Combining Feature Correspondence With Parametric Chamfer Alignment: Hybrid Two-Stage Registration for Ultra-Widefield...

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Ultra-widefield (UWF) fluorescein angiography (FA) images often exhibit extreme geometric distortions.
• Accurate registration of these images is crucial for analyzing changes in retinal structures over time.

Purpose of the Study:

• To introduce a novel hybrid framework for robust retinal image registration, specifically addressing distortions in UWF-FA images.
• To improve the accuracy and clinical utility of retinal image analysis, particularly for longitudinal studies.

Main Methods:

• A two-stage hybrid approach combining feature-based global registration and vessel-based local refinement.
• Modified RANSAC (random sample and consensus) for robust feature matching and polynomial transformation estimation.
• Parametric chamfer alignment of deep neural network-derived vessel maps for local refinement.

Main Results:

• The proposed framework significantly outperforms existing retinal image registration methods in accuracy.
• Demonstrated effectiveness on both UWF-FA and narrow-field fundus image registration datasets.
• Improved registration accuracy, especially in peripheral image regions affected by distortions.

Conclusions:

• The novel hybrid framework provides accurate registration for distorted UWF-FA images.
• Enables automated computation of vessel change metrics and standardized co-registered examinations.
• Enhances the clinical utility of longitudinal UWF imaging for detecting subtle changes.