Glaucoma diagnosis using multi-feature analysis and a deep learning technique.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Glaucoma diagnosis relies on various factors, but enhanced diagnostic tools are needed.
• Optical coherence tomography (OCT) provides detailed optic nerve head (ONH) imaging.
• Machine learning (ML) shows promise for automated disease detection.

Purpose of the Study:

• To improve glaucoma diagnostic assessment by analyzing multiple features.
• To introduce a novel cross-sectional ONH feature from OCT images.
• To compare deep learning (DL) and logistic regression (LR) for automated glaucoma detection.

Main Methods:

• Collected structural, functional, demographic, and risk factor data (n=100 per group).
• Analyzed features and trained ML algorithms (DL and LR) on significant features.
• Performed image segmentation on OCT scans to extract and analyze ONH cup area for a new DL model.
• Evaluated model performance on unseen test data (n=55) using metrics like AUC and accuracy.

Main Results:

• The DL model trained on optimal features achieved high diagnostic performance (AUC 0.98, 97% validation accuracy, 96% test accuracy).
• A second DL model using the novel ONH cup area feature showed excellent results (AUC 0.99, 98.6% accuracy).
• Both models outperformed previous automated glaucoma detection studies.

Conclusions:

• The developed DL models, utilizing optimal features and a novel ONH feature, demonstrate significant potential for glaucoma screening.
• These AI-driven tools can assist clinicians in making precise glaucoma diagnostic decisions.
• The findings suggest a promising future for automated glaucoma detection using OCT imaging and deep learning.