Generative priors-constraint accelerated iterative reconstruction for extremely sparse photoacoustic tomography boosted by mean-reverting diffusion model: Towards 8 projections

Abstract

As a novel non-invasive hybrid biomedical imaging technology, photoacoustic tomography combines the advantages of high contrast of optical imaging and high penetration of acoustic imaging. However, the conventional standard reconstruction methods under sparse view may lead to low-quality image in photoacoustic tomography. To address this problem, an advanced sparse reconstruction method for photoacoustic tomography based on the mean-reverting diffusion model is proposed. By modeling the degradation process from a high-quality image under full-view scanning (512 projections) to a sparse image with stable Gaussian noise (i.e., mean state), a mean-reverting diffusion model is trained to learn prior information of the data distribution. Then the learned prior information is employed to generate a high-quality image from the sparse image by iteratively sampling the noisy state. Blood vessels simulation data and the animal in vivo experimental data were used to evaluate the performance of the proposed method. The results demonstrate that the proposed method achieves higher-quality sparse reconstruction compared with conventional reconstruction methods and U-Net method. In addition, the proposed method dramatically speeds up the sparse reconstruction and achieves better reconstruction results for extremely sparse images compared with the method based on conventional diffusion model. The proposed method achieves an improvement of 0.52 (∼289 %) in structural similarity and 10.01 dB (∼59 %) in peak signal-to-noise ratio for extremely sparse projections (8 projections), compared with the conventional delay-and-sum method. This method is expected to shorten the acquisition time and reduce the cost of photoacoustic tomography, thus further expanding the range of applications.