Explainable Severity Ranking via Pairwise n-Hidden Comparison: Case Study of Glaucoma

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Primary open-angle glaucoma (POAG) is a leading cause of irreversible blindness.
• Accurate glaucoma severity assessment is crucial for timely intervention but faces challenges due to subjective classification by ophthalmologists.
• Patients often remain unaware of vision loss due to the gradual onset of POAG.

Purpose of the Study:

• To develop and validate a novel framework for comparing and interpreting glaucoma severity using fundus images.
• To address the inter-observer variability in glaucoma severity classification.
• To provide explainable AI insights into glaucoma severity assessment.

Main Methods:

• A siamese-based deep learning model was employed for pairwise severity ranking of fundus images.
• The framework utilizes n-hidden comparisons to analyze image pairs.
• Pairwise saliency maps were generated to explain the model's severity predictions.

Main Results:

• The proposed AI framework demonstrated superior diagnostic accuracy compared to traditional methods for glaucoma severity ranking.
• The model successfully provided saliency explanations, highlighting image regions contributing to severity classification.
• The approach showed promise in standardizing glaucoma severity assessment.

Conclusions:

• The developed AI framework offers a more accurate and objective method for assessing POAG severity from fundus images.
• Explainable AI through saliency maps enhances the interpretability and trustworthiness of the model.
• This technology has the potential to improve clinical decision-making and patient outcomes in glaucoma management.