Abstract
Purpose :
Manual ETDRS grading of 7-field color fundus photography (CFP) is laborious and time-consuming. It is highly relevant to identify the eyes at risk of developing severe NPDR. We compare the performance of several CFP, optical coherence tomography (OCT), and optical coherence tomography angiography (OCTA) metrics for the automated discrimination of eyes classified as ETDRS 35 or ETDRS 43-47.
Methods :
Imaging data from 93 eyes (66.51±6.97 years),64 with ETDRS 35 and 29 with ETDRS 43-47 acquired during the baseline of the CORDIS cohort study (NCT03696810) were used in this study.OCT and OCTA data were acquired using CIRRUS™ HD-OCT 5000 with AngioPlex® OCTA using the acquisition protocol angiography 3x3mm and the CFPs were obtained using a Topcon TRC 50DX. The performance of 18 metrics was evaluated. Vessel density (VD), perfusion density (PD), and abnormal intercapillary spaces (AIS) were measured based on the OCTA and retinal thickness metrics were measured using the OCT. Associated with AIS, two metrics were measured, the total area of the AIS and the number of abnormal areas (nAIS).All the OCTA metrics were measured in the superficial capillary plexus (SCP), deep capillary plexus (DCP), and full retina (FR). The microaneurysm turnover (MAT) was computed using the RetmarkerDR® software based on the 2-field CFP. Additional CFP acquisitions were used to compute the MAT at a 6 months interval and at a 12 months interval.Logistic regression models were used to discriminate between the ETDRS groups.The performance metric was the area under the receiver operating characteristic (AUC) measured using a 5-folder cross-validation approach. The Mann-Whitney U test was used to study the statistical significance difference between the ETDRS groups.
Results :
The best combination of features, which was found using the forward feature selected method, includes the 6 months MAT, the PD and VD in the DCP, and the nAIS in the DCP and in the FR. With these 5 features, the AUC was equal to 0.82±0.10 (mean±std). The best individual metric was the 6 months MAT (p-value < 0.05) with the value of AUROC equal to 0.79±0.1.
Conclusions :
The results support that metrics obtained from non-invasive imaging methods at the central retina can be used for automated ETDRS classification of eyes with NPDR, identifying the eyes at risk for progression to severe NPDR.
This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.