantitative accuracy of scatter estimation and improves performance of image reconstruction. For post-processing, I will showcase several AI predevelopment activities, focusing on the lung ventilation / perfusion application. Here, deep learning-based lung lobe segmentation has been developed to enable a potentially fully automated workflow for lung analysis. This prototype is available on the Siemens Frontier platform, offering a seamless integration to syngo.

Abstract

fastMRI is a collaborative research project between Facebook AI Research (FAIR) and NYU Langone Health. The aim is to investigate the use of AI to improve acceleration and robustness of MRI scans. In this talk, Tullie, a Research Engineer at FAIR, will give an overview of the work done on knee image reconstruction and reinforcement learning based active sampling. He will cover the plans going forward to investigate brain image reconstruction, motion robust reconstructions for Dynamic MRI and extensions to the active sampling work.

Abstract

Expecting to fully engage equally deep Physicists and Biologists, I will introduce the notion of “how good a macromolecular CyoEM map is”, addressing this question in a totally new way in the field, by providing a “resolution tensor” per CryoEM voxel map (instead of just a number, the so-called “local resolution”). The mathematical beauty of this tensor representation will immediately open a new university of opportunities for experimentalists in CryoEM (clearly impacting Pharma), with the capability to assess the quality of the map from the map itself (without the images), the alignement errors, the presence of problematic directions….. and much more.

Abstract

T1-weighted structural imaging with MP-RAGE is a cornerstone of brain imaging studies for both clinical and research applications. However it is sensitive to head motion, RF inhomogeneities, and provides only a single image contrast. Recently, we developed MPnRAGE which combines inversion magnetization preparation with a 3D radial rapid gradient echo readout. This sampling enables the simultaneous acquisition of n inversion recovery contrasts, which may be used to generate one or more application specific contrast images, and generate high resolution, whole-brain T1 relaxometry images. The 3D radial sampling is also highly amenable to self-navigated motion correction during the reconstruction, which provides robust and reliable high quality T1-weighted and quantitative T1 images of the brain. This technique is highly promising for brain imaging studies of children, aging and brain pathology.

Abstract

A synergistic approach in developing MRI acquisition through utilizing the interplay between hardware design, software algorithm development, and MR physics has dramatically increased MRI’s spatiotemporal resolution capability. IN this talk, I will cover some of these tailored acquisition strategies which are being pioneered by my group, focusing particularly on applications in rapid imaging, diffusion, & fMRI, and quantitative and multidimensional/time-resolved imaging of the brain. The overarching theme is in radically improving the speed, sensitivity, and specificity of in vivo brain imaging, with the goal in providing more detailed information about the brain both in health and disease.

Abstract

Dr. Ren will discuss a series of studies using dynamic and kinetic MRS, that have identified cellular energetic activities in multiple pathways. He will also demonstrate how 7T 31P MRS can serve as a powerful tool to capture aberrant brain events in remote skeletal muscle.

Abstract

Migraine is a disabling, multifactorial recurrent neurological disorder. Affecting approximately 38 million people in the United States alone, migraine is recognized by the World Health Organization as the 7th most disabling condition, due to the sufferer’s inability to perform everyday activities. The characterization, classification and diagnosis of migraine is complex due to the tremendous cohort of variable clinical triggers and symptoms reported. Collectively, the symptoms accompanying migraine implicate multiple neural networks and processes functioning abnormally. A mechanistic search for a common denominator based on the symptoms in migraine potentially involves the recruitment of the thalamic region (fatigue, depression, irritability, food cravings), brainstem (muscle tenderness, neck stiffness), cortex (sensitivity to photo and phono) and limbic response (depression anhedonia).

The prevailing consensus in the migraine community appears to indicate a combination of neuronal and vascular involvement with the trigeminal vascular system (TGVS) complicit in the progression of migraine. Broadly, various triggers initiate migraine to differing degrees and treatment methodologies target a variety of pathways with varied results; the fundamental mechanism driving change is unclear. In the absence of an identifiable locus for anatomical, biochemical or pathological change in common clinical migraine, a fundamental question remains unanswered: What endogenous media and pathways link the stimulus to perception of migraine and potentially pain?

The goal of this talk is to highlight progress made in the characterization of acute triggered migraine. To elucidate this neurovascular coupled system, two fundamental mechanisms complicit in neuronal disorders are explored, namely ionic fluxes using sodium MRI and metabolic changes by utilizing proton spectroscopy as well as ongoing efforts to characterize cerebral perfusion—with and without pharmaceutical prophylaxis.

Abstract

Automatic delineation and measurement of main organs is one of the critical steps for assessment of disease, planning and postoperative or treatment follow-up. Internal human anatomy is composed of complex shapes that exhibit a large degree of variation, which is challenging to capture using existing modeling tools. We observe that complex shapes can be learned by neural networks from large amounts of examples and summarized using a coarsely defined structure, which is consistent and robust across variety of observations. Further, shape structure can be used in the synthesis process to improve the quality of generated shapes. We study medical applications of 3D organ reconstruction from topograms and synthetic X-ray prediction and propose several ways of incorporating structure into the synthesis process, and. We also show compelling quantitative results on 3D liver shape reconstruction and volume estimation on 2129 CT scans.

Abstract

Magnetic Resonance Imaging (MRI) has emerged as one of the most powerful and informative diagnostic tools in modern medicine. While most clinical MR studies use magnetic field strengths of 1.5T or 3T, leading research is pushing these magnetic field strengths to 7T and beyond. These new ultra high‐field (UHF) technologies promise images with higher spatial resolution, higher sensitivity to subtle change, and novel contrasts, which will in turn improve our basic understanding of anatomy and physiology in both healthy tissue and disease. However, there are substantial hurdles to surmount before we will reap the promised benefits of UHF MRI in clinical applications. This talk will introduce some of the major challenges faced in UHF MRI and will summarize a number of concepts in engineering and multiphysics that are being researched to overcome these issues.

Abstract

Chronic Kidney Disease (CKD) is a cardinal feature of methylmalonic acidemia (MMA), a prototypic organic acidemia. Impaired growth, low activity, and protein restriction affect muscle mass and lower serum creatinine concentrations, which can delay the diagnosis and management of renal disease in this patient population. We have designed a general alternative strategy for monitoring renal function based on administration of a pH sensitive MRI contrast agents to acquire functional information. We have tested our methods in a mouse model of MMA, and detected robust differences in the perfusion fraction and pH maps we produce between groups with severe, mild, and no renal disease. Our results demonstrate that MRI contrast agents can be used for early detection and monitoring of CKD, particularly in disorders that alter renal pH and perfusion such as MMA.

Abstract

One grand challenge for the 21st century is to achieve an integrated understanding of brain circuits and networks, particularly the spatiotemporal patterns of neural activity that give rise to functions and behavior. Brains form highly complex circuits where circuit elements communicate using electrical and/or chemical signals. Such communications are typically facilitated through long-range projections that interconnect numerous regions, giving rise to a network-like property in the brain. Despite their importance, the functions of long-range projections remain poorly understood. Here, I will show you our recent developments in deploying multimodal techniques in-vivo on rodents to interrogate multisensory brain networks; leveraging on the strengths of optogenetics to enable cell-type specific neuromodulation, functional MRI (fMRI) to visualize brain-wide neural activity, and electrophysiology to explore the neural mechanism(s) that underlie our observations. I will present key findings from our work in the multisensory thalamo-cortical, cortico-cortical, and cortical-subcortical circuits, including the unique dynamic spatiotemporal response properties of multisensory pathways as well as their functional relevance. From this talk, I aim to show you how utilization of multimodal brain imaging techniques can be vital in our quest to achieving an integrated and systemic understanding of large-scale brain-wide multisensory interactions.

Abstract

In this talk, I will present techniques to reconstruct 3D dynamic MRI of ~100 GBs from non-gated acquisitions. The problem considered is vastly undetermined and demanding of computation and memory. I will introduce a multi scale low rank matrix model to compactly represent dynamic image sequence. This enables compressed storage, which in combination with a stochastic optimization approach, renders the reconstruction of 100s of GBs of images feasible. The proposed method is applied to dynamic contrast enhanced MRI and free breathing lung MRI, with reconstruction resolution of near millimeter spatially, and sub-second temporally. The attached animated gif shows a 3D rendered result from this talk. (Joint work with Xucheng Zhu, Joseph Cheng, Peder Larson, Shreyas Vasanawala, and Michael Lustig)

Abstract

Pain is now the #1 clinical problem in the world and, yet, our current imaging methods to correctly identify pain generators remain woefully innacurate. The fact that meniscal tears, herniated discs, arthritis and rotator cuff tears are seen in asymptomatic individuals supports the disturbing fact that standard-of-care imaging techniques are extremely poor at pinpointing the exact site of pain generation. This dearth of unreliable diagnostic tools necessarily facilitates significant misdiagnosis, mismanagement, rampant use of opioids and unhelpful surgeries. Thankfully, relatively recent developments in clinical molecular imaging (MI) are affording the opportunity to pinpoint the exact site(s) of pain generation due to advances in biomarker discovery, imaging technology and radiotracer design. Our group has developed a highly specific 18F-labeled positron emission tomography (PET) radiotracer for imaging the sigma-1 receptor (S1R), a master regulator of ion channel activity and molecular biomarker of pain generation. Additionally, we have repurposed 18F-fluordeoxyglucose (FDG) as a marker of inflammation by virtue of its proclivity for metabolically active processes. Here, we will describe our experience using these radiotracers in our ongoing PET/MRI clinical trials of patients with chronic pain. Importantly, we will illustrate how this new imaging method is enabling more accurate identification and localization of pain generators and is starting to positively impact the way we treat pain.

Abstract

Neuroinflammation is a key pathological feature of many central nervous system (CNS) diseases. Although extensive work in preclinical rodent models demonstrate a significant role for both the innate and adaptive immune response in the initiation and progression of neurological diseases, our understanding of these responses and their contribution to human disease remains very limited. ​Additionally, ​both beneficial and toxic inflammatory processes are associated with progression and remission of neurological disease, and the spatiotemporal course of these complex responses remain a mystery especially in the clinical setting. Molecular imaging using positron emission tomography (PET) has enormous potential as a translatable technique to enhance our understanding of neuroinflammation in CNS diseases. Our experience with developing new PET radioligands for visualizing the neuroinflammatory component of Alzheimer’s disease, multiple sclerosis, and stroke will be described. I will provide examples regarding our work on d​esigning radioligands for the translator protein 18 kDa (TSPO), triggering receptor expressed on myeloid cells 1 (TREM1), and two B lymphocyte surface antigens. Specifically, the in vivo role, spatiotemporal dynamics, ​peripheral contribution and different functional phenotypes of innate and adaptive immune cells throughout the progression of CNS diseases will be shown. Moreover, I will describe how we are starting to apply these tools to track disease progression, guide therapeutic selection for individual patients, and serve as surrogate endpoints in clinical trials.

Abstract

Although both MRI and CT resolutions are limited, different MRI and X-ray modalities offer possibilities for tissue microstructure analyses. Diffusion MRI is sensitive to proton displacement in the micrometer scale, whereas X-ray photons scatter off the sample’s micro- and nano-structure.

Recently, we developed techniques based on X-ray scattering that allow tomographic investigations of the sample’s fiber orientations. In brain, these techniques also allow quantifying myelin content, due to myelin’s repetitive structure.

In this talk I will give an overview of my work in CBI in the past two years; I will present applications of these techniques to mouse and human CNS, to derive fiber orientations and myelin content in healthy, diseased and treated tissue, and comparison to diffusion MRI metrics.

