Generative model for sparse photoacoustic tomography artifact removal

Sparse reconstruction in photoacoustic tomography has always faced the problem of artifacts. To address this issue, a diffusion model-based method for sparse data reconstruction in photoacoustic tomography was proposed. During the training phase, the gradient of the probability density of the image was learned as the data prior by adding noise and denoising at each step. During the testing phase, ultrasonic signals are generated by illuminating with pulsed laser and acquired by ultrasonic transducers surrounding the object, which was implemented using the k-Wave toolbox. The reconstructed image was finally obtained by reserve-time Stochastic Differential Equation (SDE). Experimental results on vascular data show that the proposed algorithm can effectively remove artifacts and improve image quality compared with conventional reconstruction methods under 32 and 64 detectors, respectively.