SAMCF: Adaptive global style alignment and multi-color spaces fusion for joint optic cup and disc segmentation

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Accurate segmentation of the optic cup (OC) and optic disc (OD) in retinal fundus images is crucial for diagnosing eye diseases.
• Deep learning models for joint OC and OD segmentation face challenges due to dataset variability, brightness interference, and poor edge definition.

Purpose of the Study:

• To propose a novel framework, Style Alignment and Multi-Color Fusion (SAMCF), for robust joint OC and OD segmentation.
• To address domain shift, color representation limitations, and edge perception issues in existing segmentation methods.

Main Methods:

• Developed a domain generalization technique for uniform image styling to mitigate dataset discrepancies.
• Implemented a multi-color space feature extraction and fusion network to enhance color representation and handle brightness variations.
• Incorporated an edge-aware loss function to improve the clarity and smoothness of segmented edges.

Main Results:

• The proposed SAMCF framework achieved superior performance compared to state-of-the-art methods on three public datasets (DGS, RIM, REFUGE).
• SAMCF demonstrated remarkable generalization ability across diverse retinal fundus image datasets.

Conclusions:

• SAMCF effectively addresses key challenges in joint OC and OD segmentation, offering improved accuracy and robustness.
• The framework shows significant potential for clinical applications in eye disease diagnosis through enhanced retinal image analysis.