Shear wave dispersion estimation using deep learning with a multi-frequency approach

Shear wave dispersion (SWD) imaging provides insight into tissue viscoelastic properties by analyzing frequency-dependent shear wave speeds. While traditional approaches generally rely on group velocity or Young's modulus estimation for quantifying tissue stiffness, this study introduces a deep learning framework capable of reconstructing shear wave phase velocity maps across multiple frequencies. We propose a neural network that integrates an encoder, ConvLSTM-based recurrent block, and decoder to generate ten frequency-specific phase velocity maps, enabling direct and accurate SWD estimation. Training was performed using 450 samples of simulated shear wave velocity fields, and validation was performed on both simulated and phantom experiments. Predicted Phase velocity values across 50-950 Hz closely matched reference data, yielding dispersion slopes of 6.45 m/s/K-Hz in simulations and 6.37 m/s/K-Hz in the phantom. Results highlight the ability to directly estimate shear wave dispersion through a deep learning approach that relies on multiple frequency-dependent phase velocity maps.