Active Diffusion Matching: Score-Based Iterative Alignment of Cross-Modal Retinal Images.

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Aligning Standard Fundus Images (SFIs) and Ultra-Widefield Fundus Images (UWFIs) is challenging due to differences in viewing range and retinal appearance.
• Existing image alignment techniques lack the accuracy needed for effective cross-modal analysis.

Purpose of the Study:

• To develop a specialized method for accurate alignment of SFIs and UWFIs.
• To address the unaddressed challenge of cross-modal fundus image alignment.

Main Methods:

• Proposed Active Diffusion Matching (ADM), a novel cross-modal alignment technique.
• Integrated two interdependent score-based diffusion models using an iterative Langevin Markov chain.
• Employed custom sampling strategies for enhanced adaptability to input image pairs.

Main Results:

• ADM achieved state-of-the-art alignment accuracy on both private SFI-UWFI and public SFI-SFI datasets.
• Demonstrated significant improvements in mean Area Under the Curve (mAUC) by 5.2 points on the private dataset and 0.4 points on the public dataset compared to existing methods.

Conclusions:

• ADM effectively bridges the alignment gap between SFIs and UWFIs, offering a robust solution for cross-modal image alignment.
• The method's joint optimization of global and local alignment enhances its utility.
• ADM has the potential to transform integrated SFI and UWFI analysis, improving clinical utility, enabling learning-based enhancements, and potentially enhancing diagnostic accuracy and patient outcomes in ophthalmology.