Deep Learning-Based Multi-class Classification of Diabetic Retinopathy Utilizing Transfer Learning with MobileNet Architecture

This study examined the performance characteristics of several architectures to establish the finest model for DR diagnosis. After evaluating alternative architectures, the MobileNet architecture was chosen to create the DR diagnostic model due to its better performance. Our findings demonstrate that, while numerous models are capable of DR classification, MobileNet stands out for its superior performance. Therefore, an architecture based on the transfer learning technique is created with the MobileNet as the base model. The results demonstrate the necessity of selecting the appropriate deep learning model for DR medical image classification problem.