An efficient optimized deep learning model for diabetic retinopathy classification

<p>One of the most common disorders worldwide is diabetes, a metabolic condition marked by elevated blood sugar levels. Complications from diabetes can lead to diabetic retinopathy (DR). Later stages of DR can result in blindness, while early stages may only produce slight vision abnormalities or no symptoms at all. Diagnosing diabetic retina (DR) is particularly challenging due to changes in the retina that occur with the stages of the illness. An autonomous DR early detection device can help ophthalmologists with DR screening while also protecting a patient’s vision. The Ensemble of EfficientNet-B0, a unique approach based on the Modified Sparrow Search Algorithm (EMSSA) that provides more accurate classification with less processing time, is presented in this study. The proposed EMSSA regularized classification is carried out after the images have been pre-processed, segmented, and dimension-reduced features constructed using the suggested algorithms. Five phases of non-proliferative images were used in the experiment: Proliferative, moderate, mild, severe, and non-proliferative. By using dimensionally reduced data, the suggested approach reduces complexity and produces an accuracy rate of 98.8%. According to an examination of performance metrics, the system performs better than other cutting-edge methods in terms of F-measures, accuracy, recall, and precision.</p>