Explainable dual-stage context-aware SwinUNet for accurate optic disc and optic cup segmentation in glaucoma...

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Accurate segmentation of optic disc (OD) and optic cup (OC) in retinal fundus images is vital for glaucoma diagnosis.
• Image variations like illumination and low contrast pose challenges to precise boundary delineation.

Purpose of the Study:

• To enhance the accuracy of OD and OC segmentation for improved glaucoma assessment.
• To develop a robust model capable of overcoming segmentation difficulties in retinal images.

Main Methods:

• Proposed an Enhanced SwinUNet model integrating Swin Transformer (ST) blocks for hierarchical feature extraction.
• Incorporated a Dual-Stage Context-Aware Feature Refinement (DCF-Refine) module in skip connections for enhanced feature fusion.
• Applied CLAHE-based contrast enhancement and min-max normalization for preprocessing retinal images.

Main Results:

• The Enhanced SwinUNet achieved superior segmentation performance on Drishti-GS and REFUGE datasets.
• Attained high accuracies of 99.3% on Drishti-GS and 99.1% on REFUGE.
• Demonstrated effectiveness in addressing segmentation challenges posed by image variations.

Conclusions:

• The proposed model offers highly accurate and reliable segmentation of OD and OC structures.
• Effectively addresses challenges in retinal image analysis for glaucoma assessment.
• Shows potential for clinical application in ophthalmology.