Self-supervised light fluence correction network for photoacoustic tomography based on diffusion equation.

Deep learning (DL) shows promise in estimating the absorption coefficient distribution of biological tissue in quantitative photoacoustic tomography (QPAT) imaging, but its application is limited by a lack of ground truth for supervised network training. To address this issue, we propose a DL-based light fluence correction method that only uses the original PAT images for network training. Our self-supervised QPAT network model, which we termed SQPA-Net, introduces light fluence estimation based on diffusion equation to the loss function, and thus guides the model to learn an implicit representation of photoacoustic light transport within tissue. Simulation and small animal imaging experiments demonstrate the effectiveness and efficiency of our method. Compared to current DL-based methods and traditional iterative correction method, the proposed SQPA-Net achieves better light fluence correction results and significantly reduces the processing time.