TUNet and domain adaptation based learning for joint optic disc and cup segmentation.

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Glaucoma, a leading cause of irreversible blindness, necessitates accurate optic nerve head assessment.
• The optic cup (OC) to optic disc (OD) ratio is crucial for glaucoma diagnosis.
• Current deep learning models struggle with generalizing optic cup and optic disc segmentation across diverse imaging devices and resolutions.

Purpose of the Study:

• To develop a robust domain adaptation framework for accurate optic cup and optic disc segmentation.
• To improve the generalization capability of deep neural networks in medical image segmentation.
• To enhance automated glaucoma screening systems through precise optic nerve head analysis.

Main Methods:

• A transformer-based segmentation network was employed as the primary architecture.
• Domain adaptation techniques were integrated to learn domain-invariant features.
• An auxiliary classifier was introduced to refine segmentation details and improve discrimination.

Main Results:

• The proposed framework demonstrated superior performance in segmenting optic cup and optic disc regions across multiple public datasets (REFUGE, Drishti-GS, RIM-ONE-r3).
• The method effectively mitigated performance degradation caused by domain shift between different scanners and resolutions.
• Experimental results confirmed the framework's ability to capture fine segmentation details.

Conclusions:

• The novel domain adaptation framework significantly improves optic cup and optic disc segmentation accuracy and robustness.
• This approach holds great potential for integration into automated glaucoma screening systems.
• Accurate segmentation of optic nerve head structures is vital for early glaucoma detection and management.