Global Information-driven Unfolding Network for Photoacoustic Imaging

Photoacoustic tomography (PAT), as a noninvasive functional imaging modality, provides a tool for visualizing the absorption properties of chromophores within biological tissue. However, factors such as energy attenuation and signal noise continue to limit imaging accuracy. In this work, we propose a perturbation unfolding network integrated global information (GAPNet) for photoacoustic reconstruction. Results on a publicly available human forearm dataset demonstrate that, compared with conventional analytical methods, model-based approaches, and existing deep unfolding networks, GAPNet effectively restores fine vascular details with enhanced noise robustness. Furthermore, reconstruction results from human breast tissue experiments indicate that GAPNet can accurately recover both superficial and deep features across multiple wavelengths. These findings highlight GAPNet as a promising solution for advancing the clinical application of PAT in medical imaging.