Parallel diffusion models promote high detail-fidelity photoacoustic microscopy in sparse sampling

Area of Science:

• Biomedical Imaging
• Medical Technology
• Artificial Intelligence in Medicine

Background:

• High spatiotemporal resolution photoacoustic microscopy (PAM) requires reconstructing sparsely sampled data for in vivo microvascular imaging.
• Existing methods like CNNs and GANs struggle with complex microvascular structures and vascular texture, leading to low-detail images.

Purpose of the Study:

• To develop a novel approach for high-fidelity reconstruction of sparsely sampled PAM data.
• To improve the detail and quality of microvascular imaging in high-speed PAM.

Main Methods:

• Proposed parallel diffusion models (PDM) with parallel learning for Noise and Image tasks.
• Noise task optimizes variational lower bounds for realistic microvascular structures.
• Image task uses image-based loss to enhance microvascular detail fidelity.

Main Results:

• PDM achieved an LPIPS of 0.199 using only 1.56% of fully sampled pixels from human oral PAM data.
• PDM effectively prevented breathing artifacts and image distortion in high-speed 16x PAM.
• Reduced standard deviation of Row-wise Self-Correlation Coefficient while maintaining high image quality.

Conclusions:

• PDM demonstrates high confidence in reconstructing detailed information from sparsely sampled data.
• This approach will advance the application of reconstructed data for high spatiotemporal resolution PAM.
• PDM offers a robust solution for enhancing microvascular imaging in challenging conditions.