Abstract

Diffusion MR imaging has become an important clinical imaging modality in breast imaging, for the detection of malignant lesions and metastases, as well as for therapy monitoring. Some studies have shown that pretreatment ADC has might be a useful biomarker to predict response to breast cancer therapy. However, non-Gaussian diffusion might potentially extract more microstructural information than the ADC, as with a high degree diffusion weighting (high b values) one increases the effects of obstacles to free diffusion present in tissues, notably cell membranes. Indeed, the “kurtosis” which reflects diffusion non-gaussianity is high in malignant lesions compared to benign lesions. Still, a particularly challenging problem for breast diffusion MRI is the detection of the non-mass enhancing lesions seen on contrast-enhanced MRI, such as with DCIS. High-resolution images using readout- segmented EPI might overcome the low sensitivity of such lesions. On the other hand, tissue perfusion which is also available from diffusion MRI images (IVIM effect) gives information on the blood fraction which appears correlated with vessel density. The IVIM fraction is usually high in malignant lesions, but there seems to be a large overlap with benign lesions. Combination of non-Gaussian diffusion and IVIM parameters appears to boost diagnosis accuracy. Still, the results have been sometimes inconsistent in the literature partly due to differences in study design (choice of b values and acquisition methods, data analysis approaches, differences in patient population), and the standardization of acquisition protocols and processing methods used for quantitative DWI analysis is a very important step for for diffusion MR imaging to become a clinically recognized biomarker.

The investigations on the relationship between the IVIM/diffusion parameters and the underlying tissue structure at microscopic level, as well as changes induced by therapy, must be pursued using animal models, MRI of specimens at ultra-high resolution and validation with histology. Reliability and reproducibility of diffusion MRI results must also be assessed to facilitate monitoring disease progression or response to therapy in individual patients.

Abstract

The understanding of the human brain is one of the main scientific challenges of the 21st century. Unraveling the biological mechanisms of our mental life should help us understanding neurological or psychiatric diseases to allow early diagnosis and treatment of patients, with obvious economical counterparts. In this quest of the human brain neuroimaging and especially MRI has become an inescapable pathway because it allows getting maps of brain structure and function in situ, non-invasively, in patients or normal volunteers of any age. MRI allows brain anatomy of individuals to be visualized in 3 dimensions with great details, as well as networks of brain regions activated by high order cognitive functions, together with stunning images of the connections between those areas. Still, images remain at a macroscopic scale (millions of brain cells), while invasive techniques in animals and tissues explore very small ensembles of neurons. This large gap must be bridged to understand how the brain works, as interaction and synergy exist between all brain levels. One approach is to rely on diffusion MRI, a concept which has been develop from the mid-1980s based on Einstein’s framework to probe tissue structure at a microscopic scale while images remain at millimeter scale through parametrization or modeling, providing unique information on the functional architecture of tissues. Since then, diffusion MRI has become a pillar of modern clinical imaging. Diffusion MRI has mainly been used to investigate neurological disorders, but is now also rapidly expanding in oncology, to detect, characterize or even stage malignant lesions, especially for breast or prostate cancer. In the brain diffusion MRI even allows to reveal dynamic changes occurring in tissue microstructure intimately linked to the neuronal activation mechanisms. On the other hand, outstanding instruments operating to field of 11.7 teslas or above are now emerging to boost the spatial and temporal resolution to not only allow us to “better” see inside our brain, confirming or invalidating our current assumptions on how it works, but also to generate new assumptions and elaborate a kind of “Gauge Theory” to help us decode the functioning of our brain.

Abstract

To overcome the lack of automation and long computational times for advanced PET image reconstruction methods, we present a novel encoder-decoder architecture that quickly reconstructs high quality images directly from PET sinogram data. DeepPET is trained and evaluated on realistic, simulated data, and resulting images have higher quality than conventional techniques, and takes a fraction of the time to generate.

Abstract

Sodium (²³Na) MRI has shown promise for monitoring neoadjuvant chemotherapy (NACT) response in breast cancer. Unfortunately, due to low sodium content in the body, its low MR sensitivity and short relaxation times in biological tissues, ²³Na MRI suffers from intrinsically low signal-to-noise ratio (SNR), which can be up to 20,000 times lower than that of proton. Such low SNR translates into low spatial resolution and long acquisition times. Efforts to alleviate these challenges generally utilize high field systems (≥ 3 T), ultra-short echo time (UTE) acquisition methods, and tailored radiofrequency coils to boost the baseline SNR. Our focus is the coil design aspect. Specifically, we present a dual-tuned multichannel 1H/²³Na bilateral breast coil consisting of volume transmit/receive (Tx/Rx) 1H coils, volume ²³Na transmit coils and an 8-channel ²³Na receive array for 7 T MRI which enabled sodium imaging in vivo with 2.8 mm isotropic nominal resolution (~5 mm real resolution) in 9:36 min. The proposed coil could enable access to even more specific biomarkers of cellular metabolism such as intracellular sodium concentration, and cellular density such as extracellular volume fraction that are still largely unexplored due to the challenges associated with ²³Na MRI.

Abstract

Transcranial Ultrasound (TUS) is an emerging field with a vast range of new potential clinical applications. Here, a series of new human scale TUS devices and the novel benchtop strategies used to develop them in the laboratory will be presented. These devices are intended for brain tumor cancer therapy and for treating neurological disorders such as epilepsy, pain, depression, and essential tremor. The presentation will include the latest developments for: 1) A neuronavigation-guided single-element transducer platform for delivering multi-target pulsed low-intensity TUS to human brain. 2) An integrated scalp sensor for simultaneous electroencephalography and acoustic emission detection. 3) A 3D passive acoustic mapping array device compatible with the FDA approved ExAblate 4000 system for localizing microbubble cavitation. Highlights of each technique relevant to current clinical investigations and future directions of each strategy will be discussed.

Abstract

Sodium(²³Na) MRI can provide unique metabolic information to study the human body and its afflictions. However, the low intrinsic SNR of sodium MRI limits the resolution of the images to 3-5 mm isotropic and necessitates long acquisition times (~10-20 min). Moreover, the necessity to perform 1H and ²³Na acquisitions sequentially prolongs the total scan time, which impedes the wide spread adoption of sodium imaging. In this talk, we will present a technique to simultaneously acquire sodium images and multi-parametric proton maps in one single scan.

Abstract

Prior histological post-mortem studies have highlighted gray matter (GM) microstructural abnormalities as a pathological feature of both schizophrenia (SZ) and autism spectrum disorder (ASD). However, these histological studies were limited by the small sample sizes and focus on restricted brain areas. In this talk, we present our work examining the feasibility of diffusional kurtosis imaging (DKI) to describe gray matter microstructural abnormalities in SZ and ASD non-invasively and in vivo. DKI is an extension of diffusion tensor imaging that accounts for non-Gaussian water diffusion contributions to the diffusion MRI signal and provides several kurtosis indices that reflect tissue microstructural complexity. The talk will review existing research investigating DKI’s use to describe GM microstructure pathology in several clinical populations and animal disease models, as well as our recent findings showing significant differences in kurtosis intensity and lateralization metrics in SZ and ASD populations.

Abstract

Diffusion MRI (dMRI) has the unique ability to study brain microstructure at a resolution much smaller than the MRI voxel itself. The strength of diffusion weighting (i.e., the b-value) strongly impacts what information is contained in the dMRI signal. Since modern scanners have much stronger gradients, high b-value dMRI is becoming more feasible, and its utilization is likely to increase. High b-value acquisitions provide information beyond what is attainable with DTI and have proven useful for fiber tractography and for calculating diffusion measures that have greater biological specificity. This presentation will revisit a high b-value technique known as fiber ball imaging (FBI) but will mostly focus on how it can be used in combination with diffusion kurtosis imaging (DKI) to estimate microstructural parameters, such as compartmental water fractions and diffusion tensors. In addition, FBI provides the opportunity to calculate compartmental transverse relaxation times (T2) while avoiding multi-exponential fitting schemes.

Abstract

Magnetic Resonance Imaging (MRI) is an invaluable diagnostic tool for imaging the human body, diagnosing and characterizing diseases, and developing new treatments. In this work, we describe two applications of a novel MRI technique, Quantitative Ultra-short Time-to-echo Contrast-Enhanced (QUTE-CE) MRI to brain disease.

In a first application, QUTE-CE is employed to quantify nanoparticle accumulation in tumors, which is of great clinical interest for stratifying cancer patients who may benefit from therapeutic nanoparticles. Using FDA-approved superparamagnetic iron oxide nanoparticle (SPION) ferumoxytol in QUTE-CE MRI, we produce quantitative measurement of contrast and delineate clear, positive-contrast brain/tumor vasculature image in mice and rats. QUTE-CE MRI is shown to improve contrast and contrast efficiency compared to conventional high-resolution T1- and T2-weighted imaging. QUTE-CE is ideally suited for non-invasive visualization and quantification of tumor nanoparticle uptake, and accordingly, it can potentially be used for identifying cancer patients who can respond to treatment with therapeutic nanoparticles.

In a second application, QUTE-CE is employed to characterize traumatic brain injury (TBI). TBI is a prevalent risk of death and disability in young people with about 1.6 million cases reported per year in the US. Some of the most devastating injuries from brain trauma are the rupturing of arteries between the dura and the skull in an epidural hematoma (blood brain barrier disruption), as well as tears in emissary veins, resulting in hemorrhagic contusions seen in subdural hematomas. This accumulation of blood can squeeze and increase pressure on the brain. Here, we introduce a novel application of QUTE-CE to image blood accumulation and detect microbleeds in mild TBI animals. Rats which underwent 3 mild concussions showed significant difference in QUTE-CE MRI measure of ferumoxytol accumulation in extravascular space indicating blood brain barrier damage following TBI. These differences were observed primarily in cortex, hypothalamus, basal ganglia, cerebellum and brainstem. This study demonstrates that QUTE-CE MRI can be used to detect blood brain barrier disruption and microbleeds in mild TBI rats.

No abstract was provided for this talk.

Abstract

Glioblastoma (GBM) is the most common primary malignant brain tumor in adults with poor prognosis. The standard of care for patients with GBM includes maximal surgical resection and concurrent chemo-radiation therapy followed by 6 to 12 cycles of adjuvant temozolomide (TMZ). Standard therapeutic approaches provide modest improvement in progression-free and overall survival, necessitating the investigation of novel therapies. Recently, FDA approved the use of tumor-treating fields for the treatment of patients with GBM. Additionally, several immunotherapeutic modalities such as chimeric antigen T cell receptors, check-point inhibitors and dendric cell vaccines hold much promise in the future treatment paradigms for these patients. In this presentation, I will discuss the potential roles of 3D-echoplanar spectroscopic imaging, diffusion and perfusion MR imaging techniques in evaluating treatment response in patients with GBM receiving established and novel treatment modalities. As non-invasive identification of patients harboring isocitrate dehydrogenase (IDH) mutant gliomas can have significant clinical implications, I will also present our initial experience on the utility of 2D-correlational spectroscopy in identifying glioma patients with IDH mutation.

Abstract

Diffusion MRI is sensitive to the length scale of tens of microns, which coincides to the scale of microstructure in the human brain tissue. By varying the diffusion time, we can evaluate the brain micro-geometry via time-dependent diffusion measurements and the biophysical modeling. To validate our model, we segmented 3-dimensional realistic microstructure of the mouse brain white matter and performed Monte Carlo simulations of the diffusion in segmented axons. This talk will focus on the time dependence either along or perpendicular to white matter axons and corresponding micro-geometries, such as axonal diameter variation.

Abstract

Fiber Ball Imaging (FBI) is a diffusion MRI method that estimates the orientation of axonal fibers in white matter from an inverse Funk transform. This approach avoids the need for numerical fitting to a signal model and for a fiber response function. FBI also yields predictions for certain microstructural parameters, including the fraction anisotropy axonal. When combined with triple diffusion encoding MRI, FBI can also be used to find the intra-axonal diffusivity and the axonal water fraction. This talk will focus on the basic concepts that underlie FBI but will also show data that support its validity and illustrate its application.

