An Efficient Hierarchical Optic Disc and Cup Segmentation Network Combined with Multi-task Learning and Adversarial...

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 diagnosing ocular pathologies.
• Challenges include complex retinal structures like blood vessels and lesions, complicating segmentation tasks.
• Convolutional neural network (CNN)-based methods show promise in fundus image analysis.

Purpose of the Study:

• To propose a robust and accurate cascaded two-stage network for OD and OC segmentation.
• To enhance the reliability and precision of automated segmentation in fundus images.
• To improve computer-aided diagnosis of ocular diseases through improved segmentation.

Main Methods:

• A U-Net like framework with an attention mechanism and focal loss for initial OD detection.
• A second-stage refined segmentation network integrating multi-task learning (contours, distance maps) and adversarial learning.
• Utilizing public datasets (RIM-ONE-r3, REFUGE) for evaluation.

Main Results:

• The proposed cascaded network achieves robust and accurate segmentation of OD and OC.
• Multi-task learning ensures segmentation smoothness and shape integrity.
• Adversarial learning promotes spatial and shape consistency with ground truth.

Conclusions:

• The developed cascaded two-stage network offers competitive performance against state-of-the-art methods.
• The approach demonstrates significant potential for advancing automated ocular disease diagnosis.
• The method provides accurate segmentation crucial for clinical applications.