SwinCup-DiscNet: A fusion transformer framework for glaucoma diagnosis using optic disc and cup features

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Glaucoma is a leading cause of irreversible blindness, characterized by optic nerve head (ONH) damage.
• Early detection of glaucoma is crucial for preventing vision loss.

Purpose of the Study:

• To develop and evaluate a novel fusion transformer pipeline for effective glaucoma screening.
• To integrate optic disc/cup segmentation with feature-based analysis for enhanced diagnostic accuracy.

Main Methods:

• A U-Net with an attention mechanism was used for Optic Disc (OD) and Optic Cup (OC) segmentation.
• Spectral shape descriptors were employed to calculate the Vertical Cup-to-Disc Ratio (CDR).
• Swin Transformer encoder extracted fundus image features, and a probabilistic fusion method merged CDR and deep learning features for classification.

Main Results:

• The SwinCup-DiscNet framework demonstrated superior performance compared to traditional CNN-based models and segmentation-only methods across three public datasets (LAG, ACRIMA, DRISTHI-GS).
• The model achieved high accuracy, DSC IoU, F1-score, and low Cup-to-Disc Ratio Mean Absolute Error (CDR MAE).

Conclusions:

• SwinCup-DiscNet is a robust, reliable, and clinically interpretable tool for early glaucoma detection.
• The proposed framework shows significant potential as an effective clinical tool in real-world settings.