MRMR optimized classification for automatic glaucoma diagnosis.

Area of Science:

• Ophthalmology
• Medical Informatics
• Biomedical Engineering

Background:

• Glaucoma diagnosis relies on complex data analysis.
• Feature selection is crucial for efficient diagnostic algorithms.
• Information theory offers novel approaches to feature selection.

Purpose of the Study:

• To apply the Min-Redundancy Max-Relevance (mRMR) principle for automatic glaucoma diagnosis.
• To develop a compact and efficient diagnostic classifier using mRMR.
• To improve the interpretability of automatic glaucoma diagnosis algorithms.

Main Methods:

• Utilized the mRMR feature selection methodology.
• Integrated clinical screening data and retinal fundus images.
• Trained an mRMR-optimized classifier on 650 subjects' eye records.

Main Results:

• Achieved a significantly compact classifier, using less than 25% of the initial features.
• Demonstrated improved diagnostic process accessibility for clinicians through feature ranking.
• Validated the methodology on a dataset from the Singapore Eye Research Institute.

Conclusions:

• The mRMR approach offers a powerful tool for developing efficient automatic glaucoma diagnosis systems.
• Feature selection via mRMR enhances classifier compactness and clinical interpretability.
• This work advances the field of automated ophthalmic diagnostics.