Abstract
Purpose :
To develop a computational method based on Deep Learning (DL) to automatically detect biomarkers of retinal vein occlusions in images acquired by optical coherence tomography angiography (OCT- A) using retrospective data.
Methods :
Images of the superficial, deep, en face, choriocapillaris and outer retina to choriocapillaris (ORCC) layers obtained from 254 patients attended in an Ophthalmology Clinic were used to train and test an artificial intelligence (AI) model. The OCT-A scans were manually annotated with four biomarkers (BMs): disruption of the perifoveal capillary plexus, non-perfusion areas (NPAs), vascular tortuosity and cystoid spaces. Segmentation and identification were subsequently provided to build and training the DL model using Deep Convolutional Neural Networks (DNN). Detection rate and Jaccard index were the main outcome measures.
Results :
The detection rate of the model for disruption of the perifoveal capillary plexus, non-perfusion areas, vascular tortuosity and cystoid spaces were 93%, 92%, 91% and 84% respectively. The Jaccard index values were 0.85, 0.77, 0.72 and 0.73 respectively.
Conclusions :
The proposed DL model may identify with good performance four key biomarkers of retinal vein occlusions in OCT-A images.
This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.