An Automatic Identification of Diabetic Macular Edema Using Transfer Learning

Abstract

The growth of fluid in the macular region is an area in the retina center, which causes Diabetic Macular Edema (DME). The early stages it is classified by extracting features from retinal images to grade DME. This paper acquaints a transfer learning approach for DME treatment from computerized fundus images and correctly grouping its reality. This paper introducing a Convolutional Neural Network (CNN) to deal with diagnosing DME from digital fundus images and precisely classifying its seriousness, and building a system with DenseNet architecture, which can convolute features engaged in the task of classification. This model is trained by utilizing an openly accessible Messidor and IDRiD dataset and shows the amazing outcome, especially for a significant level order task. The proposed model achieves 96.78% accuracy on Messidor and 96.51% on IDRiD datasets.