Chákṣu: A glaucoma specific fundus image database

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Science

Background:

• Glaucoma diagnosis relies on expert interpretation of retinal fundus images.
• Developing automated glaucoma detection systems requires large, annotated datasets.
• Existing databases may lack diversity or comprehensive expert annotations.

Purpose of the Study:

• Introduce Chákṣu, a novel retinal fundus image database.
• Facilitate the evaluation of computer-assisted glaucoma prescreening techniques.
• Provide a resource for developing artificial intelligence-based glaucoma diagnostics.

Main Methods:

• Compiled 1345 color fundus images from three camera brands.
• Included expert ophthalmologist annotations for optic disc and optic cup outlines.
• Generated segmentation ground-truths using mean, median, majority, and STAPLE algorithms.
• Calculated cup-to-disc ratios and provided image-wise glaucoma decisions.

Main Results:

• The Simultaneous Truth and Performance Level Estimation (STAPLE) algorithm achieved the highest agreement with expert annotations.
• Performance indices demonstrated varying levels of agreement for different annotation fusion methods.
• The database provides detailed annotations crucial for algorithm training and validation.

Conclusions:

• Chákṣu is the largest Indian-ethnicity-specific fundus image database with expert annotations.
• The database serves as a valuable resource for advancing AI-driven glaucoma diagnostics.
• Chákṣu will aid in the development and evaluation of computer-assisted glaucoma prescreening tools.