Abstract

Niek Van Overberghe (International Sales Manager @ MOLECUBES) will present on the unique technology at the core of the β-,γ and X-CUBE, preclinical imagers for PET, SPECT and CT. This new generation of in vivo imaging systems makes use of monolithic crystals coupled to solid state siPMs taking imaging one step further, combined with an in vivo CT system that ensures fast and low dose acquisitions. Thanks to this new technology, researchers can now inject lower activities, scan for a shorter time, hereby reducing the stress level on animals, increasing throughput, lowering radiotracer cost, and lowering the dose of the operator. Because of their unique bench top size, the instruments can be used in any lab around the world without needing building modifications. In addition, Niek will present on different applications that highlight the superior capabilities of these bench top modular systems compared to older systems.

Abstract

Regardless of improved accuracy scores and other metrics, deep learning methods tend to be overconfident on unseen data or even when predicting the wrong label. Bayesian deep learning offers a framework to alleviate some of these concerns by modeling the uncertainty over the weights generating those predictions. This talk will introduce Bayesian deep learning and present the use of Bayesian NNs for outlier detection in the medical imaging domain, particularly the application of Brain lesion detection.

Abstract

Since its inception 45 years ago, development of MRI systems has been predominantly driven by commercial entities and innovation is centered on the commercial interests of these vendors. In a relatively small ecosystem of MRI manufacturers that compete in a low quantity, high profit margin market, innovation is effectively controlled by the manufacturer’s openness to outside access and is often limited by the manufacturer’s market needs. The possibilities are even more limited when it comes to disruptive modifications of the scanner hardware. In my presentation, I’ll discuss the need for and show the prospects of an open-source MRI system for education, disruptive innovation, and accessible healthcare and will show the results of educational work with a $500 fully open-source MRI spectrometer.

Abstract

The talk will outline recent approaches for using (deep) convolutional neural networks to solve a wide range of inverse problems, such as image reconstruction in medical tomography. A key element is to use a neural network architecture for reconstruction that includes physics based models that describe how data is generated as well as its statistical properties. Another is the possibility to integrate complex task related a priori information and elements of decision making into the reconstruction procedure. The resulting approach outperforms current state-of-the-art in terms of ‘quality’, computational speed and there is no need to manually set parameters as with variational methods. Furthermore, the amount of training data and network size can be kept surprisingly small. The talk will also touch upon further developments based on using generative adversarial networks for uncertainty quantification.

Abstract

Ultrafast breast MRI provides excellent data for automated breast lesion classification. Incorporating morphology, T2 and DWI using various automated approaches, improves the classification task slightly, but late phase dynamics seem redundant. Since about a 3rd of detected breast cancers are missed on prior breast MRI while in retrospect clearly visible, there is a strong incentive for the development of systems that not only classify, but automatically detect breast lesions in MRI volumes. Using a deep learning system about 70% of these cancers can be marked at 2 false positives per volume. Since automated detection also demands automated segmentation of the breast and the fibroglandular tissue to determine the search space, these algorithms may potentially also be used for risk prediction and prognostication based upon qualified measures of fibroglandular tissue and background parenchymal enhancement.

Abstract

Intracellular metabolites that give rise to quantifiable MR resonances are unique structural probes for the intracellular space, and are oftentimes specific, or preferential enough to a certain cell type to provide information that is also cell-type specific. In the brain, N-acetylaspartate (NAA) and glutamate (Glu) are predominantly neuronal/axonal in nature, whereas soluble choline compounds (tCho), myo-inositol (mI) and glutamine (Gln) are predominantly glial. The diffusion properties of these metabolites, examined by diffusion weighted MR spectroscopy (DWS) exclusively reflect properties of the intracellular milieu, thus reflecting properties such as cytosolic viscosity, macromolecular crowding, tortuosity of the intracellular space, the integrity of the cytoskeleton and other intracellular structures, and in some cases – intracellular sub-compartmentation and exchange.

The presentation will introduce some of the methodological concepts of DWS and the particular challenges of acquiring robust DWS for accurate estimation of metabolite diffusion properties. Subsequently, the unique ability of DWS to characterize cell-type specific structural and physiological features will be demonstrated, focusing on combining DTI/DWI and DWS in a combined analysis framework aimed at better characterizing tissue microstructural properties, as well as acquisition strategies aimed at characterizing compartment-specific microscopic anisotropy (µFA) in tissue. Also presented are applications of DWS to discern cell-type specific intracellular damage in disease.

Abstract

Emerging novel analysis techniques offering insights into microstructure and tissue properties require more and more eloquent data. This talk will introduce some of our recent advances on the acquisition side, presenting multi-parametric diffusion acquisitions – extending the parameter space to allow integrated T1, T2* and Diffusion sampling (b-value,b-vector, b-shape) within reasonable imaging times. Combination with Multiband imaging and sampling strategies in this multi-dimensional space will be discussed and exploratory data-driven analysis results presented.

Abstract

AI, in its much misinterpreted form, holds the promise of revolutionizing medical imaging in healthcare. In practice, however, many challenges remain. This talk presents some of the challenges that we, as a company, have recognized on the way of building better tools for radiologists. In particular, Dr. Li Yao, the Lead Data Scientist at Enlitic, will share with the audience three project stories, one with Chest X-ray, one with Chest CT, and one with medical text reports, each of which highlights unique excitement and challenge in the real world clinical context. The talk will be overall technical on AI and machine learning side.

Abstract

The terms artificial intelligence, machine learning, deep learning, or computer vision are mentioned increasingly often in the radiology community. In this talk, Dr. Mazurowski will talk about how these methods can help radiologists in their clinical practice as well as how they can advance science by improving our understanding of cancer. The talk will be concluded with more general thoughts on the future of the radiology profession in the advent of human-level artificial intelligence. Dr. Mazurowski is an Associate Professor of Radiology, Electrical and Computer Engineering, and Biostatistics and Bioinformatics and Duke University. He leads a research laboratory with focus on applications of machine learning to cancer imaging.

Abstract

Being sensitive to tissue structural features on the micrometer level (microstructure), diffusion MRI combined with biophysical modeling has the potential to map relevant biological properties on scales far below the nominal voxel resolution. In the brain and spinal cord, much work in this direction has been based on a relatively simple biophysical model of diffusion, recently dubbed “the standard model”. This model characterizes the diffusion signal in terms of a handful of relevant parameters: neurite volume fraction, intra-neurite and extra-neurite diffusivity, and the neurite or fiber orientation distribution. In this talk, I will give some background for the standard model and an overview of our work with it, covering our efforts to validate the model in animal model systems including some comparison with histology. I will also outline some current problems with the model and ongoing attempts to overcome them.

Abstract

Access to clean and voluminous datasets is a piece of luxury confined to academic research for many machine learning applications. In practice, such datasets are hard to come by, and consequently limit the performance of deployed machine learning systems. This problem is pervasive in medical imaging applications where the cost of data acquisition and labelling is high. In this talk, I will present a method that is capable of learning more intelligently from such noisy data by modelling the human annotation process. This is particularly relevant in situations where data is labelled by multiple annotators of varying skill levels and biases.

Abstract

When cells are exposed to high voltage electric pulses their membranes become transiently permeable, i.e. molecules otherwise deprived of transmembrane transport molecules can gain access into the cytosol. This phenomenon is called electroporation. It can be reversible – cells survive or irreversible, if cells die. The former is used to introduce genes into cells for gene therapy and DNA vaccination (gene electrotransfer) or to increase effectiveness of some chemotherapeutic drugs (electrochemotherapy), while the latter is used as a non-thermal tissue ablation method.

Electroporation of cells depends on local electric field to which cells are exposed. In vivo in tissue electric field is impossible to measure directly. Therefore current density imaging and magnetic resonance impedance tomography have been used to elucidate electric field distribution and was correlated with cell membrane permeabilisation.

Abstract

Novel molecular imaging techniques are allowing us to visualize breast tumor biology in unprecedented molecular detail. These include the use of magnetic resonance spectroscopic imaging, mass spectrometric imaging, and Raman imaging for mapping molecular and metabolic pathways in breast cancer. Applications of these molecular imaging techniques are improving our understanding of metabolic and oncogenic signaling in breast cancer progression, metastasis, and response to therapy. Finally, we are also investigating the processes that lead to “molecular priming” of metastatic target organs prior to the arrival of the first metastasizing cancer cells.

Abstract

Millions of people worldwide suffer from metabolic bone diseases, predisposing them to bone fractures and devastating consequences. Within a year of a hip fracture, for example, 20-30% of patients die and 50% lose the ability to walk. Medical imaging plays an important role in the diagnosis of bone disease, staging, fracture risk assessment, and monitoring of treatment. Radiographs and dual energy X-ray absorptiometry (DXA), which provides semi-quantitative assessment, are the modalities of choice for clinical management of metabolic bone diseases. Recent advances in medical imaging technologies and analysis techniques have enabled novel non-invasive approaches for quantification of bone quality. For example, it is now possible to obtain three-dimensional high-resolution images depicting the trabecular and cortical microstructure in human subjects. Ability of obtain high resolution images has paved the way for elegant image analysis algorithms for extracting information about various aspects of bone quality not feasible previously. For example, it is now possible to characterize trabecular bone microarchitecture using multi-row detector CT and the tensor scale algorithm or estimate the hip fracture strength using high-resolution imaging based finite element analysis. More recently, MRI, CT, ultrasound, and PET techniques have been developed for extracting novel biomarkers related to bone strength. For example, bone water assessed by MRI has been proposed as a new biomarker for bone quality. Many of these imaging-based techniques could provide early differential diagnosis, periodic monitoring, and a comprehensive assessment of bone quality thereby potentially changing the way metabolic bone diseases are managed in the future.

No abstract was provided for this talk.

Abstract

The famous Shannon-Nyquist theorem has become a landmark in the development of digital signal processing. However, in many modern applications, the signal bandwidths have increased tremendously, while the acquisition capabilities have not scaled sufficiently fast. Consequently, conversion to digital has become a serious bottleneck. Furthermore, the resulting high rate digital data requires storage, communication and processing at very high rates which is computationally expensive and requires large amounts of power. In the context of medical imaging sampling at high rates often translates to high radiation dosages, increased scanning times, bulky medical devices, and limited resolution.

Abstract

While we possess rather detailed understanding of select micro- and macroscopic processes of normal human brain development, we know far less about how brain changes relate to behavioral changes over the course of life from the prenatal period to early adulthood. This lack of understanding is especially pronounced in very early years of human life, years where change is most rapid, and vulnerability heightened. The primary objective of our research is to characterize fundamental properties of human brain macroscale neural system development, and examine how early experiences, beginning in utero, influence life-long learning and neurological health. We are testing models in which early psychosocial stress and concomitant toxin exposure influence development of neural systems, particularly those that support the establishment of cognitive control and regulatory processes in childhood. Rigorous evaluation of emergent self-regulatory processes and their neurological and biobehavioral bases has potential to inform educational strategies and lead to biologically-informed behavioral interventions for those with enhanced risk.

No abstract was provided for this talk.

Abstract

Decade-old measurements of the MRI signal phase in the human brain white matter ignited a still-ongoing discussion of how to calculate the Larmor frequency of NMR-visible spins in magnetically heterogeneous media. While this discussion has somewhat decoupled from the original biomedical context going deep into NMR physics and questioning the assumptions behind the Lorentz cavity construction, its practical implications may significantly limit the applicability of quantitative susceptibility mapping (QSM). This talk will give an overview of the biophysical origins of the Larmor frequency offset. A simple model, which remains in the debate focus, is used to illustrate the relation between the microstructure and the Larmor frequency. A closed-form analytical solution for this model is obtained in the practically relevant limit of fast diffusion. This solution illustrates the microstructural correlates of the recent empirical nerve tissue description, adds to the discussion of the Lorentz cavity construction in heterogenous media, and formulates the major challenge for the QSM. The talk will conclude with a discussion of the unresolved problems on the way to building realistic models for white matter magnetic microstructure.

