Deep learning for artery-vein classification in optical coherence tomography angiography

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Retinopathies differentially affect retinal arteries and veins.
• Traditional fundus photography lacks resolution for capillary details.
• Optical coherence tomography angiography (OCTA) offers enhanced visualization of retinal vasculature at the capillary level.

Purpose of the Study:

• To review traditional and deep learning (DL) methods for artery-vein (AV) classification in OCTA.
• To highlight the importance of AV classification for retinopathy staging and diagnosis.
• To discuss the potential of DL to automate and simplify AV classification for clinical use.

Main Methods:

• Summarizing traditional AV classification techniques.
• Reviewing recent DL approaches for AV classification in OCTA.
• Discussing interpretability methods for DL models in this context.

Main Results:

• Current AV classification methods often require multiple imaging modalities and complex algorithms.
• DL algorithms show promise in automating AV classification with reduced computational complexity.
• Differential AV analysis in OCTA can improve sensitivity for retinopathy staging.

Conclusions:

• Automated AV classification using DL in OCTA is crucial for efficient retinopathy diagnosis.
• DL can overcome the limitations of traditional methods, facilitating clinical deployment.
• Further research into DL interpretability is needed for reliable clinical application.