PREDICTION OF DIABETIC RETINOPATHY SEVERITY USING A COMBINATION OF RETINAL NEURODEGENERATION AND CAPILLARY NONPERFUSION ON OPTICAL COHERENCE TOMOGRAPHY ANGIOGRAPHY

Abstract

Purpose: To generate a prediction model of diabetic retinopathy (DR) severity stages based on retinal neurodegeneration and capillary nonperfusion area (NPA) detected using optical coherence tomography (OCT) and OCT angiography (OCTA). Methods: A total of 155 treatment-naïve eyes were retrospectively included. Macular ganglion cell/inner plexiform layer (mGCIPL) thickness in six macular regions was measured. A custom algorithm was used to calculate capillary NPA from 3 × 3 mm2 and 12 × 12 mm2 field OCTA images. The region of interest was selected as circular areas of 3 mm and 12 mm diameter and divided into six subsections. Classification and regression tree analysis identified the best predictors to discriminate between the five DR stages. Results: Inferotemporal sector showed the largest mean NPA, and the inferior hemispheric NPA was significantly larger compared with the superior hemisphere. The mean mGCIPL thickness was significantly correlated with NPA of 12 × 12 mm2 field in participants with early stage DR. Inferior hemispheric NPA of 12 × 12 mm2 field and mean mGCIPL thickness were the two best variables to discriminate no DR versus mild nonproliferative DR (NPDR) and mild versus moderate NPDR (accuracy: 88.8% and 93.5%). Meanwhile, a combination of NPA of 12 × 12 mm2 and 3 × 3 mm2 fields was the best prediction model to discriminate moderate versus severe NPDR and severe NPDR versus PDR (accuracy: 91.8% and 94.1%). Conclusion: A combination model of capillary NPA and mGCIPL thickness may be a novel biomarker for predicting DR severity. Capillary nonperfusion seems to initially occur in the midperipheral retina with macular neurodegeneration and progress posteriorly.