Abstract

Epilepsy affects approximately 2.2 million people in the United States. Thirty percent of epilepsy is refractory to pharmacotherapy, and surgical treatment of refractory epilepsy can often be the most effective treatment option. Investigations of the resected tissue of MRI-negative subjects suggest that there exist focal epileptogenic lesions, amenable to resection, that are not detectable using current clinical MRI protocols.

Magnetic resonance spectroscopic imaging (MRSI) provides metabolic information which is complimentary to structural imaging. We have developed a novel B1-insensitive semi-adiabatic spectral-spatial imaging sequence (SASSI) which was designed to overcome many of the limitations of MRSI at ultra-high fields, enabling effective acquisition of high resolution grids of spectra. We have used SASSI to detect metabolic alterations in the head and body of the hippocampus of patients with focal epilepsy who were non-lesional or inconclusive in their clinical MRI exams.

No abstract was provided for this talk.

Abstract

Dynamic Causal Modelling (DCM) is an advanced biophysical model which explicitly describes the entire process from experimental stimuli to functional magnetic resonance imaging (fMRI) signals via neural activity and cerebral hemodynamics. To conduct a DCM study, one needs to represent the experimental stimuli as a compact vector-valued function of time, which is hard in complex tasks such as book reading and natural movie watching. Deep learning provides the state-of-the-art signal representation solution, encoding complex signals into compact dense vectors while preserving the essence of the original signals. There is growing interest in using Recurrent Neural Networks (RNNs), a major family of deep learning techniques, in fMRI modeling. However, the generic RNNs used in existing studies work as black boxes, making the interpretation of results in a neuroscience context difficult and obscure.

In this paper, we propose a new biophysically interpretable RNN built on DCM, DCM-RNN. We generalize the vanilla RNN and show that DCM can be cast faithfully as a special form of the generalized RNN. DCM-RNN uses back propagation for parameter estimation. We believe DCM-RNN is a promising tool for neuroscience. It can fit seamlessly into classical DCM studies. We demonstrate face validity of DCM-RNN in two principal applications of DCM: causal brain architecture hypotheses testing and effective connectivity estimation. We also demonstrate construct validity of DCM-RNN in an attention-visual experiment. Moreover, DCM-RNN enables end-to-end training of DCM and representation learning deep neural networks, extending DCM studies to complex tasks.

Abstract

In this talk, I will present 3 different roles of total variation (TV) regularization in variational methods for unsupervised image segmentation in computer vision, deconvolution in quantitative susceptibility mapping (QSM), and image reconstruction for fast multicontrast MRI. First, TV as a measure of the perimeter of a candidate partition encoded by the indicator function of a set. In unsupervised image segmentation, the total length of region boundaries is often minimized to obtain a compact partition that likely matches the way humans perceive. A statistical distance between color distributions of distinctive regions in a candidate partition is maximized with the minimization of TV for unsupervised image partitioning. Second, TV as a measure of the amount of streaking artifacts in QSM deconvolution. QSM is a noninvasive MRI method for a quantitative study of the tissue magnetic susceptibility distribution by solving magnetic field to susceptibility source inversion problem. A major challenge in the ill-posed inverse problem is streaking artifacts from noise in the field which propagates at the complementary magic angle. These artifacts can be selectively reduced by weighted TV regularization that makes use of anatomical information of the corresponding magnitude image. Lastly, TV as a measure of undersampling artifacts in image reconstruction for multicontrast MRI. In clinical MRI, multiple contrasts such as T1w, T2w, and FLAIR are sequentially acquired, consequently taking a long scan time. To shorten such a long scan time, structural information that exists between contrasts is extracted from T1w and is incorporated into the TV term as an orthogonal projector in the model-based image reconstruction for the subsequent contrasts that are highly undersampled.

No abstract was provided for this talk.

Abstract

Magnetic Particle Imaging (MPI) is a new tomographic imaging modality that offers high spatial and temporal resolutions. Compared to the other imaging modalities such as MRI/CT/PET, MPI is non-toxic, more sensitive, and fully quantitative technique. MPI addresses clinical and research needs for safe diagnostic and therapeutic applications such as cancer screening, cell tracking, and angiography. To date, a few small-bore MPI systems have been developed, however, human-size MPI scanner has yet to be built. The major challenge of scaling up of MPI is in high power consumption that is associated with the traditional approach to designing the scanner. In my talk, I will overview the basics of MPI, specifically, physics and instrumentation that includes two fundamental types of MPI topologies: field-free-point and field-free-line. Then I will describe my approach to designing MPI scanner and also will show how traditional MPI can be blended with atom optics to incorporate an atomic magnetometer as a very sensitive way of detecting the signal.

Abstract

With billions of mobile devices worldwide and the low cost of connected medical hardware, recording and transmitting medical data has become easier than ever. However, this ‘wealth’ of physiological data has not yet been harnessed to provide actionable clinical information. This is due to the lack of smart algorithms that can exploit the information encrypted within these ‘big databases’ of biomedical time series and take individual variability into account. Exploiting these data necessitates an in depth understanding of the physiology underlying these biomedical time series, the use of advanced digital signal processing and machine learning tools to recognize and extract characteristic patterns of health function, and the ability to translate these patterns into clinically actionable information.

In this talk I will present my research in electrophysiology, namely the “Fetal Holter AI” and “Cardio AI” projects. For these two research projects I leverage state-of-the-art signal processing and machine-learning techniques to harness physiological information contained in biomedical time series and provide clinically actionable information. The “Fetal Holter AI” project aims to create a novel intelligent non-invasive fetal Holter electrocardiogram system to diagnose for fetal arrhythmias and remotely monitor the fetal cardiac health. The “Cardio AI” project has two aims: (1) to better understand the physiological information contained in the heart rate variability i.e. the time interval variation between consecutive heartbeats. I will present a new software, PhysioZoo, which we developed in our laboratory at the Technion for analyzing the heart rate variability from animal models; (2) to create an artificial intelligence system which can identify cardiac pathologies from the electrocardiogram with accuracy similar to the cardiologist’s direct interpretation. I will finish my presentation by briefly mentioning the “SmartCare Sleep AI” project which aims at creating a single channel screening test for obstructive sleep apnea. I will present my work in elaborating this test using patterns recognition from the oximetry time series in order to recognize individuals with this medical condition.

Abstract

Our brain depends on continuous blood flow to deliver the oxygen and nutrients it needs to function. Disruption to this oxygen supply, as in cerebrovascular diseases, has devastating consequences, most strikingly in acute stroke. Noninvasive imaging of brain blood flow and metabolism is technically challenging, but would provide critical information to diagnose and select therapies for patients.

My mission is to engineer new imaging biomarkers of brain physiology to address this need. In this talk, I describe development of a novel magnetic resonance imaging (MRI) technique to quantify oxygenation in cerebral blood vessels. I also validated MRI methods to measure cerebral blood flow against the reference standard by positron emission tomography (PET), using state-of-the-art simultaneous PET/MRI hardware. I performed these studies in challenging cerebrovascular patient cases, including Moyamoya disease, and used imaging to inform our basic understanding of disease pathophysiology.

In the long term, the imaging tools I develop will establish a vascular “fingerprint” that succinctly captures the metabolic health of an individual, and alerts us to a broad set of neurological diseases in its earliest stages.

No abstract was provided for this talk.

Abstract

Current clinical MRI often consists of a series of qualitative or weighted measurements of tissue properties, such as T1-weighted or T2-weighted images. These qualitative measurements have some inherent limitations. The relative contrasts from these images may change depending on the set-up of the acquisitions, the type of the scanners and so on. The interpretation of images thus only relies on subjective assessment or morphological measurement. This limits the ability to diagnose pathology in a reproducible and reliable manner, to characterize tissues and lesions and to longitudinally follow up lesions or to assess response to novel therapies. The weighted contrast from multiple underlying tissue properties may also reduce the sensitivity and specificity to detect and characterize subtle and diffuse diseases. Although these limitations could be overcome by collecting fully quantitative tissue maps using quantitative MRI techniques, the adoption of quantitative MRI in clinical practice is hampered due to its long scan time, low repeatability and lack of robustness.

This talk will introduce the concept and technical advances of magnetic resonance fingerprinting (MRF), which is a robust and flexible framework for fast, multi-parametric quantitative MRI. This technology allows quantification of multiple key tissue properties, such as T1, T2, T2*, and perfusion in a clinically feasible time and with high repeatability, which overcomes the barrier of clinical adoption of quantitative MRI. Since MRF is a dictionary based method that has no requirement of the encoding methods and signal shapes, this technology also allows flexible sequence designs for various clinical applications, and flexible numerical simulation for sophisticated physical and physiological settings. Both features contribute to more robust and accurate quantitative results. Finally, the talk will discuss some clinical applications of MRF, demonstrating promising clinical translation of this technology.

Abstract

MRI with ferumxotyol as a intravascular contrast agent holds great promises for a number of clinical applications. In this talk, Dr. Hu will discuss translational ferumoxytol-enhanced cardiovascular MRI techniques that enables new paradigms for imaging congenital heart disease and beyond. In the second half of the talk, Dr. Hu will discuss his recent work in developing MRI techniques for guiding radiation therapy with regard to anatomical tracking and tumor response assessment.

Abstract

The peripheral nervous system is primarily composed of nerves that transmit motor and sensory information between the spinal cord and the body. Damage to these nerves results in a wide array of symptoms, ranging from temporary numbness, tingling, and pricking sensations to burning pain, muscle weakness, paralysis, organ failure, and death. Although clinicians have tools for assessing peripheral neuropathies (e.g., nerve conduction studies), they provide limited information in proximal and/or transected nerves. Quantitative MRI techniques (e.g., diffusion and magnetization transfer) may overcome these limitations by providing assays of myelin and axon pathologies throughout the peripheral nervous system. Unfortunately, few studies have applied quantitative MRI techniques to study peripheral neuropathies in humans in vivo. This can be attributed to the technical challenges associated with peripheral nerve MRI, including the need for higher spatial resolution in feasible scan times, a lack of contrast on standard anatomical images, and the influence of surrounding fat. In this talk, Dr. Dortch will discuss i) methods to overcome the technical challenges associated with peripheral nerve MRI and ii) applications of quantitative MRI methods in inherited neuropathies and trauma.

Abstract

Advances in simultaneous positron emission tomography (PET) and magnetic resonance imaging (MRI) have enabled novel approaches for in vivo functional brain mapping. The complementary nature of the imaging signals acquired by PET and functional MRI (fMRI) permits new insights into neurotransmission of the living brain: fMRI localizes changes in brain activity, whereas PET captures the underlying molecular and receptor-specific dynamics. One of the potentials of this technology is to provide new clinical biomarkers for the evaluation of dynamic receptor function and therapeutic interventions.

This talk will describe how simultaneous functional imaging with PET/fMRI leads to novel mechanistic insights through neuromodulation of brain function. The focus will be on interventions that target the dopamine receptor system, either through pharmacological or direct electrical stimulation. I will show that neurovascular coupling to receptors as identified by PET/fMRI can be used to classify drug properties in vivo. Together with biological and pharmacokinetic models, mechanistic insight into receptor adaptations over time can be gained. I will then talk about the in vivo effects of deep brain stimulation, and how the combined use of experimental approaches allows us to unravel receptor subtype contributions to observed signal changes. Finally, I will show how PET and fMRI can be used for evaluating the effects of flow, and how the combination of both modalities can provide alternative approaches for evaluating radiotracer probes of novel in vivo receptor targets.

