FBS-DiffNet: a deep unfolding network with diffusion prior for joint reconstruction of optical absorption and sound speed in photoacoustic tomography
Year: 2026
Type: article
Abstract: Quantitative photoacoustic tomography faces challenges arising from ill-posedness caused by acoustic heterogeneity and multi-parameter coupling. We propose the forward-backward splitting diffusion network (FBS-DiffNet), to simultaneously reconstruct optical absorption coefficient and sound speed distributions. This framework unfolds the classical forward-backward splitting algorithm into a cascaded architecture. Each level integrates a data consistency module driven by a residual gradient step network and a generative joint diffusion module. This configuration ensures physical fidelity to measured signals while exploiting spatial correlations between parameters at anatomical boundaries to provide synergistic regularization. Experimental results on simulation, phantom, and in vivo datasets demonstrate that FBS-DiffNet surpasses uniform sound speed methods and prominent sound speed estimation techniques such as K-means-GMM or NF-APACT concerning both visual quality and quantitative metrics. Ablation studies verify the contribution of each key component to these results. By combining physical interpretability with deep learning representational capacity, FBS-DiffNet offers a promising approach for high-precision quantitative photoacoustic tomography.
Source: Biomedical Optics Express
DOI: 10.1364/boe.608902
Authors: Sun Zheng, Ding Gang’ao
Language: English
Subfield: Biomedical Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $1,913
OpenAlex ID: W7204489678