Categories
Resources

Vision Transformer-Based Network for MR Electrical Properties Tomography

A neural network for reconstructing electrical properties of materials and biological tissue from MRI data.

Share on LinkedIn

We are sharing a 3D vision transformer-based neural network architecture developed for fast and accurate reconstruction of electrical properties of objects and tissues from magnetic resonance imaging (MRI) data.

Electric conductivity and relative permittivity provide valuable information about interactions between electromagnetic waves and biological tissue. Magnetic resonance electrical properties tomography (MR-EPT) is used to estimate conductivity and permittivity in tissues based on MR measurements and has the potential to enable promising applications in biomedical research, including: 

  • estimation of local specific absorption rates, of particular importance in ultra-high-field MRI
  • advancement of radiofrequency or thermal-based treatments
  • estimation of novel noninvasive biomarkers for monitoring health and evaluating response to therapy

Estimation of electrical properties is computationally challenging, and traditional methods tend to be slow and imprecise. The new approach shared here relies on deep learning to produce fast and accurate reconstructions.

Based on MRI data, including magnetic field maps and an edge mask that highlights tissue boundaries, the neural network predicts 3D conductivity and permittivity maps. The model was validated through simulations, a phantom experiment, and an in vivo human study. Results demonstrate that the approach produces more accurate reconstructions than do traditional methods. 

A schematic diagram showing inputs and outputs of the vision transformer-based neural network for MT-EPT.
Graphical abstract of the neural network for 3D MR Electrical Properties Tomography. The network takes as inputs the magnitude of the magnetic transmit field, the transceive phase map, and an edge mask generated from MRI images using Canny filtering, and is trained to predict electric conductivity and relative permittivity maps of the subject.

We hope that sharing this resource will facilitate reproducibility and comparison studies and accelerate the development of clinically viable non-invasive electrical property mapping techniques. Investigators are encouraged to use this open-source tool to explore new applications of EPT in diagnostics, treatment planning, and biomedical research.

Related Publication

Giannakopoulos II, Carluccio G, Keerthivasan MB, Koerzdoerfer G, Lakshmanan K, De Moura HL, Cruz Serrallés JE, Lattanzi R.
MR Electrical Properties Mapping Using Vision Transformers and Canny Edge Detectors.
Magn Res Med. 2024 Oct 16. doi: 10.1002/mrm.30338

Get the Code

Contact

Questions about this resource may be directed to Ilias I. Giannakopoulos, PhD, at ilias.giannakopoulos@nyulangone.org.

Related Story