No abstract was provided for this talk.

Abstract

Chemical exchange saturation transfer (CEST) and the closely related off-resonance T1ρ methods are gaining popularity for their ability to visualize chemical exchange process between protons bound to solutes and surrounding bulk water, thus providing a contrast based on proton exchange sites with different chemical shifts. The contrast is also influenced by pH, temperature and molecule concentration. Therefore, CEST/T1ρ can provide molecular level information which reflects biochemical composition of tissues and their microenvironments. As a result, many promising applications of CEST imaging are explored, including but not limited to brain tumor imaging, brain ischemia, prostate cancer, breast cancer, kidney pH measurement and cartilage quality assessment. In the meanwhile, a lot of efforts are made to develop fast and quantitative CEST imaging methods to push CEST techniques toward clinical use.

To accelerate quantitative CEST imaging, we have developed a method based on the balanced steady-state free precession sequence as an alternative way for chemical exchange detection (bSSFPX). The feasibly of bSSFPX for chemical exchange detection was proved both theoretically and experimentally on phantoms. Mathematical models for bSSFPX were developed for quantitative measurements of T1ρ and exchange rate. bSSFPX was applied in the human brain. While the exact origin of the contrast is still under investigation, we hypothesize it is due to the chemical exchange from fast exchanging metabolites with resonance frequencies close to water. Detection of these metabolites is challenging for standard CEST imaging methods at 3T.

While application of CEST to brain malignancy is increasing, its application in body imaging is still challenging. One of the difficulties is the presence of large lipid signals. We have studied the influence of non-exchanging fat on CEST imaging using simulation, phantoms, and in vivo studies at different fat fractions and echo times. To remove the fat influence on body CEST imaging, we have developed a CEST-Dixon imaging sequence for fat free CEST imaging and applied it to human breast malignancy characterization at 3T. We have demonstrated that the CEST-Dixon sequence eliminates lipid contamination robustly in breast CEST imaging. The results display a potential for improved non-invasive characterization of human breast lesions at 3T using CEST, potentially differentiating more aggressive from less aggressive tumors.

Abstract

The morphology of the white matter&once referred to as nature’s finest masterpiece&is intricately coupled with brain function. Being able to measure the white matter structure, and its pathological changes, in vivo and non-invasively would promote the study of brain function and the more specific diagnosis of brain disorders. Despite a limited spatial resolution, the sensitivity of diffusion MRI to the Brownian motion of protons restricted by cellular structures, such as axons, provides an exciting avenue to reveal the microscopic architecture. However, bridging the length-scale gap requires the development and validation of a biophysical white matter model that decomposes the signal in components that probe specific features of the underlying microstructure, e.g. axon diameters. During this talk, I will focus on my recent work on the mapping of micrometer-thin axons using MRI, from preclinical validation to a clinical perspective.

Abstract

The motivation of this talk is the computational encoding of neuroanatomy in terms of tissue characteristics as well as classical neuroanatomical knowledge.

The first problem to address will be the in vivo dissection of the human brain’s white matter from diffusion magnetic resonance imaging. We address this through representing current anatomical knowledge computationally. In this talk I will introduce computational tools to represent human anatomy. More precisely, I will introduce a domain specific programming language to represent and automatically extract the major white matter structures in the human brain’s white matter, the white matter query language (WMQL) as well as applications of these techniques to dyscalculia and schizophrenia.

Then I will move on to presenting techniques to perform group-based studies to parcel the cortical mantle based on white matter connectivity. Specifically, I will show how leveraging a consistent mathematical model of axonal-based cortical connectivity we are able to separate subject and parcel-specific characteristics in a random effects model. In particular, the motor and sensory cortex are subdivided in agreement with the human homunculus of Penfield. We illustrate this by comparing our resulting parcels with the motor strip mapping included in the Human Connectome Project data.

Finally, I will provide a prospective view on NeuroLang, project that has the goal of extending the concepts explored in the previous to takes to computational neuroanatomy of the white matter to the bulk of human neuroanatomy.

Abstract

In recent years there is an increased interest to build massively parallel transmit (pTx) systems for high spatial-temperal resolution MR imaging. A novel pTx pulse compression method (array-compressed pTx, acpTx) is proposed, validated, and implemented that allows a many-element parallel transmit array to be driven by only a few power amplifiers without significantly degrading the pulse performance. AcpTx provides a new insight into pTx array design by synergistically integrating both Maxwell and Bloch principles. By maximizing the pulse performance for a pTx system with limited number of power amplifiers, acpTx can further improve highly accelerated MR imaging with multichannel reception. Specifically, acpTx is shown to improve the motion and phase errors of multi-shot EPI that would enable short TE fMRI and dMRI applications compared to single-shot EPI. In addition, the presentation will also include recent advances in improved multichannel compressed sensing reconstruction for PRF temperature imaging for MRgHIFU and high field RF heating monitoring, as well as a novel application of dielectric materials to accelerate abdominal imaging.

Abstract

Conventional gradient and shim coils for MRI generate pure spherical harmonic “B0” magnetic fields for shimming and spatial encoding, simplifying the image acquisition process. Unfortunately, these coils are difficult to build and impose high demands for space, cooling, and power consumption. In this talk, I will discuss a recent trend toward flexible spatial encoding and field control methods that use non-orthogonal magnetic field basis sets or alternative encoding mechanisms. As an example of this trend, I will discuss progress at MGH on a portable brain scanner that uses a rotating permanent magnet array with a built-in spatial encoding field to perform imaging without conventional B0 gradient coils, greatly reducing cost, weight, and power consumption. In this approach, imperfections in the linear encoding field are accounted for in the encoding matrix during image reconstruction, shifting the engineering burden away from complex hardware and into software. As a second example of flexible spatial encoding, I will show how spatial encoding can also be performed using lightweight, tailored radiofrequency coils whose “B1+” transmit field has a linear phase variation over space, enabling phase-encoded imaging in spin echo train sequences. Finally, I will review progress on multi-coil arrays of shim coils that use a non-orthogonal basis set to dynamically null localized, high-spatial order patterns of B0 inhomogeneity in the body. I will further show how multi-coil shimming can be physically integrated with the RF receive array to save space near the body. I’ll then demonstrate how multi-coil shimming can benefit echo planar imaging, MR spectroscopy, and selective excitation, overcoming some of the limitations of conventional spherical harmonic shim coils.

No abstract was provided for this talk.

Abstract

Cerebral thermoregulation is poorly understood but critical to brain homeostasis and viability. Temperature disturbances strongly potentiate cerebrovascular and other CNS injury, and represent potent targets for neuroprotection. Interrogation of brain temperature has historically been limited to costly and highly invasive implantable probes, and pragmatic approaches to measuring spatiotemporal temperature gradients are lacking. Cerebral MR thermometry may provide safe, non-invasive, and reproducible characterization of brain temperatures across physiologic, ischemic, and other pathologic disease states. This presentation will discuss initial experience with chemical shift thermometry as a biomarker of cerebrovascular injury in human and nonhuman primates, emphasizing the critical role of brain temperature at the intersection of perfusion, metabolism, and cytotoxic injury.

Abstract

Our research focuses on developing techniques that dramatically improve the efficiency of MRI by collecting/deriving more information from each unit time of data acquisition. The overarching goal in creating these strategies is twofold:

• pushing the limits of spatial/temporal resolution and CNR of MRI to make it a better neuroscientific tool;
• improving throughput, motion robustness and efficiency of clinical exams to make MRI more cost effective and more widely used in the clinic.

We are pursuing these goals in a number of areas, including structural, diffusion and spectroscopic imaging, as well as the quantitative techniques of MR fingerprinting, susceptibility mapping and myelin imaging. The order of magnitude efficiency gain we achieved for these acquisitions has been possible through a joint design approach that combines new hardware capabilities, new sequence and readout design, and novel image reconstruction exploiting sparsity, mutual information and deep learning.

Abstract

To acquire high-resolution and –sensitivity images at ultra-high field magnetic resonance (MR) scanner, various hardware solutions can be utilized: dedicated RF coil design for a certain anatomical region of the human brain, merging high dielectric constant materials with the existing RF coil concepts, use of multi-channel transmit RF coil arrays on a parallel transmit system to steer any signal amplitude or phase. Besides these solutions, MR safety limitations have been wisely investigated, i.e. simultaneous EEG-fMRI setup at 7 T MR is simulated with finite difference time domain method to assess its RF safety. My talk will address various RF hardware solutions for 7 Tesla human proton and multi nuclei MR imaging and spectroscopy.

Abstract

This talk will present an overview of Deep Learning (DL) and discuss some recent successful applications in medical imaging. One aim is to draw connections between DL methods such as convolutional neural network (CNN), convolutional encoder-decoder (CED), cycle-consistent adversarial neural network (Cycle-GAN) and medical applications including image reconstruction, multi-modality image synthesis and image analysis. Dr. Liu will present some of his recent work using DL for medical imaging applications and will discuss relevant DL methods and their strengths and limitations. The talk will conclude with a discussion of open problems in DL that are particularly relevant in medical imaging and the potential challenges of DL in this emerging field.

Abstract

Magnetic Resonance Imaging (MRI) is one of the most state-of-the-art technologies to non-invasively acquire structural, functional, and biochemical information in the human body. Dr. Sohn will present his research topics to overcome technical barriers to increase the accessibility of MRI and improve the quality of MR imaging for next-generation MRI that realizes ultra-low RF power, miniaturization, lightweight, low cost, and safety. Most of his researches are related to the oscillating field (B1) of RF coils and interface circuits between RF coils and RF signal chains in transmit (Tx) and receive (Rx). Especially, he is focusing on the development of RF/mixed-signal circuits for simultaneous transmit and receive (STAR) and automatic correction systems of frequency tuning, impedance matching, and RF coupling as well as novel RF coil structures. His research results show the ultra-low RF peak power capability and replacement of manual adjustments to obtain human MR images. These RF hardware-engineering approaches can contribute to a wide variety of MRI researches and industries.

No abstract was provided for this talk.

Abstract

Medical imaging has been targeted as a source of inordinate — and sometimes unnecessary — health care spending. Cross-sectional imaging use has increased markedly over recent decades, and yet the contribution of imaging to overall care has not been well characterized. Rather than parsimony, the goal of our system is to improve evidence-based clinical practice so that the right patients receive the right interventions. Outcomes research is necessary to drive this effort. In this talk, I will discuss the ways in which outcomes research can underscore the value of imaging, and review study designs that explore the best uses of imaging tests. MRI techniques that may perform better than available tests or offer similar performance at lower costs can be evaluated using intermediate health outcomes. Meanwhile, techniques for which larger, prospective studies are available can be evaluated for population-level benefits and harms. Finally, I will discuss some of the ongoing efforts at NYU to bridge the use of MRI to patient health outcomes and decision-making.

Abstract

Patients with epilepsy uncontrolled by medications are potential candidates for epilepsy surgery. Surgical removal of an epileptic lesion can lead to alleviation or elimination of seizures. Majority of epileptic lesion(s) can be detected as structural abnormalities on anatomical (1.5T) MRI scans. However, anatomical MRI scans in a significant proportion of medically refractory epilepsies can be ambiguous or negative. In these non-lesional patients, PET-FDG is indicated for detection of the epileptic focus. Recent technological advances in medical imaging have led to the development of hybrid PET/MRI scanners which combine the two versatile imaging modalities in one scanner. It is predicted that PET/MRI will allow higher rates of lesion localization in medically refractory epilepsies, leading to improved surgical outcomes. In Ontario, access to PET/MRI scanners have improved from one scanner in 2012 to four scanners by the end of 2018. In this talk, I will share experiences from the Epilepsy Imaging Program at London Health Sciences Center in establishing indications for the use of PET/MRI in clinical management of medically refractory epilepsy in Ontario. In addition, I will briefly discuss some of the technical developments in advanced MRI (7T, BOLD-fMRI, DTI) that are implemented in London for clinical epilepsy imaging.

