Automated Detection of Diabetic Retinopathy: A Comprehensive Study

With rapidly increasing cases of diabetes mellitus around the globe,it is of serious concern to diagnose the visual impairment related to this condition at early stages. Computer aided automated detection is a potential solution to reduce need of human expert for screening or grading of visual impairment of diabetic patients. The automated detection of diabetic retinopathy involves use of extensive and advanced image processing techniques with accuracy comparable to the human experts. It also holds high significance as it may become a promising tool to reduce burden of medical experts to diagnose the visual impairment due to diabetes. The acquired fundus images undergo various pre-processing, image segmentation, feature extraction and classification algorithms to provide final results in terms of severity level of impairment. This paper aims towards a comprehensive study of existing image processing techniques which can lead to develop an efficient automated classification system. Various popular pre-processing, image segmentation, feature extraction and classification techniques are studied with their respective advantages and disadvantages. The performance comparison is carried out in terms of classification accuracy, sensitivity and specificity. Major findings of study are summarized in concise manner. It is also concluded that the classification accuracy depends on a number of factors like image quality, size of dataset, number of severity levels to be graded, suitability of image processing algorithms with respect to images of concern etc