A Deep Learning Algorithm for Classifying Diabetic Retinopathy Using Optical Coherence Tomography Angiography.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Diabetic retinopathy (DR) is a leading cause of vision loss.
• Accurate staging of DR is crucial for timely treatment and management.
• Current diagnostic methods can be time-consuming and require specialized expertise.

Purpose of the Study:

• To develop and validate an automated system for diabetic retinopathy staging.
• Utilize optical coherence tomography angiography (OCTA) imaging with convolutional neural networks (CNNs).
• Assess the feasibility and performance of the automated system compared to human experts.

Main Methods:

• A retrospective cross-sectional study involving 1118 eyes with OCTA images (3x3 mm2 and 6x6 mm2).
• A deep CNN and traditional machine learning models were trained on expert-annotated OCTA images.
• CNN performance was compared against traditional machine learning classifiers and two human experts.

Main Results:

• The CNN achieved high accuracy (0.728), sensitivity (0.675), specificity (0.944), F1 score (0.683), and quadratic weighted κ (0.908) for six-level DR staging.
• The CNN significantly outperformed traditional machine learning methods and human experts.
• Larger OCTA image sizes (6x6 mm2) and combined OCTA layers improved CNN performance.

Conclusions:

• CNN-based classification using OCTA images offers a reliable method for DR staging.
• The automated system can assist clinicians in DR classification, improving diagnostic workflow.
• This technology has translational relevance for primary care settings, guiding referrals and clinical decisions.