Retinal vessel segmentation via a Multi-resolution Contextual Network and adversarial learning.

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Ophthalmology

Background:

• Timely diagnosis of retinal diseases is crucial for preventing blindness.
• Accurate retinal vessel segmentation is vital for disease progression monitoring and diagnosis.

Purpose of the Study:

• To propose a novel Multi-resolution Contextual Network (MRC-Net) for accurate retinal vessel segmentation.
• To enhance computer-aided diagnosis of retinal diseases through improved segmentation.

Main Methods:

• Utilized a Multi-resolution Contextual Network (MRC-Net) extracting multi-scale features.
• Employed bi-directional recurrent learning to model feature dependencies.
• Incorporated adversarial training with region-based score optimization for foreground segmentation.

Main Results:

• Achieved superior performance in retinal vessel segmentation across DRIVE, STARE, and CHASE datasets.
• Demonstrated improved Dice and Jaccard index scores.
• Maintained a comparatively low number of trainable parameters.

Conclusions:

• MRC-Net offers a high-performance solution for retinal vessel segmentation.
• The proposed adversarial strategy effectively enhances segmentation accuracy.
• This approach contributes to more timely and affordable computer-aided diagnosis of retinal diseases.