VTG-Net: A CNN Based Vessel Topology Graph Network for Retinal Artery/Vein Classification

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Accurate retinal artery/vein (A/V) classification is crucial for diagnosing cardiovascular and diabetic diseases.
• Existing methods, including graph-based and convolutional neural network (CNN) approaches, have limitations in utilizing hierarchical features or incorporating vessel topology.
• These limitations hinder the effectiveness of current A/V classification techniques.

Purpose of the Study:

• To propose a novel CNN-based framework, VTG-Net (vessel topology graph network), for improved retinal A/V classification.
• To integrate vessel topology information with deep hierarchical features extracted by CNNs.
• To enhance the accuracy of retinal A/V classification by leveraging both image and topological data.

Main Methods:

• Developed VTG-Net, a CNN-based framework incorporating vessel topology information for retinal A/V classification.
• Transformed CNN-extracted vessel features into a graph representation that preserves vessel topology.
• Utilized a graph convolutional network (GCN) to learn CNN features and vessel topological features simultaneously.
• Fused CNN and GCN outputs for final A/V classification prediction.

Main Results:

• VTG-Net demonstrated high performance in retinal A/V classification on both the publicly available AV-DRIVE dataset and an in-house dataset.
• The proposed method achieved approximately a 2% improvement in accuracy on the AV-DRIVE dataset compared to state-of-the-art methods.
• The integration of vessel topology information proved effective in enhancing classification accuracy.

Conclusions:

• VTG-Net effectively improves retinal artery/vein classification accuracy by incorporating vessel topology.
• The framework successfully combines deep hierarchical features from CNNs with topological information using GCNs.
• VTG-Net represents a significant advancement over existing methods for retinal A/V classification, aiding in disease diagnosis.