MKD-Net: A Novel Neuro Evolutionary Approach for Blockchain-Based Secure Medical Image Classification Using Multi-Kernel DLM

Virendra P. Vishwakarma, Abhay Kumar Yadav

Abstract: In the era of digital healthcare, secure and accurate classification of medical images is of utmost importance, particularly in situations of data imbalance. Owing to advancements in the domain of nature-inspired optimization and human brain simulations, this paper presents a novel approach to a neuro-evolutionary algorithm for medical image classification. In the proposed methodology, we develop a non-iterative learning-based neural network that can handle class imbalance with the help of class-specific regularizations, named multi-kernel deep neural networks (MKD-Net). Multiple kernels are employed in the proposed MKD-Net for different data representations in the classification of heterogeneous data. The MKD-Net parameters are optimized using the Enhanced-JAYA algorithm. ResNet-50, known for its incomparable feature drawing ability, has been used in MKD-Net as an input image feature extractor. Experimental evaluations conducted on benchmark medical image datasets demonstrate that MKD-Net achieves superior classification performance, particularly in imbalanced datasets. Key metrics such as classification accuracy, sensitivity, specificity, precision, and F1-score consistently improve over traditional models. Specifically, the proposed MKD-Net obtained accurate classification on ATT and ALL-IDB1 datasets; 97.44% and 97.46% accuracy on ALL-IDB2 and CoVid datasets respectively. Moreover, the integration with blockchain provide secure feature storage and also ensures data integrity, traceability, and enhanced privacy. By combining advanced feature extraction, optimization, and security mechanisms, MKD-Net sets a new benchmark in medical image classification, showcasing its potential for scalable and secure healthcare applications.