| Popis: |
Diabetic retinopathy (DR) is one of the common eye diseases caused by diabetes, which is seen mainly among adults. Severe DR, if not treated, can result in loss of vision. There are mainly four types of retinopathy conditions—mild, moderate, severe, and proliferative. Once retinopathy reaches the proliferative stage, the person will most likely become blind. In this study, a Neural Architecture Search (NASNet) model was trained to predict diabetic retinopathy from retina images. With the help of the architecture search technique, a best fit architecture for the given application was obtained. To enhance training of the model, a transfer learning approach was used in which pretrained weights were used as the initial weights of the model. The model was trained on 3113 images and 549 images were used for validation. The trained model had an accuracy of 85% and testing accuracy of 82%. |
