Enhanced photoacoustic tomography via accelerated mean-reverting generative diffusion model

Abstract

Photoacoustic tomography (PAT), as a novel non-invasive hybrid biomedical imaging technology, combines the advantages of high contrast from optical imaging and deep penetration from acoustic imaging, and its applications in biomedical imaging are becoming increasingly widespread. However, the conventional standard reconstruction methods under sparse view may lead to low-quality image in photoacoustic tomography. To address this issue, this paper proposes a sparse sinogram (projection domain) data reconstruction method based on mean-reverting diffusion model. By simulating the forward and reverse of the Stochastic Differential Equation (SDE) processes from high-quality images (full-view projection data) to degraded low-quality images (sparse-view projection data), this method enables the restoration of sparse-view projection data to full-view projection data without relying on any task-specific prior knowledge. Blood vessels simulation data, circular phantom data, the animal in vivo experimental data, and data acquired from an actual PAT system were used to evaluate the performance of the proposed method. In the experimental tests on the “T”-shaped sample data acquired from the actual imaging system, even under extremely sparse projections (16 projections), the proposed method demonstrated significant improvements in peak signal-to-noise ratio compared to Cycle-GAN and U-Net, with increases of 16.46 dB (∼69.2 %) and 0.86 dB in the projection domain, respectively. This method enhances the sparse reconstruction capability of PAT in the sinogram domain, which is expected to reduce the costs and shorten the acquisition time of PAT in the practical applications, thus further expanding the application scope of PAT.