Physics-Informed Conditional Diffusion for Motion-Robust Retinal Temporal Laser Speckle Contrast Imaging

2026-04-22 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision & Pattern Recognition · Depth: Expert, quick

Summary

RetinaDiff, a physics-informed conditional diffusion framework, enhances retinal temporal laser speckle contrast imaging (tLSCI) by enabling robust blood flow monitoring from only a few frames. Conventional tLSCI requires long speckle sequences for stable statistics, making it susceptible to acquisition disturbances and limiting temporal resolution. RetinaDiff addresses this by first applying phase correlation-based registration to stabilize raw speckle sequences, reducing interframe misalignment and providing a motion-corrected physics prior. Subsequently, a conditional diffusion model performs inverse reconstruction, jointly conditioning on the registered sequence and the corrected prior. Experiments using an in-house retinal LSCI system demonstrate that RetinaDiff achieves improved structural continuity and statistical stability compared to direct reconstruction from limited frames and other baselines, even in challenging cases where conventional methods fail.

Key takeaway

For AI Scientists developing medical imaging reconstruction algorithms, RetinaDiff offers a robust approach to overcoming data limitations and motion artifacts. You should consider integrating physics-informed priors and conditional diffusion models into your frameworks to enhance image quality and stability, especially when working with inherently noisy or sparse medical data. This method can significantly improve diagnostic reliability from limited input.

Key insights

RetinaDiff uses physics-informed diffusion to reconstruct robust retinal blood flow from minimal speckle frames.

Principles

• Motion correction improves flow dynamics accuracy.
• Physics priors guide robust inverse reconstruction.

Method

RetinaDiff registers raw speckle sequences via phase correlation, then a conditional diffusion model performs inverse reconstruction, jointly conditioned on the registered sequence and the motion-corrected physics prior.

In practice

• Stabilize raw speckle sequences with registration.
• Integrate physics priors into diffusion models.

Topics

• Retinal LSCI
• Temporal Laser Speckle Contrast Imaging
• RetinaDiff Framework
• Conditional Diffusion Models
• Motion Robustness

Code references

• QianChen113/RetinaDiff

Best for: Computer Vision Engineer, AI Scientist, Research Scientist

Related on AIssential

• IG-Diff: Complex Night Scene Restoration with Illumination-Guided Diffusion Model — Illumination-guided diffusion models effectively restore complex night scenes by integrating illumination priors and synthetic data.
• Reproducing DragDiffusion: Interactive Point-Based Editing with Diffusion Models — DragDiffusion's interactive image editing relies on single-timestep latent optimization and LoRA for precise, identity-preserving control.
• PPDM: Pixel Puzzling Diffusion Model for Speed and Memory Efficient Volumetric Medical Image Translation — PPDM efficiently translates 3D medical images by trading spatial resolution for channel depth and focusing on task-relevant residuals.
• S23DR 2026 Winning Solution — A flow-matching DiT conditioned on scene tokens effectively reconstructs 3D wireframes from sparse multi-modal inputs.
• Estimation of instrument and noise parameters for inverse problem based on prior diffusion model — A Gibbs-Diffusion Posterior Sampling (G-DPS) based method efficiently estimates observation parameters and images in inverse problems with diffusion priors.
• A physics-informed foundation model for quantitative diffusion MRI — PIGMENT uses a physics-informed generative prior to enable reliable quantitative diffusion MRI from sparse, heterogeneous data.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.