Full‑waveform inversion for sound speed and absolute attenuation in a living mouse

Attenuation is a key parameter in ultrasound computed tomography for identifying malignant cells, as cancerous regions typically exhibit elevated sound speed and attenuation compared to healthy tissue. We develop a timedomain full-waveform inversion (FWI) approach to reconstruct sound speed and attenuation with sub-millimetre resolution on a cross-section of a living mouse. We use different misfit functions to effectively decouple the inverse problem into a sound speed and an attenuation part. First, we perform straight-ray tomography to construct smooth initial sound speed and attenuation models that mitigate cycle-skipping at the beginning of the FWI. Then, we successfully obtain an accurate sound speed model through FWI using an optimal transport misfit. This misfit function isolates phase differences, while being more robust than the conventional L2 waveform misfit against cycle-skipping. We extend the analysis to obtain an attenuation model by applying FWI with an amplitude-based misfit function. This misfit allows us to emphasise amplitude over phase discrepancies to better resolve attenuation features. From the amplitude misfit, we define an attenuation sensitivity kernel, derived from the adjoint method, which captures dispersion and energy loss for a more accurate inversion. After inversion, our simulations fit the observations and we observe a clear misfit reduction. Despite the presence of the spine, we obtain plausible sound speed and attenuation images of the mouse with the acoustic FWI. We compare our results with an existing frequency-domain solver and conclude that our time-domain inversion method, although more time-consuming, gives more plausible attenuation values. Our method allows for the identification of high absolute attenuation and sound speed regions, which is critical for the detection of cancerous tumours.