Simple methods for the lesion detection and severity grading of diabetic retinopathy by image processing and transfer learning

Diabetic retinopathy (DR) has become one of the major causes blindness. Due to increased prevalence diabetes worldwide, diabetic patients exhibit high probabilities developing DR. There is a need develop labor-less computer-aided diagnosis system support clinical diagnosis. Here, we attempted simple methods for severity grading and lesion detection from retinal fundus images. We developed DR by transfer learning with recent convolutional neural network called EfficientNet-B3 publicly available Kaggle Asia Pacific Tele-Ophthalmology Society (APTOS) 2019 training dataset, which includes artificial noise. After removing blurred duplicated images dataset using numerical threshold, trained model achieved specificity sensitivity values ? 0.98 in identification retinas. For grading, classification accuracy 0.84, 0.95, were recorded 1st, 2nd, 3rd predicted labels, respectively. The utility EfficientNets-B3 as well detailed areas referred confirmed via visual explanation networks. Lesion extraction was performed applying an empirically defined threshold value enhanced Although blood vessels red lesions occurred simultaneously, white lesions, including both soft hard exudates, clearly extracted. detected further ground truth DIARETDB1 database general accuracy. easily applicable proposed this study will aid DR, might help selection appropriate treatment strategies