Non-stationary estimation of optical absorption in ultrasound-modulated diffuse optical tomography

Ultrasound-modulated diffuse optical tomography (US-DOT) is an emerging imaging modality where spatial distributions of optical parameters of an imaged target are estimated from measured near-infrared light modulated by a focused ultrasound through an acousto-optic effect. The use of acoustic modulation enables increased resolution when compared to conventional diffuse optical tomography. Generally, estimation methods in US-DOT have assumed stationary tissue properties, limiting their capability to capture temporal variations of the unknown optical parameters within biological tissues that can occur, for example, due to the time needed to scan with multiple ultrasound focuses. In this work, we aim to estimate the dynamic optical parameters of tissues by approaching the image reconstruction problem of US-DOT in the Bayesian framework for non-stationary estimation. We implemented two time-varying models: a random walk and a vector autoregression model. The approach was evaluated with numerical simulations for the difference (linear) and absolute (non-linear) imaging of absorption coefficients. The results were compared to a conventional approach where the state estimation model was not used. The results demonstrate that state estimation significantly improves the reconstruction of non-stationary targets in US-DOT, even if relatively simple time-evolution models are used.