Efficient optic cup localization using regional propagation based on retinal structure priors.

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Glaucoma detection relies on identifying the optic cup in 2D fundus images.
• Accurate optic cup localization is essential for clinical diagnosis and monitoring.
• Existing pixel-based methods often require manual labeling and struggle with accuracy.

Purpose of the Study:

• To develop a regional propagation approach for accurate optic cup localization in 2D fundus images.
• To improve glaucoma diagnosis by enhancing optic cup segmentation.
• To reduce reliance on manually labeled training data for optic cup detection.

Main Methods:

• Processing fundus images at the superpixel level for enhanced feature extraction.
• Utilizing structural priors on relative optic cup and disc positions, eliminating the need for manual labels.
• Implementing a refinement scheme with local context information to boost localization accuracy.

Main Results:

• Achieved a 34.9% non-overlap ratio with manual ground-truth on the ORIGA-light dataset.
• Obtained an absolute cup-to-disc ratio (CDR) error of 0.104.
• Demonstrated significantly higher accuracy than state-of-the-art pixel-based techniques.

Conclusions:

• The superpixel-based regional propagation approach offers a robust and accurate method for optic cup localization.
• This technique provides a viable, label-free alternative for glaucoma screening tools.
• The method achieves high accuracy with comparable or reduced computational cost.