Abstract
Purpose :
Optical coherence tomography angiography (OCTA) has gained popularity since it can delineate retinal vessels without a contrast dye or the risk of allergic complications. However, a major drawback of OCTA is its low detection rate of microaneurysms (MAs) compared to fluorescein angiography (FA). The identification of MAs is important in the therapeutic planning of diabetic retinopathy (DR). In this study, we evaluated the ability of a new OCTA-based artificial intelligence (AI)-inferred FA system (Murata T et al. Biomed Opt Express. 2023) to detect MAs in patients with DR.
Methods :
We obtained OCTA (OCT-S1, Canon, Tokyo, Japan) images (6 mm × 6 mm) and FA images from 56 eyes of 30 patients with DR and constructed a dataset of approximately 17,000 image pairs. We also designed a GAN architecture to generate FA-like (AI-inferred FA) images from OCTA images at various imaging times and trained it using the dataset. MAs detected by FA were considered positive results and were compared with those detected by OCTA and AI-inferred FA (Figure 1).
Results :
Eighteen eyes from 10 patients (62 ± 17 years; 4 female, 6 male) were finally evaluated. A total of 1,053 MAs were identified using FA images. OCTA images revealed 357 MAs (true positive, 343; false positive, 14). In contrast, AI-inferred FA images identified
737 MAs (true positive, 715; false positive, 22). Therefore, OCTA and AI-inferred FA accurately identified 33.9% and 67.9% of MAs, respectively. Comparing the images case-by-case, AI-inferred FA detected significantly more true positive MAs than OCTA (39.7 ± 23.0 vs. 19.1 ± 35.2; P < 0.01).
Conclusions :
OCTA-based AI-inferred FA was more accurate than OCTA at detecting a much larger number of MAs. AI-inferred FA could be valuable in treating DR because, unlike FA, it can delineate MAs noninvasively.
This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.