Detection of diabetic retinopathy using computational model of human visual system

Abstract

Background: Diabetic retinopathy is one of the main causes of blindness and the most important complication of diabetes. The accurate analysis of retinal images is important in diagnosing this disease. In this study, a powerful and accurate algorithm for diagnosis of diabetic retinopathy, inspired by the human visual system, is presented based on the rapid sensitivity of the human visual system to intensity, direction and color. Materials and Methods: In this study, DIARETDB1 database containing selected images for diagnosis of diabetic retinopathy has been used. The Matlab R2013a software is used in Windows 7 with 2.5 GHz processor and 4 GHz RAM memory to implement various algorithms for images saliency maps. Results: The proposed method and four models of the existing methods on the DIARETDB1 database were tested, the results of which were based on visual comparison of the results, drawing the ROC curve and calculating the AUC of the models, showed the optimal performance of the proposed algorithm in comparison with other algorithms. The AUC of the proposed method was 0.9012, which in comparison with other methods is the highest value, indicating the proper function of this method in determining the correctness of image saliencies. Conclusion: The positive results from the proposed algorithm, which are based on image processing techniques and inspired by the human visual system, suggest that using this method can help ophthalmologists to diagnose fast, accurate, and reliable diabetic retinopathy.