Skeleton-guided diffusion reconstruction from sparse views for high-fidelity photoacoustic imaging

Photoacoustic tomography (PAT), as an emerging medical imaging technique, demonstrates broad application prospects in vascular imaging, tumor detection, and tissue visualization due to its high resolution and deep penetration capabilities. However, acquiring high-quality PAT images relies on dense view sampling, which directly leads to issues such as time-consuming acquisition and high computational costs during imaging. To enhance imaging efficiency, sparse view photoacoustic tomography has become a common approach, but this leads to significant image quality degradation, particularly in preserving fine vascular structures. Addressing this challenge, we propose a skeleton-guided diffusion reconstruction framework. This framework ingeniously utilizes vascular skeleton structures extracted from low-quality reconstructions as prior information, feeding both the extracted skeleton and the original sparse view images into the diffusion model. Through this innovative design, our method achieves high-fidelity image reconstruction under sparse sampling conditions. Compared to methods such as DAS, BP, TR, CS, and U-net, our approach demonstrates significant improvements in both SSIM and PSNR metrics. This not only substantially enhances structural coherence but also markedly improves the visual clarity of the reconstructed images.