Data-driven electrical conductivity brain imaging using 3 T MRI.

Area of Science:

• Biomedical Engineering
• Medical Imaging
• Computational Electromagnetics

Background:

• Magnetic resonance electrical properties tomography (MR-EPT) non-invasively measures tissue electrical properties (EPs) like conductivity.
• Tissue conductivity shows potential as a biomarker in clinical studies.
• Conventional MR-EPT conductivity reconstructions face inaccuracies due to numerical assumptions.

Purpose of the Study:

• To develop an artificial neural network (ANN)-based non-linear conductivity estimator for improved brain imaging.
• To overcome limitations of traditional model-based MR-EPT conductivity reconstructions.
• To validate the ANN method using simulated, in-silico, and in-vivo data.

Main Methods:

• Trained an ANN on 201 synthesized T2-weighted spin-echo (SE) datasets from finite-difference time-domain (FDTD) electromagnetic simulations.
• The training dataset included T2-w SE magnitude and transceive phase information.
• Evaluated the ANN against conventional phase-based EPT methods (e.g., S-G Kernel, cr-EPT, Poly-Fit, Integral-based) using in-silico, volunteer, and patient data.

Main Results:

• The ANN method produced more accurate conductivity maps with better structural preservation compared to conventional methods in in-silico experiments.
• ANN-based reconstructions showed improved quality, generalizability, and robustness on in-vivo data, including pathologies.
• The method demonstrated reliable performance across various signal-to-noise ratio (SNR) levels and repeatability conditions.

Conclusions:

• The proposed ANN-based MR-EPT conductivity estimator significantly enhances the accuracy and quality of brain conductivity imaging.
• The network's ability to generalize from simulated to in-vivo data, including pathologies, highlights its clinical potential.
• This approach offers a more robust and accurate alternative for quantitative conductivity mapping in medical applications.