TT-PADM: A Time-Driven Transformer Diffusion Model for Robust Sparse-View and Limited-View Photoacoustic Tomography

Area of Science:

• Biomedical imaging
• Medical physics
• Computational imaging

Background:

• Photoacoustic tomography (PAT) provides high-resolution biological tissue visualization.
• Limited-view and sparse-view PAT acquisitions lead to ill-posed inverse problems and degraded image quality.
• Existing diffusion models for image restoration can be computationally intensive and lack expressiveness for PAT.

Purpose of the Study:

• To develop a high-performance reconstruction framework for high-quality PAT imaging under limited-view and sparse-view constraints.
• To reduce the number of acoustic transducers required for PAT while preserving image quality.
• To offer a practical and cost-efficient solution for biomedical PAT imaging.

Main Methods:

• Introduction of a time-driven transformer-based photoacoustic diffusion model (TT-PADM).
• TT-PADM utilizes a time-driven transformer within a time-dependent noise-estimation network.
• This approach reduces model parameters by over 80% and enhances the diffusion process's generative capacity.

Main Results:

• TT-PADM achieves high-fidelity reconstructions from severely limited acquisition data.
• Image quality is comparable to full-view PAT systems.
• TT-PADM surpasses state-of-the-art methods in structural accuracy and noise suppression.

Conclusions:

• TT-PADM is a robust, parameter-efficient, and effective solution for PAT image restoration under hardware constraints.
• The model has strong potential for deployment in resource-limited biomedical imaging.
• This framework addresses practical challenges in PAT acquisition and reconstruction.