Development of an MR-Gleason Score for Prostate Cancer using Advanced MRI Metrics

Development of an MR-Gleason Score for Prostate Cancer using Advanced MRI Metrics

Project Summary

The lack of a means of reliably separating cases of prostate cancer (PCa) with a low and high Gleason score pre-operatively has resulted in over-treatment of many patients via radical prostatectomy (RP). The goal of this collaborative project is to develop a comprehensive imaging-based Gleason score for PCa, combining a multitude of metrics including advanced diffusional- and T2-based coefficients, perfusion, tissue sodium concentration, and 18F-fluorocholine (FCH) uptake, that can reliably distinguish low and high grade tumors. The imaging-based Gleason score will be validated as a more accurate predictor of RP-based Gleason score than is biopsy-based Gleason score and as a reliable predictor of surveillance failure for men who select active surveillance (AS) for their cancer. This project takes advantage of an existing team of radiologists, urologists, uropathologists, physicists, and industry collaborators at NYULMC with a strong publication track record. The innovative technologies encompassed by the BTRC will be critical for advancement of this project, which in turn will provide a spectrum of unique opportunities for testing and applying BTRC technologies.

Aim 1: To determine the combination of advanced imaging parameters that best correlates with Gleason score. We will initially compare mean diffusivity, mean kurtosis, and bi-exponential T2 parameters with Gleason score from whole-mount RP and subsequently incorporate additional metrics including perfusion, sodium concentration, and FCH uptake.

Aim 2: To validate the imaging-based Gleason score developed in Aim 1 in men undergoing RP. The predicted Gleason score will be determined in a large prospective patient cohort before RP and compared with actual RP Gleason score and biopsy Gleason score.

Hypothesis: The imaging-based Gleason score better predicts RP Gleason score than biopsy Gleason score.

Aim 3: To evaluate use of the imaging-based Gleason score developed in Aim 1 in men undergoing AS. Hypothesis: High imaging-based Gleason scores will be a significant predictor of AS failure, as determined by standard clinical criteria incorporating rapid PSA doubling time and tumor upgrade on repeat biopsy. Specific interactions with the BTRC are spelled out at the end of this project summary. Impact: This project will provide a powerful tool for establishing prognosis and guiding treatment selection for PCa, thereby improving clinical outcomes, lowering over-treatment, and reducing overall health-care costs

Significance

PCa is the most commonly diagnosed cancer and second most common cause of cancer death in US men. Despite its frequency, the risk of dying from PCa is very low given that a large majority of tumors are clinically insignificant. RP is the standard curative therapy for PCa in the US and is over-treatment for many patients given its substantial negative impact on quality-of-life, including sexual and urinary side effects. A means of reliably distinguishing indolent and aggressive cancers would drastically reduce overtreatment by enabling appropriate selection of patients for radical therapy. Currently, the histological PCa Gleason score shows the greatest potential for achieving this purpose, having been validated in multiple studies as a significant predictor of mortality. However, there is no reliable means of identifying the Gleason score pre-operatively at the time that treatment decisions are made. The Gleason score determined by transrectal biopsy is an unreliable predictor of the score at RP due to substantial undersampling. This inability to confidently identify low-grade tumors by biopsy is a primary reason for current substantial over-treatment.

The availability of advanced, quantitative MRI now offers the opportunity to characterize tumor aggressiveness. Of note, over a dozen studies have identified an inverse correlation between apparent diffusion coefficient (ADC) values and Gleason score, leading to ADC receiving the greatest attention as a noninvasive biomarker of PCa aggressiveness. However, correlation between ADC and tumor grade is insufficient to predict Gleason scores in individual patients given substantial overlap in values between low and high grade tumors in the literature. While these parameters largely reflect the average amount of fluid within the tissue, i.e., the glandular fraction, the Gleason score is influenced by an additional range of factors that includes cellularity, cell size, uniformity of cell size and shape, size of glandular spaces, and presence of intermixed histologic components such as benign stroma. Therefore, a new diagnostic tool is needed that better reflects the histologic complexity of PCa in prediction of the Gleason score

We hypothesize that imaging metrics that provide detailed information on tissue microstructure will provide a more complete representation of Gleason score. Specifically, we propose using diffusion kurtosis measurement to assess glandular size and cellularity, bi-exponential T2 measurement to provide a more accurate measure of glandular fraction, and additional metrics as noted below. These parameters will contribute to an imaging-based Gleason score. Given the ability to completely image the prostate, this score will be more reliable than biopsy in identifying significant cancers that warrant radical therapy.

Diffusional kurtosis metrics
ADC calculations assume that diffusion is “Gaussian”, i.e. unaffected by the presence of impediments to diffusion such as cell walls. Diffusion kurtosis imaging (DKI), introduced at NYU in 2005, takes the non- Gaussian diffusion found in tissue explicitly into account. This has two advantages. First, DKI provides a more accurate estimate of diffusivity, D, than ADC. Second, and more important, DKI provides an estimate of the diffusional kurtosis, K, that is a measure of the microstructural complexity of tissue. Compelling preliminary data from our group shows that the increase in tissue complexity seen with increasing Gleason score is reflected in an increase in K. Furthermore, our preliminary data in 48 patients shows significantly greater AUC for K than ADC in differentiating low and high grade tumors (0.70 vs. 0.62, p=0.010).

