Diabetic Retinopathy Image Analysis Using Deep Learning Techniques

In the whole world, one among the leading factors of blindness in the citizens is the diabetic retinopathy (DR). Detection of DR at an early stage can provide crucial assistance in managing with this disease. Deep learning (DL) has excelled in a variety of domains, particularly in the analysis of biomedical images. The diagnosis of DR is made by scanning retinal fundus images. We have done an empirical evaluation for the automatic detection of DR as part of our research. Convolutional Neural Networks (CNN) approach has been used in this research by using two of its architectures i.e. MobileNetV2 and DenseNet-201. For this research work, we have utilized the Indian Diabetic Retinopathy Dataset (IDRID) dataset, which is an online dataset available containing retinal fundus images of size 4288×2848. The outcomes of this research work shows that the DenseNet-201 model detects DR better than MobileNetV2 with the following metrics: accuracy, recall, precision and fl-score of 87.6%, 97.2%, 82.6%, and 88.4%, respectively are the most dependable findings from the DenseNet-201 model's case testing.