Background: To explore the value of Optical Coherence Tomography Angiography (OCTA) metrics in the early diagnosis of vascular complications in diabetes.
Methods: All participants underwent OCTA with a swept-source OCT device. Automated measurements of the foveal avascular zone (FAZ) area, vessel density (VD), and blood flow density (BFD) of both 3 × 3 and 6 × 6 windows were then obtained after a quality check.
Results: Diagnostic models based on multiple risk factors were developed separately for diabetic retinopathy and carotid atherosclerosis using random forest and multivariate logistic regression methods. The addition of specific OCTA metrics improved the diagnostic prediction of DR compared with the models of risk factors alone (Inner Retinal Blood Flow Density in 3 × 3 window, IRBFD; Brier score 0.124 vs. 0.149; AUC, 0.887 vs. 0.836) (Central Retinal Blood Flow Density in 3 × 3 window, CRBFD; Brier score 0.142 vs. 0.149; AUC, 0.851 vs. 0.836). Adding diabetic peripheral vascular disease (DPVD) indicator improved the prediction of carotid atherosclerosis (Brier score, 0.180 vs. 0.192; AUC, 0.802 vs. 0.781. The FAZ in the 3 × 3 window also achieved this effect when targeting only T2DM patients (AUC, 0.797 vs. 0.766; Brier score, 0.183 vs. 0.195).
Conclusions: Focusing on IRBFD and CRBFD in the 3 × 3 window of OCTA allows for a more sensitive prediction of the occurrence of DR in diabetic patients. Meanwhile, the quantitative microvascular information provided by OCTA and the occurrence of DPVD may be crucial for diagnosing carotid atherosclerosis. For T2DM patients, we also propose the possibility of FAZ in the 3 × 3 window as a potential diagnostic indicator.
Keywords: Carotid atherosclerosis; Diabetic retinopathy; Machine learning; OCTA; Prediction model.
Copyright © 2023 Elsevier B.V. All rights reserved.