Physics Informed Deep Unfolded Full Waveform Inversion for Edema Detection

Accurate detection of edema is clinically important but remains challenging due to the subtlety of its quantitative indicators. Ultrasound (US) offers a safe, accessible, and cost-effective imaging modality, yet conventional beamforming methods such as B-mode do not directly recover the tissue’s physical parameters. In this work, we present a physics-informed deep learning approach that performs inverse reconstruction of tissue properties directly from raw channel data, enabling quantitative estimation of the speed of sound and density. Our method, called Deep-Unfolded Full Waveform Inversion (DUFWI), unfolds the iterative steps of a classical inverse solver into a trainable neural network, preserving physical interpretability while learning efficient update rules from data. We demonstrate results on both simulated datasets and real hardware experiments using a Verasonics US system with phantom setups containing cylindrical rods of known speed of sound, showing substantial improvement over traditional FWI and MB-QRUS in performance and computational demand. The framework can be used for a wide range of inverse US imaging tasks, offering a practical path toward real-time, physics-based diagnostic imaging.