Edge-Aware Short-Chain Diffusion Enables High-Fidelity Sparse-Sampling Optoacoustic Tomography

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Computational Imaging

Background:

• Multispectral optoacoustic tomography (MSOT) offers valuable biochemical and molecular contrast in tissues.
• MSOT image quality is frequently compromised by system noise and sparse sampling.
• Existing methods struggle to maintain image fidelity under these challenging conditions.

Purpose of the Study:

• To develop a novel framework for enhancing MSOT image quality.
• To address limitations posed by strong noise and extreme sparse sampling in MSOT.
• To reduce hardware requirements while improving image reconstruction.

Main Methods:

• A hybrid diffusion-based framework, the optoacoustic universal denoising network (OA-UDNet), was proposed.
• The model was trained on over 250,000 in vivo images.
• A joint denoising and image restoration approach was employed, integrating an edge-aware module with diffusion-based generation.

Main Results:

• Using only 32 detectors, OA-UDNet improved image quality and increased peak signal-to-noise ratio (PSNR) by approximately 14 dB.
• The framework demonstrated consistent structural fidelity preservation under extreme undersampling compared to a 256-detector reference.
• Validation across mouse models, tumor imaging, and human samples showed reduced artifacts and enhanced anatomical/functional feature recovery.

Conclusions:

• OA-UDNet effectively enhances MSOT image quality, particularly under sparse sampling conditions.
• The method provides a practical solution for accelerating MSOT imaging and simplifying hardware complexity.
• This advancement holds potential for broader applications in preclinical and clinical imaging.