GlaucomaNet

GlaucomaNet diagnoses primary open-angle glaucoma (POAG) from fundus photographs using deep convolutional neural networks to enable automated, accurate detection.

Key Features:

• Dual-Convolutional Neural Network Architecture: Employs two CNNs that learn discriminative features from fundus images and fuse those features to mimic human grading for POAG.
• High Diagnostic Accuracy: Achieves an AUC of 0.904 on the Ocular Hypertension Treatment Study (OHTS) dataset and an AUC of 0.997 on the Large-scale Attention-based Glaucoma (LAG) dataset.
• Ensemble Approach for Enhanced Accuracy: Integrates multiple network architectures to improve robustness to variability in image data.
• Transparency and Comprehensiveness: Reports comprehensiveness scores of 97% and 36%, reflecting measured aspects of interpretability and output coverage.
• Addresses Diagnostic Challenges: Utilizes diverse fundus image data to enhance generalizability and reduce reliance on perimetry tests.

Scientific Applications:

• Automated POAG Diagnosis: Provides automated classification of primary open-angle glaucoma from fundus photographs.
• Population Screening and Early Detection: Processes large volumes of fundus images to support screening and earlier clinical intervention for POAG.
• Explainable AI in Ophthalmology: Simulates human grading and supplies interpretable outputs for research into explainable diagnostic models.

Methodology:

Training of convolutional neural networks, including a dual-CNN architecture with feature fusion and an ensemble of network architectures, on fundus photograph datasets OHTS and LAG.