Patch-Based Output Space Adversarial Learning for Joint Optic Disc and Cup Segmentation.

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Glaucoma is a primary cause of irreversible blindness globally.
• Accurate segmentation of the optic disc (OD) and optic cup (OC) in retinal fundus images is crucial for glaucoma screening and diagnosis.
• Existing deep learning models struggle with generalization across different datasets due to domain shift.

Purpose of the Study:

• To develop a robust framework for joint optic disc and optic cup segmentation across diverse fundus image datasets.
• To address the challenge of domain shift in deep learning models for medical image segmentation.

Main Methods:

• Proposed a novel patch-based output space adversarial learning (pOSAL) framework.
• Utilized a lightweight segmentation network as a backbone.
• Incorporated a morphology-aware segmentation loss for accurate and smooth segmentation.
• Employed unsupervised domain adaptation to mitigate domain shift.
• Implemented a patch-based adversarial approach for fine-grained segmentation detail discrimination.

Main Results:

• Demonstrated significant improvement in OD and OC segmentation performance on three public datasets (Drishti-GS, RIM-ONE-r3, REFUGE).
• Achieved first place in the OD and OC segmentation tasks at the MICCAI 2018 Retinal Fundus Glaucoma Challenge.
• The pOSAL framework shows effectiveness in improving segmentation robustness and generalization.

Conclusions:

• The proposed pOSAL framework offers a robust solution for joint OD and OC segmentation, overcoming domain shift limitations.
• This approach enhances the reliability of automated glaucoma screening and diagnosis tools.
• The method shows strong potential for clinical application in ophthalmology.