Comparison of diagnosis of early retinal lesions of diabetic retinopathy between a computer system and human experts

Objective: To investigate whether a computer vision system is comparable wi th humans in detecting early retinal lesions of diabetic retinopathy using color fundus photographs. Methods: A computer system has been developed using image processing and pa ttern recognition techniques to detect early lesions of diabetic retinopath y (hemorrhages and microaneurysms, hard exudates, and cotton-wool spots). C olor fundus photographs obtained from American Indians in Oklahoma were use d in developing and testing the system. A set of 369 color fundus slides we re used to train the computer system using 3 diagnostic categories: lesions present, questionable, or absent (Y/Q/N). A different set of 428 slides we re used to test and evaluate the system, and its diagnostic results were co mpared with those of 2 human experts-the grader at the University of Wiscon sin Fundus Photograph Reading Center (Madison) and a general ophthalmologis t. The experiments included comparisons using 3 (Y/Q/N) and 2 diagnostic ca tegories (Y/N) (questionable cases excluded in the latter). Results: In the training phase, the agreement rates, sensitivity, and speci ficity in detecting the 3 lesions between the retinal specialist and the co mputer system were all above 90%. The kappa statistics were high (0.75-0.97 ), indicating excellent agreement between the specialist and the computer s ystem. In the testing phase, the results obtained between the computer syst em and human experts were consistent with those of the training phase, and they were comparable with those between the human experts. Conclusions: The performance of the computer vision system in diagnosing ea rly retinal lesions was comparable with that of human experts. Therefore, t his mobile, electronically easily accessible, and noninvasive computer syst em, could become a mass screening tool and a clinical aid in diagnosing ear ly lesions of diabetic retinopathy.