A multi-scale convolutional neural network with context for joint segmentation of optic disc and cup

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Glaucoma is a primary cause of irreversible blindness worldwide.
• Accurate segmentation of the optic disc (OD) and optic cup (OC) in fundus images is vital for calculating the cup-to-disc ratio (CDR), a key glaucoma screening indicator.

Purpose of the Study:

• To propose a novel residual multi-scale convolutional neural network with a context semantic extraction module for joint OD and OC segmentation.
• To enhance the accuracy of CDR estimation for large-scale glaucoma screening.

Main Methods:

• A W-shaped backbone network with image pyramid multi-scale input and side output layers for early classification.
• Integration of a context extraction module to capture multi-level receptive field information and recalibrate channel-wise features.
• Joint segmentation of OD and OC to reduce semantic gaps between deep and shallow features.

Main Results:

• The proposed method achieved low overlap errors in OC segmentation (0.0492-0.0684) and OD segmentation (0.1777-0.2547) across four diverse datasets.
• Validation on DRISHTI-GS1, REFUGE, RIM-ONE r3, and a private dataset demonstrated robust performance.
• The method effectively extracts global information and fuses semantic features from multiple levels.

Conclusions:

• The developed residual multi-scale convolutional neural network with context semantic extraction is effective for joint OD and OC segmentation.
• Accurate CDR estimation using this method supports large-scale glaucoma screening.
• The approach shows potential for improving early detection and management of glaucoma.