A Computational Intelligence-Based Approach for Detection of Diabetic Retinopathy Using Residual Recurrent Neural Network

This research work aims to detect diabetics in their early stages using retinopathy image classification, to help patients take treatment. A mechanism of hybrid attention, paired with a Residual Recurrent Neural Network (RRNN) method is presented to address the issue of extracting small focus while classification and increasing the classification accuracy of Diabetic Retinopathy (DR) images. To derive the more fundamental properties of pictures, a multi-scale hybrid attentiveness-oriented Deep Learning (DL) network approach is initially developed. The method used for sampling equalizes various sample formats, which enhances the spatial and channel attention of the retrieved characteristics. The network model performance is optimized using a loss function, a small-step training technique, and a beginning set of variables. The five classifications are evaluated using a classifier based on a multi-scale hybrid attention network. According to experimental findings, the suggested approach can efficiently increase the classification efficiency of the DR by learning more features of small targets. Comparing the classification accuracy of the method to various current classification models, an in-depth evaluation of Kaggle's freely and openly accessible dataset of DR revealed a classification accuracy of 93.58%. The recognition of the characteristics of tiny spots is essential for the categorization diagnosis of DR, and it is challenging to retrieve the characteristics of too-tiny spots using conventional extraction techniques combined RRNN model proposed.