Graph deep network for optic disc and optic cup segmentation for glaucoma disease using retinal imaging.

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Accurate segmentation of the optic disc (OD) and optic cup (OC) in fundus images is crucial for detecting eye diseases.
• Existing segmentation methods face challenges in achieving high precision for OD and OC.
• Deep learning approaches show promise but require further refinement for complex feature extraction.

Purpose of the Study:

• To propose a novel three-layer graph-based deep architecture with an enhanced fusion method for accurate OD and OC segmentation.
• To improve the connection of local and global information for more robust segmentation.
• To evaluate the proposed method's performance against established benchmarks.

Main Methods:

• Developed a three-layer graph-based deep architecture incorporating a CNN encoder-decoder.
• Implemented an extended graph network and fusion-based rules to integrate local and global features.
• Utilized feature masking and channel fusion for regularization and improved feature representation.

Main Results:

• Achieved high performance metrics, including Dice Similarity Coefficient (DSC) up to 0.97 for OD and 0.93 for OC.
• Obtained Intersection over Union (IOU) scores of 0.96 for OD and 0.90 for OC on benchmark datasets.
• Demonstrated superior performance compared to existing methods in OD and OC segmentation accuracy.

Conclusions:

• The proposed graph-based deep architecture effectively enhances OD and OC segmentation in fundus images.
• The enhanced fusion method significantly improves the integration of multi-scale features.
• This methodology offers a promising advancement for automated eye disease screening and diagnosis.