Automated Nonperfusion Quantification in Diabetic Retinopathy on Ultra-Widefield Swept-Source OCT Angiography.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Diabetic retinopathy (DR) is a leading cause of vision loss.
• Accurate assessment of nonperfusion area (NPA) is crucial for DR staging.
• Ultra-widefield swept-source optical coherence tomography angiography (UWF SS-OCTA) provides detailed retinal vasculature imaging.

Purpose of the Study:

• To develop and evaluate a deep learning algorithm for automated NPA segmentation on UWF SS-OCTA images.
• To assess the algorithm's performance across different scan sizes.
• To determine the algorithm's utility in grading diabetic retinopathy severity.

Main Methods:

• A convolutional neural network (CNN) with a multiscale U-Net backbone and attention mechanism was developed for NPA segmentation.
• The algorithm processed en face SS-OCTA images from 6x6 mm, 12x12 mm, and 29x24 mm scan patterns.
• Ground truth was established by expert graders, and segmentation accuracy was measured using the F1 score. Nonperfusion index (NPI) agreement was assessed via Bland-Altman analysis.

Main Results:

• The algorithm achieved high F1 scores for NPA segmentation across all scan sizes (0.82-0.84).
• Algorithm-derived NPI showed strong agreement with expert grading (ICC > 0.979).
• Both human and algorithm-derived NPI increased with DR severity, particularly in larger scan areas.

Conclusions:

• The validated deep learning algorithm accurately segments and quantifies NPAs on UWF SS-OCTA.
• The algorithm demonstrates scalability across various scan sizes.
• This tool has potential for integration into objective biomarker analysis for diabetic retinopathy.