Deep proximal gradient network for absorption coefficient recovery in photoacoustic tomography.

The optical absorption properties of biological tissues in photoacoustic (PA) tomography are typically quantified by inverting acoustic measurements. Conventional approaches to solving the inverse problem of forward optical models often involve iterative optimization. However, these methods are hindered by several challenges, including high computational demands, the need for regularization, and sensitivity to both the accuracy of the forward model and the completeness of the measurement data. The aim of this study is to introduce a novel learned iterative method for recovering spatially varying optical absorption coefficients (OACs) from PA pressure measurements.The study introduces a deep learning-based approach that employs the proximal gradient descent mechanism to achieve optical inversion. The proposed framework consists of multiple cascaded structural units, which iteratively update the absorption coefficients through a learning process, unit by unit.The proposed method was validated through simulations, phantom experiments, andstudies. Comparative analyses demonstrated that the proposed approach outperforms traditional nonlearning and learning-based methods, achieving at least 12.85% improvement in relative errors, 3.50% improvement in peak signal-to-noise ratios, and 3.53% improvement in structural similarity in reconstructing the OAC distribution.This method significantly improves the accuracy and efficiency of quantitative PA tomography. By addressing key challenges such as computational demand and sensitivity to the accuracy of the forward model and the completeness of the measurement data, the proposed framework offers a more reliable and efficient alternative to traditional methods, with potential applications in medical imaging and diagnostics.