Efficient optic cup detection from intra-image learning with retinal structure priors.

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Glaucoma diagnosis relies on identifying structural changes in the optic nerve head from fundus images.
• Accurate localization of the optic cup is crucial for glaucoma assessment.
• Current pixel-based or sliding window methods have limitations in accuracy and computational efficiency.

Purpose of the Study:

• To develop a superpixel-based learning framework for automated glaucoma diagnosis using retinal structure priors.
• To improve the accuracy and computational efficiency of optic cup localization in digital fundus photographs.
• To enable glaucoma diagnosis without relying on pre-labeled training samples.

Main Methods:

• A superpixel-level processing approach for fundus images to extract more descriptive features.
• A novel classifier learning process using structural priors from the test image itself.
• A classification refinement scheme incorporating structural priors and local contextual information.

Main Results:

• The method achieved a 26.7% non-overlap ratio with manual ground-truth on the ORIGA(-light) dataset (650 images).
• An absolute cup-to-disc ratio (CDR) error of 0.081 was obtained, demonstrating high diagnostic accuracy.
• The framework demonstrated a speedup factor of tens to hundreds compared to state-of-the-art techniques.

Conclusions:

• The superpixel-based framework offers a computationally efficient and accurate method for glaucoma diagnosis.
• The approach effectively localizes the optic cup using retinal structure priors, outperforming existing methods.
• This technique holds significant potential for improving automated glaucoma screening and diagnosis.