Abstract

Developing precision medicine requires accurate prediction markers and/or models to identify the personalized disease (e.g., cancer) risk and prognosis or response to the different treatment. Radiographic medical imaging is widely used in clinical practice and carries much useful information to phenotype disease risk and prognosis. However, how to reliably and quantitatively extract and compute the useful image features, which can be used to develop new and highly performed clinical prediction models remain a very challenged and hot research topic in the biomedical imaging and informatics field. In this presentation, I will discuss the general concept of applying the quantitative image feature analysis in this research field and report several research work recently conducted in our laboratory to identify new quantitative imaging markers and apply machine learning technology to develop new prediction models, which include (1) using a new imaging marker based on the bilateral mammographic density asymmetry computed from the negative mammograms to predict risk of cancer detection in the next subsequent mammography screening; (2) extracting image features from breast MR images to predict complete response (CR) of breast tumors to the neoadjuvant chemotherapy; (3) using tumor density heterogeneity features computed from lung CT images to build a prediction model to assess lung cancer recurrence risk after surgery; and (4) using image features computed from abdominal CT images to predict response of ovarian cancer patients to chemotherapy at the early stage of the clinical trials.

Abstract

This talk will describe recent developments in several areas of magnetic resonance hardware and sequences which have been applied to clinical research and patient scanning at field strengths between 1.5 and 7 Tesla. Topics will include the design of very high permittivity materials/metamaterials for improved magnetic field homogeneity and lower power deposition, new ceramic-based resonators for multi-element transmit arrays, methods for the rapid non-invasive estimation of tissue conductivity, high resolution motion-free imaging of the eye, and whole-body optical-based measurement of temperature changes. Clinical applications include studies of patients with eye tumours, epilepsy, early-onset Alzheimers as well as muscular and neuromuscular dystrophies.

Abstract

In 2016, it is estimated that internet IP traffic reached 10^21 bits – within striking distance of the Avogadro number. Given that data sizes are reaching thermodynamic proportions, and that relevant calculations have often to be performed in a distributed manner, it can be expected that phenomena and methods from the statistical physics of many particle systems are relevant.

This talk will examine a couple of examples where phase-transition like phenomena occur, with network performance going from a “good” to a “bad” phase sharply as a function of a relevant global parameter. The examples include the so called network consensus problem, and feature selection in multivariate regression using an L1 norm.

Abstract

In quantitative MRI (QMRI), one seeks to accurately and rapidly localize biomarkers (i.e., measurable tissue properties) using MR data. One key challenge of QMRI is that ‘accurate’ and ‘rapid’ are often competing goals: more physically accurate MR signal models typically depend on more biomarkers, but estimating more markers usually requires longer acquisitions and greater computation. In this talk, I will discuss two recently developed methods to systematically limit these QMRI resource burdens. First, I will describe a method to assemble fast, statistically informative acquisitions that enable min-max optimally precise biomarker estimation. Second, I will describe a machine-learning inspired method to “learn” an extremely fast and scalable biomarker estimator from purely simulated training data. Finally, I will describe our ongoing efforts to apply these methods for fast, accurate myelin water fraction imaging. This talk discusses joint works with Prof. Jeffrey Fessler, Dr. Jon-Fredrik Nielsen, and Prof. Clayton Scott, all at the University of Michigan.

Abstract

Compressed sensing (CS) MRI enables fast imaging. Conventional CS MRI reconstruction algorithms are time-consuming and often lead in undesired over-smoothing or artifacts. Recently, various methods have been proposed to apply deep learning models for more efficient and accurate MRI reconstruction. However, there are still open question on how to ensure realistic and consistent Deep Reconstruction. In this talk, a MRI reconstruction technique using deep learning and generative adversarial network (GAN) is introduced. Evaluated on clinical MRI datasets with both quantitative metrics and radiologists’ ratings, the proposed method demonstrates superior performance compared with conventional iterative reconstruction and Deep Learning models trained with pixel-wise loss. Similar deep learning models can also be applied for PET reconstruction and quantitative MRI.

No abstract was provided for this talk.

Abstract

Evidence that conventional (linear) diffusion encoding is not enough to probe all relevant features of microstructure has accumulated for 20 years. Recent developments have seen the canonical Stejskal-Tanner experiment complemented with techniques that all contribute more specific information about the underlying structure. The lecture will survey several methods based on diffusion encoding with non-conventional gradient waveforms, and what microstructural features that they can resolve.

Abstract

Technology leaders have recently announced the goal of translating thoughts into text directly from brain recordings. Existing work on decoding linguistic meaning from imaging data has been largely limited to concrete nouns, and trained and tested with similar stimuli from a few semantic categories. I will present a new approach for building a brain decoding system, based on a procedure for broadly sampling a semantic space constructed from massive text corpora. By efficiently selecting training stimuli shown to subjects, we ensure the ability to generalize to new meanings from limited imaging data. To validate this approach, we trained the system on imaging data of individual concepts, and showed it can decode imaging data of sentences from a wide variety of concrete and abstract topics in two separate datasets.

Abstract

Brief overview of the current research activities at Siemens Healthcare Technology Center, Medical Imaging Technologies. Located in Princeton, NJ, we are the central research and development lab of Siemens Healthineers. Our team of over 80 research scientists and software engineers specializes in using large collections of data to build artificial intelligence solutions for healthcare. We also work closely with Siemens’ customers in submitting grant proposals to government funding agencies. Our research has resulted in multiple scientific contributions in the fields of medical imaging, modeling, and image-guided therapy and has been incorporated into many clinical products.

Abstract

The advancement of diffusion MR imaging (dMRI) acquisition, post-processing, and clinical diagnostic precision would be accelerated with a cross-laboratory anisotropic diffusion phantom providing paramedic control of shape geometry, packing density and routing. Our group is developing such a phantom matched to histology geometry on a 1 to 1 scale. We have created idealized axons (iAxons) that are textile-based hollow fibers at nanometer scale. They provide controlled geometrical configurations and packing density patterns. The iAxons have a diameter range from 0.2 to 36 microns filled with water covering and exceeding the biological range allowing parametric tests of dMRI precision. We create Standard iAxon Fasciculi (SIF) that contains 950-nanometer internal diameter water filled tubes with a density of a million per mm2. We can create cortical networks such as the eye to LGN of millions of iAxons with precise 50 micron routing positional control. We use non-MRI measurement with Micro CT, light, and electron microscope imaging of iAxons to to quantify dMRI precision. We are creating matched histology and phantoms for pig harvested and human cadaver tissue. We are testing bio-physical models like NODDI or spherical mean techniques (SMT) for packing density pattern and amount of iAxons. We have found the intra-cellular volume fraction correlates with a number of iAxons (r = 0.96). For geometrical configuration, we have tested Constrained Spherical Deconvolution techniques which show promising results to resolve more than 45-degree crossing. We will also present the effect of small/big delta on diffusivities at multiple packing densities of the iAxon bundle. We plan to provide phantoms across laboratories and release public data sets to drive MRI-based quantitative calibration and discovery of improved techniques. We have done cross instrument measurement and found large systematic errors in measurement (35%) across instruments at five sites. We are developing correction methods for clinical scanners. We expect the phantoms to provide a set of ground truth challenges to advance MRI diffusion physics and tractography.

Abstract

Focused Ultrasound is a novel treatment modality that displaces (minimally) invasive surgery with a totally non-invasive approach using a focused beam of ultrasound energy. Depending on the parameters used, the effect at the focal point can be purely mechanical, thermal or a combination thereof. Coupled with real-time feedback of MRI enables to accomplish a spatio-thermal closed-loop procedure, which may lend itself to automation.

In my talk I will review the history of MRgFUS, the current clinical indications it is being used for and some new emerging applications. I will also describe the role of the Focused Ultrasound Foundation, a non-profit aimed at accelerating clinical adoption, in how NYU may benefit from research grants provided by the Foundation.

No abstract was provided for this talk.

No abstract was provided for this talk.

No abstract was provided for this talk.

Abstract

Quantitative Magnetic Resonance Imaging (qMRI) enables the non-invasive measurement of microstructural properties of living tissue, thus providing useful imaging biomarkers with strong clinical potential. In practice, while qMRI is rather popular and successful in the brain, qMRI of the spinal cord is more difficult due its proneness to noise, field inhomogeneity and phyisological artifacts, which hamper the clinical translation of most qMRI methods. In this talk, I will provide an overview of spinal cord qMRI and illustrate its challenges and report on recent developments. In particular, the talk will focus on recent spinal cord qMRI approaches for neuronal morphology and myelin measurement, which hold promise for more accurate diagnosis and prognosis in conditions such as multiple sclerosis.

Abstract

Positron Emission Tomography (PET) has been nowadays established as a molecular imaging modality capable of providing non-invasive, diagnostic and treatment response assessments of the activity of specific molecular processes underlying a spectrum of oncologic, cardiovascular and neurologic diseases. In the first part of this talk we will introduce a clinically adoptable WB dynamic 18F-FDG PET/CT scan protocol coupled with a family of robust direct 4D PET image reconstruction methods to enable for the first time WB multi-parametric PET imaging in humans. The presented framework exploits current state-of-the-art clinical PET systems technologies, such as Time-of-Flight and Resolution modeling, to also support combined WB static and parametric PET imaging from only the standard-of-care scan time window to deliver to clinic additional and highly quantitative information content beyond the standardized uptake value (SUV) metric. Later in the talk, we will also present a novel dual-tracer 18F-FDG:18F-NaF PET/MR imaging framework designed to improve PET attenuation correction in PET/MR studies by robustly segmenting the bone tissues from the 18F-NaF kinetic analysis. Finally, we will demonstrate a clinically adoptable dual-tracer dual-modality imaging protocol for the simultaneous and co-registered anatomical and molecular assessment of both inflammation and micro-calcification, two major molecular mechanisms considered to be associated with atherosclerosis, in human carotid vessel walls.

Abstract

Chronic social stress is an integral part of our busy contemporary lives. Abundant data show that severe chronic psychosocial stress is a risk factor for cardiovascular disease and a predictor of myocardial infarction and stroke. The mechanisms by which stress contributes to the higher cardiovascular event rates are primarily attributed to secondary effects on behavior, including smoking or food intake. How stress’ effect on the brain can directly impact cardiovascular disease is uncharted territory.

Preclinical data describe a direct causal link between social stress, neural signals, and atherosclerosis, the lipid-driven chronic inflammatory disease that is the underlying cause of myocardial infarction and stroke. The key connecting component is the macrophage, a large phagocytic leukocyte that originates in the bone marrow and accumulates in atherosclerotic lesions. Informed by abundant published and unpublished data, we hypothesize that chronic variable stress aggravates cardiovascular disease by interfering with macrophage dynamics.

Specifically, we wish to (1) understand how stress biologically affects macrophage dynamics in atherosclerosis; (2) develop technology that monitors macrophage dynamics non-invasively; and (3) elucidate the mechanism by which post-traumatic stress disorder (PTSD) leads to atherosclerosis.

This work is based on technological developments (such as motion compensation and fast imaging) in biomedical imaging and systems imaging using PET/MR and using novel targeted approaches (such as molecular imaging and nanomedicine) to study and treatment of inflammation in preclinical and clinical studies. I will describe our overarching and long-term goal is to collectively institute a sound scientific foundation for the biomedical and clinical community as how the link between stress and cardiovascular disease can be best approached and integrated in patient care.

