Learning Algorithms using Optimization Techniques for Detection of Diabetic Retinopathy

Abstract

Diabetic Retinopathy (DR) damages retinal blood vessels due to elevated blood newlineglucose levels. Manual examination of retinal images is labor-intensive and timeconsuming. newlineTo address this, we propose an automated approach utilizing the Hybrid newlineHorse-Herd-based Convolutional ResNet Segmentation Framework (HHbCRSF) to detect newlinediabetes-induced changes in retinal images. Our method demonstrates a sensitivity newlineof 90%, a specificity of 99.7%, and an error rate of 0.25, outperforming existing techniques. newlineAdditionally, we have developed a novel approach called Squirrel Search-based Extreme newlineBoosting (SSbEB) for the segmentation and severity assessment of diseased retinal newlineimages. This method integrates squirrel-inspired optimization into extreme boosting, newlineresulting in superior feature analysis and segmentation performance. The algorithm newlineis noted for its simplicity, high efficiency, robustness, and adaptability to various data newlinetypes without requiring significant processing power. Our results show an accuracy of newline94%, a specificity of 98%, and a sensitivity of 96.4%. This proposed method achieves newlinea higher recognition rate compared to existing methodologies. newlineA Meta Learn Enhanced Recurrent Neural Network (ML-ERNN) model was developed newlinefor retinopathy image classification. It combines APTOS and Kaggle datasets newlineto reduce overfitting, lower computational demands, and improve interpretability. The newlinemodel achieved high accuracy, with 99.96% on the Kaggle dataset and 99.93% on the newlineAPTOS dataset. This model will be useful for identifying patterns associated with patients newlineat higher risk of developing severe diabetic retinopathy. newline