Time-domain reconstruction of the speed of sound in ring-array ultrasound computed tomography with randomized super-shots

Abstract

Ring-array ultrasound computed tomography has recently achieved sufficient maturity for clinical applications like breast imaging. Image reconstruction is achieved with state of the art iterative algorithms, full-waveform inversion in the frequency domain. Equivalent iterative techniques in the time-domain are available, but they are not practical due to prohibitive computing times. To overcome these limitations, time-domain methods based on stochastic optimization have been developed, with promising initial results. In this Letter, we revisit the time-domain stochastic reconstruction by introducing the notion of multiple super-shots and stochastic ensembles. We test our inversion algorithm on synthetic data and publicly available patient data. Whereas a time-domain randomized inversion remains significantly slower with respect to a deterministic inversion in the frequency domain, we show that image quality may, indeed, be comparable to state of the art methods. In particular, the methods presented in this Letter allow us to obtain images with sufficient quality for clinical tasks such as the detection of masses in breast tissue and their characterization as benign or malignant.