Dual Encoder-Based Dynamic-Channel Graph Convolutional Network With Edge Enhancement for Retinal Vessel Segmentation.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Deep learning is vital for retinal vessel segmentation in diagnosing fundus diseases.
• Current deep learning methods struggle with fine blood vessels due to lost edge information during down-sampling.
• Existing approaches inadequately capture channel characterization by ignoring dynamic topological correlations in feature maps.

Purpose of the Study:

• To develop an advanced deep learning model for improved retinal vessel segmentation.
• To address limitations in preserving edge information and utilizing channel features in current methods.
• To enhance the accuracy of segmenting fine blood vessels for better clinical diagnosis.

Main Methods:

• Proposed a novel dual encoder-based dynamic-channel graph convolutional network with edge enhancement (DE-DCGCN-EE).
• Implemented an edge detection-based dual encoder to preserve vessel edge information during down-sampling.
• Utilized a dynamic-channel graph convolutional network for topological mapping and feature synthesis across channels.
• Incorporated an edge enhancement block to fuse edge and spatial features.

Main Results:

• The DE-DCGCN-EE model demonstrated superior performance in retinal vessel segmentation across five datasets.
• Achieved more remarkable segmentation results compared to existing state-of-the-art methods.
• Effectively preserved edge information and improved the utilization of channel features.

Conclusions:

• The proposed DE-DCGCN-EE model significantly enhances retinal vessel segmentation accuracy, particularly for fine vessels.
• The method shows potential for clinical application in diagnosing fundus diseases.
• The novel architecture effectively addresses limitations of previous deep learning approaches in this domain.