Abstract
Purpose :
To investigate ultrawide-field optical coherence tomography and deep-learning algorithms for visualizing pathologic features in eyes with diabetic retinopathy (DR).
Methods :
We obtained 12x12-mm, 18x18-mm, and ultrawide-field 26x21-mm optical coherence tomography (OCT) and OCT Angiography (OCTA) scans of the central macular region in one eye from each study in healthy and diabetic patients with varying levels of retinopathy using a commercial investigational swept-source OCT system (DREAM OCT; Intalight, Inc). The images demonstrated retinal neovascularization (RNV) (Fig.1. A), nonperfusion area (NPA) (Fig.1. B), and retinal fluid (Fig.1. D). One deep-learning algorithm segmented and quantified NPA from OCTA, and another quantified retinal fluid volumes from structural OCT. We then examined the impact of field-of-view size on visualizing these features and included a comparison of the visibility of RNV with OCTA and fluorescein angiography (FA) (Fig. A, C).
Results :
Ten healthy controls and 10 DR participants were included in the study. Ultrawide-field OCTA enabled the visualization of clinically unsuspected RNV (Fig. 1. A) and detected all RNV revealed by the FA within its field of view. OCTA also detected all intraretinal microvascular abnormalities seen on FA in its field of view (Yellow arrows in Fig.1 A, C). Using a previously developed deep-learning algorithm, 12x12-mm and 18x18-mm scans captured 8.35%±6.20% (mean± standard deviation) and 53.52%±8.93% of the NPA detected in the 26x21 scan, respectively. The fluid volume can be detected by the deep-learning algorithm in ultrawide-field OCT (Fig.1 D), and the segmented fluid regions on cross-sections were verified by a grader.
Conclusions :
Ultrawide-field OCT powered by deep learning algorithms can facilitate the visualization and quantification of retinopathy features in eyes with diabetic retinopathy.
This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.