Patrick Stähli, Martin Frenz, Michael Jaeger
Computed ultrasound tomography in echo mode (CUTE) is a multimodal extension to conventional pulse-echo ultrasound (US), with promise for improving the US-based diagnosis of e.g. cancer or liver disease. It is based on measuring the phase shift of echoes when detected under a variety of transmit and receive steering angles. By inverting a forward model describing the relation between speed-of-sound (SoS) and phase shift, the SoS is reconstructed. This inversion is ill-posed and thus requires some way of regularization. We recently obtained promising results of imaging the SoS inside the liver when employing a Bayesian framework for regularization. This allows to greatly reduce phase-noise related artifacts by including a priori spatial statistical properties of the SoS distribution, which is readily available by segmenting the anatomy in the conventional US image. In this talk, we first detail the basics of data acquisition and the forward model. Then we introduce the Bayesian approach and demonstrate its strengths compared to a more naïve Tikhonov inversion, in simulations, phantom studies, and in vivo. Finally, we show preliminary in-vivo liver imaging results where the segmentation step is automated using deep learning, resulting in a real-time framerate of the full processing chain.