Abstract
Purpose :
To propose a predictive model for visual acuity (VA) in eyes with diabetic retinopathy (DR) using optical coherence tomography (OCT) and OCT angiography (OCTA) scans.
Methods :
We acquired 3×3-mm central scans with a commercial OCT system (Avanti RTVue-XR, Optovue, Inc.). Two OCTA parameters, 3-dimensional para-FAZ vessel density (3D-PFVD) and fractal dimension of deep capillary plexus (DCP), and two OCT parameters, central 1-mm fluid volume and ellipsoid zone (EZ) defect area, were derived from the scans (Figure 1). The central macular thickness (CMT) and foveal avascular zone (FAZ) area, available from the commercial software, were also recorded. We applied a linear multivariate regression to correlate macular parameters with VA at baseline and at 12 months. The Mann-Whitney U test evaluated the VA prediction errors (VApredicted -VAactual) of the model created from baseline visits as applied to the baseline vs. 12-month follow up visits.
Results :
This study included 234 eyes with DR of varying severity (Early Treatment of Diabetic Retinopathy Study (ETDRS) letter score range: 40-93), 81 eyes of them had 12-month follow up visits (ETDRS letter score range: 51-92). 3D-PFVD (r=0.610, P<0.001), DCP fractal dimension (r=0.606, P<0.001), EZ defect area (r=-0.574, P<0.001) and central fluid volume (r=-0.536, P<0.001) all achieved better correlations than CMT (r=-0.464, P<0.001) or FAZ area (r=-0.117, P=0.075). The regression model combining both OCTA and OCT parameters had significantly (P<0.001) better VA prediction performance (r=0.75) than that including OCTA parameters only (r=0.65). The model created from baseline visit achieved comparable performance when applied to 12-month follow up visits (P=0.18).
Conclusions :
We have demonstrated OCT- and OCTA-derived parameters that correlate better with VA than CMT and FAZ area. A multivariate regression model combining the parameters created from the baseline visit achieved excellent correlation with VA for both baseline and 12-month follow up visits.
This is a 2021 ARVO Annual Meeting abstract.