The Survey of the Effectiveness of Electronic Learning for Early Detection of Retinal Hemorrhages in Diabetic Retinopathy

Introduction: World health organization estimates that it will increase from 130 million patients to more than 350 million patients by next 25 years. Harmful effects of this disease on retinal are named diabetic retinopathy (DR). Consequently the purpose of this paper is with electronic learning and automatic identification of color retinal images for fast and precise identification of spot-shaped red color retinal pathologies is named hemorrhages for identification diabetic retinopathy in its first stages.Materials and Methods: 68000 pixels have been extracted from 85 color retinal images which have formed a learning data set of hemorrhage and non-hemorrhage. That our learning data set sample includes 35000 hemorrhage pixels and 33000 non-hemorrhage pixels, and then selected features have been extracted from processed images by morphological technique and Pixel-level Hemorrhages recognition technique. Finally retinal lesions are classified in two class hemorrhage and non-hemorrhage by a classifier named decision trees. Finally Results this system compared with specialist physicians.Results: After extracting and classifying 68000 pixels from retinal images, in the testing stage this method using formula classifier decision tress achieve 98% sensitivity, 97.14% Specificity features and 97.57% accuracy.Conclusion: therefore, we have solved these problems by computer-aided diagnosis techniques and achieved results show that morphological technique has a high efficiency and is more precise than clinical technique.