Physics-Informed Complex-Valued Network for Electrical Properties Tomography in 7T MRI Using Multi-Channel Receive Arrays

Motivation: Current MR-Electrical Properties Tomography (MREPT) methods for ultra-high-field MRI depend on transceiver coil arrays and second-order differential operations on magnetic fields, often leading to errors and inaccuracies in reconstructions. Goal(s): To create a neural network that reconstructs electrical properties (EPs) using $B₁⁻$ data collected from regular receive arrays. Approach: A complex-valued, physics-informed neural network was trained on simulation data, then validated on both simulated and experimental 7T human head data. The network leverages the $B₁⁻$ magnitude and relative phase from each receive channel to extract EPs Results: The network effectively reconstructed EPs in both simulation and 7T human brain MRI experiment. Impact: This approach utilizes $B₁⁻$ magnitude and relative phase data from regular receive array to reconstruct EPs. It requires no additional hardware or complex software, offering potential for clinical MREPT applications.