References

1. “Systems biology and noninvasive imaging of atherosclerosis.” Arteriosclerosis, Thrombosis, and Vascular Biology. 2016; 36:e1-e8. doi 10.1161/ATVBAHA.115.306350
2. “Relation between resting amygdalar activity and cardiovascular events: a longitudinal and cohort study.” Lancet. 2017; 389: 834-845. doi 10.1016/S0140-6736(16)31714-7
3. “Imaging systemic inflammatory networks in ischemic heart disease.” Journal of the American College of Radiology. 2015; 65: 1583-1591. doi 10.1016/j.jacc.2015.02.034

Abstract

I will introduce a novel analytical method based on high order statistics and nonlinear coherences that enables to obtain directed pathways of signal progression among coupled time-series. Assuming a consistent phase relationship between neuronal and MRI signals, the method is demonstrated in the human brain with resting-state fMRI data. Pathways in the visual and the motor systems were characterized by appealing to a hierarchy based upon temporal or phase differences. I will describe the different organizations of the ventral and dorsal visual systems, the frequency dependency of the thalamo-cortical connections and how it changes with age.

Abstract

In biomedical preclinical research we have dreamt about a magnifying glass that would allow us to e.g. see neurotransmitters in action, that would simultaneously quantify mechanical function, perfusion and various local cell functions in the heart, and in cancer research for (simultaneous) detailed dynamic distributions of pharmaceuticals and indicators of tumor response. In recent years many groups have been involved in the development of pinhole imaging SPECT systems for imaging rodents.

At MILabs and TU-Delft, a ultra-high resolution Single Photon Emission Computed Tomography (U-SPECT-CT) has been developed that can quantify tracers in 0.15 mm structures, enable low dose imaging (sub-MBq), or visualize extremely fast tracer dynamics (sub-second time frames) by developing highly advanced imaging geometries, novel image acquisition and reconstruction. An option on this system to perform 0.6 mm Positron Emission Tomography (PET) simultaneous with 0.4mm SPECT (VECTorTM) was developed. It also enables for the first time ultra-high energy SPECT (up to 1MeV) and imaging of sub-mm resolution of theranostic isotopes to real time monitor and steer cancer therapy.

In this presentation, scientific results recorded by worldwide users of a full integrated platform combining SPECT, PET, ultra-fast and ultra-high resolution CT, Cherenkov, bioluminescence and fluorescence imaging will be discussed. Finally the results of translating U-SPECT technology into a clinical device (G-SPECT: WMIS Innovation of the Year 2015), an Ultra-fast, Ultra-high resolution (< 3 mm resolution) will be presented.

Abstract

Diffusion MRI is an excellent method for detecting microscopic changes of the living human brain, but often fails in assigning the observed changes to a specific structural property such as cell density, size, shape, or orientation. When attempting to solve this problem, we have decided to simply ignore the entire field of diffusion MRI, and instead translate data acquisition and processing schemes from multidimensional solid-state NMR spectroscopy. Key elements of our approach are q-vector trajectories and correlations between isotropic and directional diffusion encoding. To emphasize the origin of the new methods, we have selected the name “Multidimensional diffusion MRI.” Assuming that the water molecules within a voxel can be divided into groups exhibiting approximately Gaussian anisotropic diffusion, the composition of the voxel can be reported as a diffusion tensor distribution where each component of the distribution is directly related to a specific tissue environment. Our new methods yield estimates of the complete diffusion tensor distribution or well-defined statistical properties thereof, such as the mean and variance of isotropic diffusivities, mean-square anisotropy, and orientational order parameter, which are straight-forwardly related to cell densities, shapes, and orientations. This presentation will give an overview of the multidimensional diffusion MRI methods, including basic physical principles, pulse sequences, data processing, and examples of applications in healthy and diseased brain.

Abstract

Disconnections of cortical networks may underlie various cognitive deficits that take severe clinical tolls on patients with schizophrenia. Historically, the neuropsychopharmacology of cognitive deficits is mostly conceptualized and studied in terms of neurons, neurotransmitters and synaptic receptors. We hypothesized that the dynamics of the extended lifetime development trajectory of the brain’s white matter, and the consistency of connectivity deficits in schizophrenia, posit white matter as the key loci responsible for these cognitive deficits. Using novel diffusion weighted imaging (DWI) techniques and a milestone development of identifying key white matter tracks most relevant to schizophrenia, we are now able to show that specific white matter pathways are responsible for shared vs. unique contributions to some of the key cognitive deficits in schizophrenia.

No abstract was provided for this talk.

Abstract

Hyperpolarized contrast media prepared via dissolution dynamic nuclear polarization or parahydrogen-induced polarization provide tremendous in vivo signal enhancements for dilute tracer molecules labeled with nuclei such as 13C or 15N. These signal enhancements provide a tool for monitoring tissue function and metabolism, particularly in cancer and cardiac disease. In pre-clinical models of lung, prostate and breast cancer, hyperpolarized pyruvate can detect tumor response to therapy within hours of the onset of treatment, potentially providing a new tool for personalized medicine by rapidly identifying the best therapy for each patient. Clinical translation of hyperpolarized imaging will require new approaches to MR spectroscopic imaging. Spectroscopically selective balanced steady-state techniques offer improved sensitivity and speed relative to conventional echo-planar spectroscopic methods that can be leveraged for imaging in patients.

Abstract

Nuclear Magnetic Resonance (NMR) and Magnetic Resonance Imaging (MRI) is common in medical research and widely used for medical diagnosis. However, NMR and MRI systems are expensive to install and cause substantial maintenance costs. Its use is often restricted to radiology centres or hospitals in larger cities. Here we report on recent research which may help to turn inexpensive, mobile low field NMR systems into medical devices. One challenge in low field NMR is the magnetic field inhomogeneity. It introduces a distribution of Larmor frequencies and magnetic field gradients. However, field distributions can be determined (see Fig. 1 left) and may be corrected for, thus enabling these magnet systems for the use in NMR diffusometry [1]. Another challenge is the reduced signal-to-noise ratio at lower magnetic fields. Therefore, conventional imaging approaches may not be feasible. We have shown that the sample averaged fractional anisotropy (FA) can be determined without the use of imaging [2]. However, if imaging is needed, the amount of acquired data may be reduced dramatically using prior knowledge [3]. More recently we have also demonstrated that single sided NMR systems such as the NMR MOUSE [4] can be used (see Fig. 1 right) for the determination of the total volume-to-bone volume ratio, a parameter linked to the micro structure of bones and therefore to the risk factor for osteoporosis [5]. We anticipate the use of mobile low field NMR systems as diagnosis and screening tools, affordable for general practitioners as well as mobile point-of-care medical devices on the bedside, in ambulances, operational theatres and ICU’s.

Abstract

Immunotherapies are becoming increasingly important for improved treatment of a variety of cancer types. However, the development of these novel therapies has outstripped our understanding of underlying mechanisms and how best to apply them. It is therefore crucial that we use tools such as MRI, and other molecular imaging techniques, to evaluate immunotherapies in both the clinic and in preclinical studies, and develop new probes and biomarkers to increase their efficacy. Studies have shown high degrees of variability in individual response to cancer, increasing the necessity of a more personalized approach, and optimized methods for combinations of multiple therapies are not well understood.

This talk will touch on a number of molecular imaging methods used to study immunotherapy response, including the use of MRI cell tracking for monitoring both adoptive cell immunotherapies, and immune cell population migration in response to other immunotherapy subtypes. Other techniques to be discussed include use of PET (using standard FDG imaging, and novel probes specifically developed for immunotherapies) and PET/MRI multimodal imaging for monitoring both anatomical and functional changes with MRI (using DCE, T1/T2-weighted imaging, etc.)

Abstract

Although much clinical progress has been made in harnessing the immune system to recognize and target cancer, there is still a significant lack of an understanding of how tumors evade immune recognition and the mechanisms that drive tumor resistance to both T cell and checkpoint blockade immunotherapy. Our objective is to understand how tumor-mediated signaling through inhibitory receptors, including PD-1, combine to affect the process of T cell recognition of tumor antigen and activation signaling, with the goal of understanding the basis of resistance to PD-1 blockade and the potential identification of new molecular targets to enable T cells to overcome dysfunction mediated by multiple inhibitory receptors. Potential combinatorial immunotherapeutic strategies of combining T-cell therapy strategies with checkpoint blockade will also be discussed.

No abstract was provided for this talk.

Abstract

Functional connectomics using resting state fMRI (rsfMRI) is a rapidly expanding task-free approach, which has the potential to complement task-based fMRI for presurgical mapping in patients with neurological disease. However, high sensitivity to head movement and physiological noise, the low frequency range of rsfMRI (< 0.2 Hz), and considerable spatial-temporal non-stationarity compromise mapping of resting state networks (RSNs).

Recently, several studies using volumetric and multi-band high-speed fMRI have reported resting state connectivity at much higher frequencies (up to 5 Hz). This approach has the potential for addressing principal limitations of mapping low frequency resting state connectivity, such as high sensitivity to signal drifts and long scan time necessary for separating major RSNs in single subjects. However, other studies have been more cautious regarding the possible signal sources or were unable to replicate the findings.

The first part of this talk will discuss recently developed ultra-high speed fMRI and confound-tolerant seed-based resting-state fMRI analysis methodology that enabled sensitive detection of high frequency signal fluctuations in auditory cortex and default mode network. Experimental findings were validated by analyzing non-physiological signal sources using simulations of auto-correlations in Rician image noise. The second part of the talk will describe initial experience using high-speed resting state fMRI for presurgical mapping in patients with brain tumors, arteriovenous malformation and epilepsy, and integration of this approach with multi-modal diagnostic imaging.

Abstract

The focus of the Seiberlich Lab is to develop MR imaging techniques to capture structural and functional information from moving organs, specifically in the abdomen and heart. This lecture will cover the recent developments using Magnetic Resonance Fingerprinting for quantitative tissue property mapping of the myocardium. Additionally, work on real-time cardiac and abdominal imaging using non-Cartesian parallel imaging techniques in conjunction with Gadgetron will be discussed.

Abstract

This talk will cover some novel radio frequency pulse and pulse sequence designs to overcome some of the main limitations of operating at high magnetic fields, thereby enabling high-resolution whole-brain anatomical, spectroscopic and diffusion imaging. Translation of these techniques to improve diagnosis, treatment and surgical planning for a range of neurological diseases and disorders will be discussed. Specific clinical applications that will be covered include: Improved localization of epileptogenic foci; imaging to reveal the neurobiology of depression; and development of imaging methods to better guide neurosurgical resection of brain tumors.

Abstract

Functional MRI (fMRI) provides the most versatile imaging platform for mapping the brain activities in vivo. More recently, resting-state fMRI (rsfMRI) has emerged as a valuable tool for mapping large-scale and long-range brain networks. However, both methods only reflect the gross outcome of the complex and cascaded activities of various cell types and networks, posing limitations when dissecting brain networks. Optogenetics technology can provide spatiotemporally precise modulation of genetically defined neuronal populations in vivo. Here we combine fMRI with optogenetic perturbations and electrophysiology to capture and analyze whole brain activity and long-range circuits with much improved specificity and causality. We deploy this capability to interrogate the spatiotemporal response properties of two distinct long-range networks, namely, thalamo-cortical and hippocampal-cortical networks. We examine the functional effects of low-frequency optogenetic stimulation within these two networks on brain responses to external sensory stimuli, on brain-wide functional connectivity at resting-state, and on cognitive behaviors. Our findings reveal that low frequency activity governs large-scale, brain-wide connectivity and interactions through long-range excitatory projections to coordinate the functional integration of remote brain regions. This low frequency phenomenon contributes to the neural basis of long-range functional connectivity as measured by rsfMRI. I this talk, I will also briefly introduce our recent diffusion MRI works in brain and MSK, including diffusion MR spectroscopy.

