Multi-Plexus Nonperfusion Area Segmentation in Widefield OCT Angiography Using a Deep Convolutional Neural Network.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Diabetic retinopathy (DR) diagnosis relies on identifying nonperfusion areas (NPAs) in retinal vascular plexuses.
• Widefield optical coherence tomography angiography (OCTA) provides detailed imaging of these plexuses.
• Accurate segmentation of NPAs is crucial for assessing DR severity and guiding treatment.

Purpose of the Study:

• To develop and validate a convolutional neural network (CNN) for segmenting NPAs in multiple retinal vascular plexuses using widefield OCTA.
• To assess the CNN's ability to differentiate between signal reduction artifacts and true flow deficits.

Main Methods:

• A deep CNN with a parallel U-Net module was trained on widefield OCTA scans from 202 participants with DR and 39 healthy controls.
• Images were acquired using a commercial 70-kHz OCT system and processed to generate widefield montages.
• Expert manual segmentation served as the ground truth, with sixfold cross-validation used for evaluation.

Main Results:

• The CNN achieved high agreement with manual segmentation for NPA detection across superficial, intermediate, and deep capillary plexuses (F-scores: 0.84-0.87).
• The deep capillary plexus demonstrated high sensitivity in differentiating diabetic eyes from healthy controls and various DR severity levels.
• Combined three-plexus analysis showed the best performance in distinguishing vision-threatening DR from non-vision-threatening DR (81.0%).

Conclusions:

• A deep learning network can accurately segment NPAs in individual retinal vascular plexuses on widefield OCTA.
• This automated segmentation approach shows potential to significantly improve the diagnostic accuracy of diabetic retinopathy.