Multi-Task Segmentation and Classification Network for Artery/Vein Classification in Retina Fundus.

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Accurate automatic classification of arteries and veins (A/V) in fundus images is crucial for detecting vascular abnormalities and diagnosing systemic diseases.
• Challenges in A/V classification include variability in vessel structures and subtle distinctions between arteries and veins.

Purpose of the Study:

• To propose a novel Multi-task Segmentation and Classification Network (MSC-Net) to enhance A/V classification accuracy in fundus images.
• To address limitations in current A/V classification methods by utilizing extracted vessel features.

Main Methods:

• Developed MSC-Net, incorporating three key modules: Multi-scale Vessel Extraction (MVE), Multi-structure A/V Extraction (MAE), and Multi-source Feature Integration (MFI).
• MVE module identifies vessel pixels using semantic information.
• MAE module classifies A/V by combining original images with MVE-extracted features.
• MFI module integrates outputs from MVE and MAE for final A/V classification.

Main Results:

• MSC-Net demonstrated high performance in retinal A/V classification.
• The proposed method outperformed state-of-the-art methods on multiple public datasets.
• Empirical experiments validated the effectiveness of MSC-Net in improving A/V classification accuracy.

Conclusions:

• MSC-Net effectively enhances automatic artery and vein classification in fundus images.
• The novel network architecture addresses existing challenges in A/V classification.
• MSC-Net shows significant potential for clinical applications in diagnosing vascular-related diseases.