An Efficient Framework Classification for Diabetic Retinopathy Using MobileNetV2

Diabetes is a disorder marked by unregulated glucose levels in the body, potentially causing retinal damage and leading to irreversible visual impairment or loss. Diabetes affects the eyes, resulting in diabetic retinopathy, a prevalent medical disorder, especially among the elderly. Consequently, timely and precise diabetic retinopathy screening is essential for disease prediction. Nonetheless, conventional manual detection is labor-intensive and often results in misdiagnosis for several individuals. In this study, we devised a deep learning (DL) model using transfer learning from the MobileNetV2 architecture, accompanied with an innovative color version preprocessing approach. It decreased the training duration and achieved an average accuracy of 0.93 when applied to the new Kaggle dataset “APTOS 2019 Blindness Detection.” Furthermore, to mitigate the issue of overfitting in the long term, we used Stratified K-fold cross-validation.