MEAL: Meta enhanced Entropy-driven Adversarial Learning for Optic Disc and Cup Segmentation.

Area of Science:

• Ophthalmology
• Medical Image Analysis
• Computer Vision

Background:

• Accurate segmentation of the optic disc (OD) and optic cup (OC) is crucial for glaucoma diagnosis.
• Domain shift in cross-domain data negatively impacts the performance of segmentation models on new datasets.

Purpose of the Study:

• To propose a novel domain adaptation model, Meta enhanced Entropy-driven Adversarial Learning (MEAL), for robust OD and OC segmentation.
• To enhance model adaptability and feature representation for improved segmentation accuracy across different datasets.

Main Methods:

• The MEAL model incorporates a meta-enhanced block (MEB) for high-level feature adaptability.
• An attention-based multi-feature fusion (AMF) module integrates multi-level features.
• An entropy map-driven adversarial loss function is employed for domain adaptation.

Main Results:

• The proposed MEAL model demonstrates effective domain adaptation for OD and OC segmentation.
• Evaluations on public fundus image datasets show superior performance compared to existing domain adaptation methods.
• Ablation studies confirm the effectiveness of the MEB and AMF modules.

Conclusions:

• The MEAL model offers a promising solution for overcoming domain shift in medical image segmentation.
• This approach can significantly improve the reliability and efficiency of glaucoma diagnosis through automated image analysis.