A deep learning framework for light propagation modelling for quantitative photoacoustics

Photoacoustic (PA) imaging is a hybrid modality based on optical absorption and ultrasound (US) detection. Quantitative PA imaging provides valuable functional information such as blood oxygen saturation, sO<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf>, by estimating chromophore concentrations from multispectral PA images. The quantification remains challenging due to unknown light attenuation in heterogeneous tissues. Monte Carlo (MC) simulation is regarded as the gold standard for modelling light propagation in turbid media. It leverages stochastic modelling methods through the simulation of the random walk of photon packets, thus it is computationally demanding and not suitable for real-time applications. In this work, for the first time, we propose a deep learning (DL) framework for light propagation modelling, with a focus on quantitative PA imaging. Compared to the MC simulation, our method reduced the computation time by 4 orders of magnitude, from 33 minutes to 46 milliseconds for a 3D simulation. In addition, the DL-based light fluence estimation improved sO<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf> quantification accuracy, achieving an average estimation error of 0.3% with a blood phantom, and showed no significant difference compared to the MC simulation. This framework aims to provide a time-efficient solution for light propagation modelling in turbid media, thereby enhancing the quantification accuracy for real-time PA imaging applications.