AUTOMATED DIAGNOSIS OF DME: A COMPARATIVE ANALYSIS

Leakage and accumulation of fluid in macular region of eye causes Diabetic Macular Edema (DME). Early detection of intricate features characterizing DME is very important as it can help restoring the vision of diabetic patients. Automated detection can prove to be a revolutionary tool as it can save a significant amount of time and can reduce the human induced errors of image interpretation. This paper aims towards exploring the potential held by image processing techniques in direction of automated detection of DME. This study discusses the results of various image classifiers on retinal images undergone different preprocessing, image segmentation and classification algorithms. Also, the popular available datasets are enlisted with their respective number of images and resolution details. The performance comparison of image classifiers is done in terms of sensitivity, specificity and accuracy. It is found that the performance of image classifier in order to detect clinical features which show presence of DME in retinal images depends on a number of different dataset based parameters like resolution of image captured, no. of severity grades, size of dataset etc. It is also proposed that to design and develop a robust classification system all performance parameters i.e. accuracy, sensitivity and specificity are equally important and must be achieved without any tradeoff.