Effect of algorithms and covariates in glaucoma diagnosis with optical coherence tomography angiography

To assess the effects of algorithms and covariates in glaucoma diagnosis with optical coherence tomography angiography (OCTA).In this prospective cross-sectional study, one eye each of 36 normal controls and 64 patients with glaucoma underwent 4.5 mm disc-centred and 6 mm macula-centred OCTA scans. The peripapillary nerve fibre layer plexus capillary density (NFLP-CD) and macular superficial vascular complex vessel density (SVC-VD) were measured using both a commercial algorithm (AngioAnalytics) and a custom algorithm (Center for Ophthalmic Optics & Lasers Angiography Reading Toolkit (COOL-ART)). The nerve fibre layer and ganglion cell complex thicknesses were measured on structural OCT.The overall peripapillary NFLP-CD and macular SVC-VD measured with the two algorithms were highly correlated but poorly agreed. Among the normal controls, the perfusion measurements made by both algorithms were significantly correlated with age. AngioAnalytics measurements were also correlated with signal strength index, while COOL-ART measurements were not. These covariates were adjusted. The diagnostic accuracy, measured as the area under the receiver operating characteristic curve for glaucoma detection, was not significantly different between algorithms, between structural and perfusion parameters and between the peripapillary and macular regions (All p>0.05). The macular SVC-VD in the 6 mm square had a significantly higher diagnostic accuracy than that of the central 3 mm square area (p=0.005).AngioAnalytics and COOL-ART vessel density measurements are not interchangeable but potentially interconvertible. Age and signal strength are significant covariates that need to be considered. Both algorithms and both peripapillary and macular perfusion parameters have similarly good diagnostic accuracy comparable to structural OCT. A larger macular analytic area provides higher diagnostic accuracy.