Artery vein classification in fundus images using serially connected U-Nets.

Area of Science:

• Medical imaging analysis
• Deep learning for ophthalmology
• Computational pathology

Background:

• Retinal vessels are crucial biomarkers for diagnosing systemic and ocular diseases.
• Manual segmentation of retinal images is time-consuming and impractical.
• Distinguishing between retinal arteries and veins is essential for accurate analysis.

Purpose of the Study:

• To develop an automated method for simultaneous retinal vessel segmentation and artery-vein classification.
• To evaluate the performance of a novel convolutional neural network architecture.
• To compare the proposed method against state-of-the-art techniques and human experts.

Main Methods:

• A convolutional neural network utilizing serially connected U-nets was proposed.
• The network was trained and validated on public (DRIVE, HRF) and proprietary datasets.
• Ablation experiments were conducted to assess the contribution of key network components.

Main Results:

• Achieved F1 scores of 0.829 (DRIVE) and 0.814 (HRF) for vessel segmentation.
• Reached F1 scores of 0.952 (DRIVE) and 0.966 (HRF) for artery-vein classification.
• Outperformed state-of-the-art methods on the HRF dataset and human experts on the DRIVE dataset for classification.

Conclusions:

• The proposed method demonstrates competitive and superior performance in vessel segmentation and classification.
• The architecture effectively segments vasculature and classifies vessels, even in pathological images.
• Serial connection of U-nets significantly enhances artery-vein classification accuracy.