Related Experiment Video
Updated: Jun 19, 2025
08:33
Author Spotlight: Methodologies and Advancements of Chronic Pain Management Research
Published on: January 5, 2024
1.1K
Unrolled Optimization via Physics-Assisted Convolutional Neural Network for MR-Based Electrical Properties
Sabrina Zumbo1, Stefano Mandija2,3, Ettore F Meliado3
1Department DIIESUniversità Mediterranea di Reggio Calabria 89124 Reggio Calabria Italy.
IEEE Open Journal of Engineering in Medicine and Biology
|July 25, 2024
Summary
This study introduces a novel, AI-driven method for Magnetic Resonance imaging based Electrical Properties Tomography (MR-EPT) to map tissue electrical properties. The approach achieves high-quality 2D reconstructions comparable to existing methods but with significantly reduced computation time.
Area of Science:
- Biomedical Imaging
- Medical Physics
- Computational Electromagnetics
Background:
- Magnetic Resonance imaging based Electrical Properties Tomography (MR-EPT) is a non-invasive method for assessing biological tissue electrical properties (EPs).
- Accurate EPs mapping is crucial for various MRI applications, including diagnostics and treatment planning.
- Current MR-EPT reconstruction methods can be computationally intensive, limiting their clinical applicability.
Purpose of the Study:
- To develop and evaluate an unrolled, physics-assisted deep learning method for accelerated 2D MR-EPT reconstructions.
- To assess the performance of the proposed method in terms of accuracy and computational efficiency compared to established techniques.
- To investigate the use of Convolutional Neural Networks (CNNs) within an iterative reconstruction framework for MR-EPT.
Main Methods:
- A novel unrolled, physics-assisted method employing a cascade of CNNs for 2D MR-EPT reconstruction was developed.
- Each CNN in the cascade was designed to compute contrast updates, incorporating physical principles via gradient descent directions.
- The method was trained and validated in silico using realistic 2D brain models at 128 MHz.
Main Results:
- The proposed physics-assisted deep learning method successfully reconstructed electrical properties (EPs) maps from 2D MR-EPT data.
- Reconstruction quality was found to be comparable to the widely used Contrast Source Inversion-EPT (CSI-EPT) method.
- A significant reduction in computational time was achieved compared to traditional methods.
Conclusions:
- The developed unrolled, physics-assisted deep learning approach offers a promising alternative for fast and accurate 2D MR-EPT.
- This AI-driven technique has the potential to accelerate EPs mapping in MRI, facilitating broader clinical adoption.
- The integration of physics-based constraints within deep learning enhances the robustness and efficiency of MR-EPT reconstructions.
