Multicenter Assessment of OCT and OCT-A Measurement Reproducibility Using AI-Based Analysis in Healthy Eyes

PURPOSE To assess repeatability and reproducibility of retinal measurements from spectral-domain optical coherence tomography (SD-OCT) and optical coherence tomography-angiography (OCT-A) using artificial intelligence tools Discovery® and CapillaryX® in healthy individuals in two centers. METHODS Twenty-three healthy participants underwent OCT and OCT-A examination at two time points in two centers. Retinal nerve fiber layer (RNFL) thickness, ganglion cell layer plus inner plexiform layer (GCL + IPL), and inner nuclear layer plus outer plexiform layer (INL + OPL) were automatically extracted from 3 x 3 mm OCT scans using Discovery® and Cohort Builder software’s. Foveal avascular zone (FAZ) in superficial (SCP) and deep capillary plexus (DCP), vessel density (VD) and vascular parameters (bifurcation, diameter and tortuosity in arteries and veins) were derived from OCT-A scans using CapillaryX®. Inter-and intraclass correlation coefficients were calculated. RESULTS Among the performed measurements, all retinal layer thicknesses showed good repeatability and reproducibility, except for central RNFL. FAZ area and perimeter in SCP also showed good repeatability and reproducibility. Repeatability and reproducibility of vessel density was moderate to good, bifurcation and tortuosity were moderate, and vessel diameter was poor. CONCLUSIONS Most retinal layer thickness measurements with SD-OCT are repeatable and reproducible. OCT-A vascular measurements show variable reproducibility. FAZ metrics in the SCP and VD yield promising results.