SDRG-Net: Integrating multi-level color transformation encryption and ICNN-IRDO feature analysis for robust diabetic retinopathy diagnosis

Abstract

The Internet of Medical Things (IoMT) has emerged as a potential solution to various challenges in disease grading, offering enhanced communication between patients and doctors and providing more robust guidance for disease management. Diabetic Retinopathy (DR) grading is crucial for the timely diagnosis and treatment of this common complication of diabetes, which can lead to blindness if left untreated. Existing methods for DR grading often need more accuracy and efficiency due to challenges such as variations in image quality, subtle lesion features, and imbalanced datasets. Furthermore, existing DR grading methods have exhibited lower security properties due to the need for image encryption algorithms in the IoMT environment. A Secure DR Grading Network (SDRG-Net) is proposed to address these issues, integrating several advanced techniques. Firstly, preprocessing techniques are applied to normalize the EyePACS and Messidor datasets and prepare the images for subsequent analysis. Next, Multi Level Color Transformation (MLCT) based image encryption is employed to enhance the robustness and security of the data, ensuring patient privacy while maintaining diagnostic accuracy. The encrypted images are then fed into an Iterative Convolutional Neural Network (ICNN) architecture for feature extraction, leveraging deep learning capabilities to learn discriminative features from the retinal images automatically. This step enables the model to capture intricate patterns and abnormalities indicative of DR. Furthermore precisely, an Improved Red Deer Optimization (IRDO) algorithm is introduced for feature selection, which iteratively refines the feature space to retain the most informative features while discarding redundant or noisy ones. This enhances the efficiency and interpretability of the model, leading to improved performance in DR grading. Finally, a Bagging classifier is employed for classification, leveraging ensemble learning to combine multiple base classifiers trained on different subsets of the data. Finally, the proposed SDRG-Net achieves high performance with an accuracy of 99.65 % on the EyePACS dataset and 99.14 % on the Messidor dataset, demonstrating its robustness and effectiveness in DR grading.