Abstract
Purpose :
To investigate the application of artificial intelligence (AI) in identifying OCT biomarkers for macula-off RRD, correlating them with final visual outcomes.
Methods :
Retrospective clinical of patients with successful repaired macula-off RRD with single vitrectomy and gas tamponade. OCT volumes were uploaded on an AI-derived platform (Discovery® OCT Biomarker Detector, RetinaAI AG, Switzerland), which performed quantitative measurements of several different retinal layer thicknesses, including outer nuclear layer (ONL) and photoreceptor and retinal pigmented epithelial (PR+RPE), and fluids, such intra-subretinal fluid (IRF, SRF) and biomarker probability detection, including hyperreflective foci (HF). Automated segmentation of retinal layers, volumes and biomarkers probability detection were measured between first visit and last-follow-up visit
Results :
59 patients with macula-off RRD were considered eligible for the study, whose 71% (n=29) were males and 29% (n=17) females. Baseline mean best-corrected visual acuity (BCVA) was 0.5 LogMAR ± 0.1 and last visit it improved to 0.3 LogMAR ± 0.1, (p<0.001). Average thickness analysis showed a significant increase of ONL thickness from first visit to last follow-up visit (from 95.4 µm ± 6.8 to 101.1 µm ± 5.6, p=0.0156) and PR+RPE layer (60.9 µm ± 2.6 to 66.2 µm ± 1.8, p=0.0002. AI-assisted biomarker detection showed mean occurrence rate of HF of 0.11 ± 0.07 at first visit and 0.05 ± 0.05 at last follow-up visit, (p=0.005). Multivariate analysis showed significant correlations between baseline PR+RPE, ONL thicknesses and HF presence with final visual outcomes.
Conclusions :
This study demonstrated the potential of AI-automated OCT segmentation in identifying prognostic biomarkers for visual outcomes after macula-off RRD. Increasing ONL and PR-RPE complex thickness correlated with better final BCVA; by contrast, HF presence at baseline seems to be a novel biomarker predictive for worse visual outcomes in patients with macula-off RRD.
This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.