Novel artificial intelligence-based OCT angiography and OCT metrics provide new insights into diabetic retinopathy: A cross-sectional study

Background Diabetic retinopathy (DR) is a leading cause of vision loss. With advances in optical coherence tomography (OCT) and OCT angiography (OCTA), improved quantification of microvascular and structural damage may enhance the understanding of DR pathology. This study aimed to introduce new OCT and OCTA-derived metrics for the evaluation of DR and to investigate the association between structural and vascular changes. Methods In this cross-sectional study, 89 eyes with DR and 20 control nondiabetic eyes were imaged via 6x6 mm swept-source OCT and OCTA. Eyes were classified into five groups: controls, diabetic with no DR (DNDR), mild, moderate and severe DR. A convolutional neural network segmented capillaries in both the superficial (SVC) and deep vascular complex (DVC), and three OCTA-derived metrics were quantified: vascular index (VI), intercapillary area (ICA), and maximum ischemic point (MIP). The Retinal ischemic perivascular lesion (RIPL) area was measured via structural en face-OCT by using a custom-built analyzer. Group comparisons were performed using ANOVA with post hoc analysis. A t test was used to compare the MIP of eyes with and without RIPL. A p value < 0.05 was considered significant. Results Twenty-seven eyes were included in the DNDR group, 23 in the mild DR group, 20 in the moderate DR group, and 11 in the severe DR group. Across the increasing DR severity groups, the VI decreased, and both the ICA and MIP increased in the SVC and DVC (p < 0.05). The moderate and severe DR groups exhibited greater RIPL areas compared with controls and DNDR (p < 0.01). Eyes with RIPL had higher MIP in the SVC than did eyes without RIPL (p < 0.001). Conclusions Novel OCT and OCTA metrics are correlated with DR severity. MIP is associated with RIPL. Trial registration: NCT03765112 (clinicaltrials.gov)