Joint optic disc and cup segmentation using feature fusion and attention

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Science

Background:

• Glaucoma is a leading cause of irreversible vision loss and remains incurable.
• Early detection and treatment are critical to slow vision loss and prevent blindness.
• Accurate Cup to Disc Ratio (CDR) measurement via optic disc (OD) and optic cup (OC) segmentation aids glaucoma diagnosis.

Purpose of the Study:

• To propose a deep learning architecture, FAU-Net, for accurate joint segmentation of OD and OC.
• To improve upon existing U-Net architectures for enhanced segmentation accuracy, particularly for OC segmentation.

Main Methods:

• Developed FAU-Net (feature fusion and attention U-Net), an improved U-Net architecture.
• Incorporated a feature fusion module to minimize information loss during feature extraction.
• Integrated channel and spatial attention mechanisms to focus on relevant features and suppress irrelevant ones.
• Utilized a multi-label loss function for joint OD and OC segmentation.

Main Results:

• FAU-Net demonstrated superior performance in joint OD and OC segmentation compared to state-of-the-art methods.
• The proposed architecture achieved high accuracy across multiple benchmark datasets: Drishti-GS1, REFUGE, RIM-ONE, and ODIR.

Conclusions:

• FAU-Net offers a promising deep learning solution for precise OD and OC segmentation in glaucoma diagnosis.
• The model's effectiveness in improving segmentation accuracy can aid in earlier glaucoma detection and management.