A QUANTITATIVE SYSTEM TO EVALUATE DIABETIC-RETINOPATHY FROM FUNDUS PHOTOGRAPHS

Purpose. To evaluate a quantitative system to measure the early lesion s of diabetic retinopathy seen in stereoscopic fundus photographs. Met hods. Using a quantitative classification system, photographs of 4657 eyes (7 stereo pairs of 35-mm slides per eye) were scored for 16 diabe tic lesions. A single severity level (identical to the ETDRS Interim S cale) was calculated for each eye. The reliability of this technique, and its reproducibility by independent examiners, was evaluated for in dividual lesions and severity levels using percent agreement, kappa, a nd weighted kappa statistics. Results. This quantitative technique dem onstrated an ''almost perfect'' agreement (weighted kappa greater than or equal to 0.810) on all but one lesion by independent observers. Fo r the severity levels, there was a 95.7% perfect agreement (kappa = 0. 9428). The reproducibility of agreement over time was ''almost perfect '' on all but four lesions; with 88% perfect agreement (kappa = 0.8394 ) for severity levels. Conclusions. When used to evaluate the early le sions of diabetic retinopathy, the Vanderbilt Classification System is highly reliable between graders and over time. This system can gather quantitative data and evaluate incremental changes in an accurate, re producible manner.