AVA-Net

AVA-Net performs arterial-venous area segmentation on optical coherence tomography angiography (OCTA) to enable quantitative analysis of vascular perfusion intensity for ophthalmic research and early detection and differentiation of diabetic retinopathy (DR).

Key Features:

• Automated arterial-venous segmentation: Deep learning-based automatic segmentation of arterial and venous areas in OCTA images.
• Arterial Area (AA): The segmented area representing arteries.
• Venous Area (VA): The segmented area representing veins.
• AV Area Ratio (AVAR): A ratio derived from AA and VA indicating vascular distribution.
• Total Perfusion Intensity Density (T-PID): An aggregate measure of perfusion intensity across the vasculature.
• Arterial Perfusion Intensity Density (A-PID): The perfusion intensity specific to arteries.
• Venous Perfusion Intensity Density (V-PID): The perfusion intensity specific to veins.
• Arterial-Venous Perfusion Intensity Ratio (AV-PIDR): A ratio comparing arterial and venous perfusion intensities, reported as sensitive for distinguishing control, NoDR, and mild DR.

Scientific Applications:

• Diabetic retinopathy stratification: Use AA, VA, AVAR, T-PID, A-PID, V-PID, and AV-PIDR to distinguish healthy controls, diabetic patients without DR (NoDR), and patients with mild DR.
• Early diabetic retinopathy assessment: Perfusion parameters such as T-PID and A-PID differentiate mild DR from healthy controls, and combining A-PID with AV-PIDR improves discrimination.
• Quantitative retinal vascular analysis: Provide quantitative measures of arterial and venous areas and perfusion intensity for ophthalmic research and analysis of other ocular vascular conditions.

Methodology:

A deep learning network performs automated arterial-venous area segmentation on OCTA images and extracts AA, VA, AVAR, T-PID, A-PID, V-PID, and AV-PIDR, with statistical comparisons reported using Bonferroni correction.