An automatic quantitative measurement method for performance assessment of retina image registration algorithms

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Accurate registration of retina fundus images is crucial for diagnosing and monitoring eye diseases.
• Existing methods for evaluating retina image registration performance often rely on subjective visual assessment.
• There is a need for objective, quantitative metrics to assess the performance of image registration algorithms.

Purpose of the Study:

• To develop a novel automatic quantitative measurement method for evaluating retina image registration algorithms.
• To introduce an objective performance assessment tool for retina fundus image registration.

Main Methods:

• Proposed an automatic quantitative measurement method using edges and edge dissimilarity for performance assessment.
• Introduced a novel edge dissimilarity measure termed "robustified Hausdorff distance".
• Input consists of registered pairs of retina fundus images from existing algorithms.

Main Results:

• Demonstrated the feasibility of the proposed automatic quantitative measurement method.
• The method's performance was validated by comparison with visual evaluation results.
• Tested on the DRIVERA and G9 datasets, showing comparable results to visual assessments.

Conclusions:

• The proposed robustified Hausdorff distance offers a feasible and automatic quantitative method for assessing retina image registration performance.
• This approach provides an objective alternative to subjective visual evaluations.
• The method can be applied to evaluate any existing retina image registration algorithm.