Abstract

The aim of this presentation is to evaluate the use of PET/CT with 18F-FDG for an assessment of the testicular function and to optimise and standardise the acquisition protocol and the testicular volume analysis in order to do that. By ways of introduction there will be a literature overview where we establish why the 18F-FDG uptake is correlated with the spermatogenesis. There will follow an overview of the public health problem of male infertility where the different possible clinical applications for testicular functional imaging with PET/CT will be addressed.

In the second part of the talk we’ll discuss the correlation between 18F-FDG uptake in terms of intensity and volume of uptake and the testicular function via the parameters of the sperm analysis based on the published article of our group.

The third part of the presentation will be on the subject of some of the technical issues where the focus will be on the standardisation of the acquisition protocol for this specific indication. In the last part of the presentation, we’ll address the overall important subject, and even more so in this andrological context, of the radioprotection related issues of a PET/CT with 18F-FDG.

Finally, there’ll be an overview of some of the issues still to be addressed and the future perspectives.

No abstract was provided for this talk.

Abstract

Small-angle X-ray scattering (SAXS) occurs when part of the X-ray beam that probes a sample is scattered at small angles, due to differences in electron density distributions within the sample. Moreover, it gives a particularly strong signal in the presence of ordered and periodic systems. The recently developed small-angle X-ray scattering tensor tomography (SAXSTT) takes X-ray tomography a step further: it uses two sample rotation axes and an iterative reconstruction algorithm to tomographically reconstruct local tissue anisotropy. The method was demonstrated for reconstructing the orientation of mineralized collagen fibrils in bone trabeculae of human vertebrae, based on the 65-nm D-spacing of collagen. Similar experiments have also very recently been performed in mouse brain, taking advantage of the ~17.5 nm spacing of the myelin sheath. Providing directly structural information, SASTT was used to derive neuronal fiber directionality and myelin content in a quantitative way. The results are being compared with MRI methods such as diffusion-weighted imaging and magnetization transfer, as well as with 2D and 3D histology (CLARITY).

Abstract

MRI has historically performed poorly when imaging ultra-fast relaxing tissues such as bone, lung tissue, and tendons as well as components of other connective tissues including cartilage and myelin. Specialized pulse sequences such as ultrashort echo time (UTE) and zero echo time (ZTE) MRI offer the potential to image these tissues, and have several promising new applications that will broaden the capability of MRI. These include

1. PET/MRI – Hybrid PET/MRI systems require attenuation correction for accurate PET reconstructions, which should include estimates of bone density. This talk will present work using ZTE MRI for generating pseudo/synthetic CT images that include bone density estimates in the head and pelvis. Most recently, we have applied Deep Learning for this synthetic image generation task.
2. Lung Imaging – Pulmonary MRI has been very challenging due to the short T2* of lung parenchyma and motion, but is important for assessing pulmonary nodules in PET/MRI and for dose reduction in pediatric populations. This talk will present an approach using UTE MRI, where self-navigation is achieved through a local low-rank reconstruction of dynamic 3D image navigators and motion-corrected images are reconstructed similarly to XD-GRASP.
3. Myelin Imaging – Myelin facilitates crucial long-range communication across the brain, and is typically assessed in MRI through diffusion-weighting, magnetization transfer, and myelin water imaging. It has been shown through ex vivo studies that there are fast relaxing components in myelin associated with protons in the myelin phospholipid membranes, which are not captured in these conventional approaches. This talk will present in vivo characterizations of the ultrashort-T2* components in the brain that have the potential to provide a more direct measurement of myelination.

Abstract

Diffusion MRI is sensitive to the length scale of tens of microns, which coincides to the scale of microstructure in the human brain tissue. By changing the diffusion time or diffusion gradient pulse width, we can probe the brain micro-geometry via time-dependent diffusion measurements. To increase the sensitivity to the microstructure, STEAM sequence is often used for extending the range of diffusion time. However, water exchange between myelin water and intra-/extra-axonal water may bias the parameter estimations. This talk will focus on the time dependence either along or perpendicular to white matter axons and corresponding micro-geometries, and the correction for the time dependence measured by STEAM.

Abstract

Diffusion of water molecules is directly influenced by the mountainous landscape of biological tissue. By modeling time-dependent diffusion, it is possible to reverse engineer various features of this landscape. The proposed model will depend on the underlying tissue microstructure, which poses an additional challenge of model selection. This talk will focus on the efforts of modeling diffusion time-dependence in the prostate, which embodies modeling problems, which concern partial volume and model selection, as well as imaging problems, such as geometric distortion and low SNR.

Abstract

PET radiochemistry can be a great resource for imaging, treatment and point-of-care response/monitoring in Cancer and Cardiovascular disease. A novel small molecule targeting the cyclin-dependent kinases CDK4/6 and a series of radiolabeled nanobodies and peptides for atherosclerotic plaques imaging will be presented. The seminar will also focus on the fundamentals of Radiochemistry and how the newly established CAI²R Radiochemistry facility will operate.

Abstract

Over the past decade, the number of microstructural imaging papers has been doubling every 2.7 years. With such growth, it is becoming increasingly difficult to perform a fair comparison between competing approaches. Some simplify the tissue modelling and overlook physiological constraints. Others overparametrize the models and amplify the noise. The outcome is a field of research with great promise, but few checks and balances.

This lecture will introduce several frameworks for interpreting, validating and communicating microstructural imaging data. Examples will be drawn from myelin imaging in the brain, focusing on the challenges associated with mapping the longitudinal relaxation time (T1), the axon caliber, and the myelin thickness (g-ratio). The last part of the lecture will put these frameworks in a broader science communication context, discussing how medical imaging researchers can set new standards for reviewing, publishing, and publicizing their findings.

Abstract

Recently, Chemical Exchange Saturation Transfer (CEST) has emerged as an attractive MRI contrast mechanism. In CEST, the MRI contrast is generated by transferring the magnetic labeled water-exchanging protons (OH, NH, or NH2) from a CEST agent to its surrounding water molecules. Many natural biological compounds naturally carry exchangeable protons, making them possibly detected by CEST MRI directly in a “label-free” manner. In our studies, we utilized this unique feature to directly detect drugs and drugs carriers, which makes MRI-guided drug delivery possible even without any chemical labeling, a strategy we called “natural labeling”. This new MRI labeling strategy in principle can be tailored to many existing drug delivery systems, and portends a new path to safe, rapid clinical translation of image-guided drug delivery.

Abstract

A high-throughput imaging pipeline is presented to characterize the heterogeneity in longitudinal disease progression in mouse models of human brain cancer and to test the efficacy of novel anti-cancer therapeutics in accurate mouse models of sporadic human cancer.

Abstract

Huntington disease (HD) is a dominantly inherited and progressive neurodegenerative disorder, caused by a CAG trinucleotide repeat expansion (≥ 39 repeats) within the HD gene. The median age at which HD occurs, is around 40 years and the disease progresses over time and is invariably fatal 15–20 years after the onset of the first symptoms. The major goals of current HD research are to improve early detection and monitor pathological changes in individuals both at early and advanced stages of the disease. Animal models of inherited neurological diseases provide an opportunity to test potential biomarkers of disease onset and progression and evaluate treatments for translation to clinical care. Using several diffusion MR techniques we studied two different rat models of HD. In this talk I will present data that shows that diffusion MRI is a sensitive and quantitative method to detect HD related neurodegenerative changes, at both microstructural and subcellular levels.

Abstract

There is an ongoing need for non-invasive identification of macromolecular changes in tissue. An important application is to the diagnosis of early osteoarthritis (OA). Our work in this area combines basic science studies in magnetic resonance imaging and relaxometry with emerging methodologies that carry translational potential. We will discuss multi-exponential transverse relaxation analysis as a means to identify underlying macromolecular compartments in normal and degraded cartilage, as well as important extensions of this work, based on higher dimensional relaxometry and compressed sensing. We will describe the mathematical setting for this work as a linear inverse problem. Further work in human subjects requires introduction of a nonlinear model system. We will describe several approaches to these problems and indicate the potential for improved detection of early cartilage degradation. Our methods are also applicable to directly mapping myelin in the brain, and we have obtained results showing myelination pattern alterations with age and in cognitive impairment. All of these studies are centered around the clinical goal of improving the ability of magnetic resonance methods to diagnose pathology and to monitor disease progression.

Abstract

Separately, PET and MRI have longstanding roles in diagnosis, prognosis and monitoring of breast cancer. Since the recent advent of the simultaneous PET/MRI platform, intense research has taken place to identify unrealized applications of their fusion. Initial work around the world has included study of a range of practical advantages (feasibility, efficiency, patient retention, physiologic simultaneity, co-registration), but always with an eye toward future ‘breakout’ applications beyond those with separate scans. I will describe efforts within our breast cancer research group that pursue both practical and fundamental benefits with the unique capabilities in our research center. Whole body evaluation of metastatic breast cancer patients is nearly equivalently done with PET/MRI as with PET/CT but with half the radiation dose. Dynamic contrast enhanced (DCE) MRI and intra-voxel incoherent motion (IVIM) MRI offer a range of quantitative characterizations of the primary tumor microenvironment (cellularity, vascular volume, vascular permeability) that when combined with fluorodeoxyglucose (FDG) uptake provide a comprehensive characterization of malignancy in one imaging session. Simultaneity also supports detailed intralesional correlations that may increase classification ability even further. Finally, future planned work with more specific microenvironment tracers and integrated PET and MRI pharmacokinetic modeling holds remarkable potential for oncologic management with noninvasive imaging.

Abstract

Analyzing moving organs such as the heart in MRI is a challenging task. In clinical routine images are acquired over several heartbeats to reconstruct all contraction phases of one representative cardiac cycle using ECG-gating and breath-hold techniques.

Real-time MRI techniques allow the acquisition of serial 2D images with a temporal resolution of up to 20 ms under free breathing. The analysis of real-time MRI sequences, however, requires adapted segmentation techniques as well as an advanced analysis providing information about temporal evolution of parameters during individual heart cycles in amulti cycle analysis workflow.

No abstract was provided for this talk.

No abstract was provided for this talk.

Abstract

Many problems in signal and image processing, machine learning, and estimation require optimization of convex cost functions. For convex cost functions with Lipschitz continuous gradients, Nesterov’s fast gradient method decreases the cost function at least as fast as the square of the number of iterations, a rate order that is optimal. This talk describes a new first-order optimization method called the optimized gradient method (OGM) that converges twice as fast as Nesterov’s famous method yet has a remarkably similar simple implementation. Interestingly, Drori recently showed that OGM has optimal complexity among first-order methods. I will discuss other recent extensions and show examples in machine learning and X-ray computed tomography (CT). Combining OGM with ordered subsets provides particularly fast reconstruction for CT. This work is joint with Donghwan Kim.

Abstract

In recent years ocular imaging has become the cornerstone for clinical diagnosis and disease monitoring as well as a primary research tool in ophthalmology. In this presentation we will discuss state-of-the-art, in-vivo, high resolution ocular imaging technologies. We will present the utility and challenges of the technologies to advance clinical practice and research of glaucoma—a leading cause of blindness and visual morbidity.

Abstract

Advances in neuroimaging provide us with greater insight to brain injury than ever before. Magnetic resonance spectroscopy is a non-invasive method of measuring brain chemistry altered by bran injury using readily available MRI, thus providing a virtual biopsy of concussions. A review of the technology and current findings from the acute to chronic stages of mild brain injury, including the rising concern of chronic traumatic encephalopathy in sports and military-related repetitive brain trauma, will be discussed.