Classification of Diabetic Retinopathy stages using Deep Learning Model- DenseNet 121

Diabetic retinopathy (DR) is one of the potentially blinding complications of diabetes and has a significant impact on global blindness and visual disability. Better identification of DR stages and proper Screening for DR and accurate determination of the stage of DR is very important in order to carry out an adequate management. The growth of knowledge in deep learning has tremendously improved the computer aided screening and diagnosis of DR from retinal fundus images. In this study, we employed DenseNet-121 a Convolutional neural network (CNN) model for feature extraction and DR severity level classification with the help of the APTOS dataset. The DenseNet-121 model was fine-tuned using transfer learning, and the model accuracy was tested using fundus images dataset, and a more than eighty-one percent resultant accuracy was obtained for the diagnosis of DR on five clarity levels. The model turned out to be quite generalizable with very little overfitting, as seen by the continuing increments in both training as well as validation loss. Additionally, the model covers predeterminants of DR-related lesions such as microaneurysms, hard exudates, hemorrhages, and cotton wool spots which helped to perform accurate disease staging.