Bi-exponential T2 modeling
The prostate consists of fluid filled glands surrounded by cellular epithelium and stroma. The diffusion distance of water in 100ms is about 20μm, compared to an average gland diameter of about 200μm. Exchange between the two compartments is thus relatively slow so that each will display its own characteristic T2. Monoexponential T2 measurements will yield only an average of these two T2s weighted by the relative size of each of the two compartments. Bi-exponential T2 measurements allow separate estimation of the two compartments and also an estimate of the relative size of the compartments. In our preliminary T2 modeling data, biexponentials gave a better fit in 96% of 25 PCa patients. We hypothesize that high grade tumors will demonstrate lower long T2 fractions, corresponding to glandular volume, than low grade tumors.

Innovation

Trans-rectal prostate biopsies are invasive, painful, unreliable, and expensive. Yet over one million are performed in the US each year. The unreliability of the procedure leads to aggressive surgical treatment at the least suggestion of PCa, with consequent sexual and urinary complications in a large number of men who have no significant disease. The ultimate aim of this research is to comlement biopsy with an imaging examination that will be non-invasive, less costly, and, most importantly, more reliable. In doing so, we propose replacing current MRI metrics, such as ADC, that are dependent on multiple different factors that contribute to Gleason score in a non-trivial manner with metrics such as diffusion kurtosis that can be directly related to specific cellular features. DKI has not been applied for grading of prostate cancer by other groups. Rather than relying on a single metric as has been the norm, we will derive a score based on a broad combination of metrics, just as the Gleason score is derived from multiple histological features. This rarely used approach will allow incorporation of other imaging-derived metrics such as sodium concentration, FCH uptake, and perfusion, in facilitation with development of the BTRC technologies

Approach

Specific Aim 1 (Creation of Imaging-Based Gleason score). Our preliminary data provides compelling evidence to support the role of DKI and bi-exponential T2 modeling in PCa assessment. In this aim, we will compare these and additional parameters (high temporal-resolution perfusion based on novel MR acquisition schemes, sodium concentration, and FCH uptake) with the Gleason score determined from whole-mount radical RP to develop a model that combines these imaging metrics into a single predictor of high grade (Gleason score greater than 6) tumor. During years 1 and 2, 80 patients (40 per year) with biopsy-proven PCa who have elected to undergo RP will undergo imaging including diffusion kurtosis, bi-exponential T2, perfusion, sodium concentration, and FCH PET. Following RP, whole-mount processing of the specimen will be performed and a uropathologist will map the location of all significant tumor foci (high Gleason score or diameter of at least 1 cm). In-house software for computer-assisted registration of MRI and pathologic images of the prostate will be used to map these tumors to the pre-operative imaging. Logistic regression and ROC analyses will be used to develop a model based on the imaging parameters to predict the pathologic Gleason score of dominant tumors. This sample size will have at least 80% power, at the 5% significance level, to identify imaging-based predictors of Gleason score with an odds ratio of at least 2.0 based on inclusion of at least one significant tumor per patient.

Specific Aim 2 (Validation of Imaging-Based Gleason score developed in Aim 1 in men undergoing RP) Biopsy misses high-grade tumors in approximately 40% of cases, leading to uncertainty that contributes to over-treatment of patients with low-grade cancer. In this aim, we will demonstrate that the imaging-based Gleason score created in aim 1 is more accurate than the biopsy Gleason score in predicting the presence of high grade tumor at RP and thus provides a better means of treatment selection for newly diagnosed cases. The imaging-based Gleason score will be measured in a large prospective patient cohort before RP and compared with RP Gleason score and biopsy Gleason score. We hypothesize that the maximal Gleason score in the RP specimen will be better predicted by a patient’s maximal imaging-based Gleason score than by maximal Gleason score in any biopsy core. This aim will occur during years 3 and 4, comprising a different patient cohort than used to create the imaging-based Gleason score in Aim 1. 80 patients (40 per year) with biopsy-proven PCa who have selected RP will be included. This sample size was chosen to achieve 80% power, at the 5% significance level, to show that the imaging-based Gleason score is better than biopsy at predicting RP Gleason score when the true prediction accuracy of the imagingbased Gleason score is at least 15 percentage points higher than the accuracy of biopsy.

Specific Aim 3 (Validation of Imaging-Based Gleason score developed in Aim 1 for surveillance) Active surveillance is increasingly selected for managing low-risk PCa, yet a sizeable fraction of patients quickly abandon this approach. In this aim, we will demonstrate that the imaging-based Gleason score predicts likelihood of progression during surveillance. 35 patients per year will be recruited for this aim and undergo initial imaging in years 1-3. These subjects will then undergo surveillance for the remainder of the 5-year study duration. Standard clinical criteria incorporating rapid PSA doubling time and tumor upgrade on repeat biopsy will be used as determinants of potential surveillance failure. The imagingbased Gleason score will be calculated as developed in aim 1 and compared with clinical surveillance outcomes. If ≤15% of the subjects are lost to follow-up, then this sample size will provide 80% power at the 5% significance level to accurately predict the correct clinical outcome in over 50% of patients.

Key Personnel: 
Andrew B. Rosenkrantz

Sponsors

Latest Updates

Philanthropic Support

We gratefully acknowledge generous support for radiology research at NYU Langone Medical Center from:
 
• The Big George Foundation
• Raymond and Beverly Sackler
• Bernard and Irene Schwartz

Go to top