Full-waveform inversion (FWI) is a powerful reconstruction technique for generating high-resolution models of tissue and bone structure using non-invasive ultrasonic measurements. Quantitative information on the model parameters, such as the sound speed in soft tissue, are obtained via an iterative data fitting procedure that accounts for the nonlinear relationship between the ultrasonic wavefield and the model parameters. However, owing to its computational complexity, FWI has yet to see widespread adoption in medical practice. This study applies acoustic FWI to in-vivo data with the objective to showcase the potential gain in resolution of reconstructed models. We identify three key components that are essential to reconstruct such high-resolution models and to accurately identify anatomical features: (1) an accurate model of the source wavelet, obtained by inverting for the source-time function using the calibration data in water; (2) a misfit functional that reduces nonlinearity and dependence on the initial model, which is achieved with a definition of the misfit in terms of graph-space optimal transport; and (3) a resolution analysis based on point perturbations, providing proxies for local spatial smearing of reconstructed features.
Ulrich et al. (Fri,) studied this question.