Abstract: Objective To automatically identify relevant diabetic retinopathy(DR) characteristic lesions in the eyes using computer vision technology, develop computer automated screening system which can be used to screen for DR. Methods Using mathematical morphology and support vector machine(SVM) classification technology, we designed algorithm to screen for DR, including detecting hemorrhage, exudation, microaneurysm and other types of lesions, Then automatic grading and diagnosis of fundus imaging based on clinical diagnostic criteria for DR, and accomplish automatic DR screening. Results The established completely automated screening system was used for international Messidor databases as screening judgment, comparing with the validation criteria for diagnosis of experts, in a total of 1 200 fundus images, the system sensitivity was determined to be 93.8%, the specificity was 94.5%, detection time was 9.83 s. Conclusion These results show that the completion of the development of computer vision algorithms based DR automated screening system can accurately and efficiently screen ophthalmic images, with better prospects for clinical application and social benefits.

Key words: diabetic retinopathy, computer vision, fundus imaging, automatic screening