Abstract
Objective: To develop a high-performance reconstruction framework that enables high-quality photoacoustic tomography (PAT) imaging under limited-view and sparse-view acquisition constraints. Impact Statement: The proposed method reduces the number of required acoustic transducers while maintaining image quality comparable to full-view systems, providing a practical and cost-efficient solution for biomedical PAT imaging. Introduction: PAT offers high-resolution visualization of biological tissues. However, restrictions such as reduced transducer counts or incomplete detection geometries render the inverse problem severely ill-posed, leading to marked degradation in reconstructed images. Although diffusion models have recently shown strong promise for image restoration, existing architectures can be computationally intensive or insufficiently expressive for the complexities of PAT.Methods: We introduce a time-driven transformer-based photoacoustic diffusion model (TT-PADM) that directly restores high-quality images from limited-view and sparse-view PAT reconstructions. TT-PADM uses a time-driven transformer within a time-dependent noise-estimation network, reducing model parameters by over 80% relative to conventional transformer designs while enhancing the generative capacity of the diffusion process. Results: Simulations and experimental results show that TT-PADM delivers high-fidelity reconstructions even under severely limited acquisition conditions, producing image quality comparable to full-view PAT systems. Quantitative and qualitative analyses show that TT-PADM consistently surpasses state-of-the-art reconstruction approaches, providing notable improvements in structural accuracy and noise suppression. Conclusion: TT-PADM offers a robust, parameter-efficient, and highly effective solution for PAT image restoration under practical hardware constraints, with strong potential for deployment in resource-limited biomedical imaging scenarios.