Enhanced score-based priors-boosted generative model for resolution breakthrough in photoacoustic tomography

Area of Science:

• Biomedical imaging
• Medical physics
• Signal processing

Background:

• Photoacoustic tomography (PAT) offers optical contrast and acoustic depth resolution.
• Limited bandwidth of ultrasonic transducers degrades conventional PAT image resolution.
• Advanced reconstruction methods are needed to overcome bandwidth constraints in PAT.

Purpose of the Study:

• To develop a novel sinogram-domain signal spectral reconstruction method for enhanced PAT imaging.
• To improve image resolution in PAT by recovering missing frequency components.
• To address the limitations imposed by industry-standard ultrasonic transducer bandwidth.

Main Methods:

• Proposed an enhanced score-based diffusion model (ESDM) for sinogram data reconstruction.
• Employed denoising score-matching to learn sinogram data distributions.
• Utilized an auxiliary network to capture cross-bandwidth dependency priors for optimized reconstruction.

Main Results:

• The ESDM demonstrated superior recovery of missing frequency components compared to U-Net and Cycle-GAN.
• Achieved approximately 1.60 times higher resolution than the limited-bandwidth baseline.
• Validated performance using both simulated and real-system experimental PAT data.

Conclusions:

• The proposed ESDM significantly enhances PAT imaging quality under bandwidth constraints.
• This advancement broadens the clinical applicability of photoacoustic tomography.
• The method offers a promising solution for improving resolution in limited-bandwidth PAT systems.