PREETHI KOLLURU RAMANAIAH,NIKILA GS,VINOD HC,DEEPAK L
DOI: https://doi.org/Diabetic retinopathy is the major visual impairment of persons with diabetes mellitus and early diagnosis is the key to effective management of the patient. The current work explores the state-of-the-art machine learning-based pipeline of diabetic retinopathy classification of fundus images. It is a model built upon VGG16, Inception V3 and DenseNet 201 to extract discriminative feature of retinal imaging. As clinical datasets are heterogeneous, this study uses the concept of transfer learning which allows a strong fine-tuning of a large image repository that contains the severity range of diabetic retinopathy. Model training is complemented by the data augmentation technique that serves as a two-in-one method of enhancing generalization and reducing overfitting. The evaluation of the performance is based on a complete set of measures, namely accuracy, precision, recall, and F1-score, which will be integrated to decide on the diagnostic performance of each model. The findings show that the proposed methodology provides better accuracy and thus it is appropriate to apply it in a clinical setting as a decision support tool to detect diabetic retinopathy and stratify it by risk to allow timely evidence-based interventions to be implemented before the loss of visual loss occurs.
