Enhancing ECG Secure Authentication and Security of Wearable Implantable Medical Devices Using Martino Homomorphic Encryption with an Optimized Granger Causality Inspired Graph Neural Network

Research on personal authentication security is crucial for both cyber security and privacy. Combining fingerprint and facial recognition have proven to be invaluable for individual authentication. These technologies do have drawbacks, too, such the ability to fake fingerprints and external obstacles. In order to address the issues of forging or faking authentication, many techniques using artificial intelligence have been put forth. Researchers are becoming increasingly interested in user authentication based on electrocardiograms (ECG). In this manuscript, enhancing ECG secure authentication and security of wearable implantable medical devices using Martino Homomorphic Encryption with an Optimized Granger Causality In-spired Graph Neural Network (ECG-MHE-GCIGNN-TTAO) is proposed. Initially, data is taken from ECG-ID Dataset. Then, data is pre-processed using Implicit Bulk Surface Filtering (IBSF) for scaling and de noising. Then, pre-processing data is fed into extraction process using Signed Cumulative Distribution Transform (SCDT). It extracts relevant features such as name, shape, edge and color. Then, extracted features are given into Martino Homomorphic Encryption Algorithm (MHEA) to secure the historical data. Then, historical data are given into Causality-Inspired Graph Neural Network (GCIGNN) to identify the user Authentication based on ECG. The TTAO is implemented to optimize the hyper parameters of GCIGNN. The performance of the suggested ECG-MHE-GCIGNN-TTAO accurately displays the outcomes of Secure Authentication. The efficiency of the proposed method is evaluated utilizing some metrics, like Accuracy, Recall, F1-score, precision, specificity, computational time is analyzed. Performance of the ECG-MHE-GCIGNN-TTAO approach attains 32.34%, 31.45%, and 25.31% higher Accuracy; 32.13%, 34.21%, and 26.27% higher Precision; 22.34%, 21.54%, and 27.56% higher specificity when analyzed with existing methods like ECG authentication Long Short-Term Memory for predictive health monitoring (ECG-LSTM-PHM), ECG biometric authentication based self-supervised learning using Convolutional Neural Network (ECG-SSL-CNN) and Multi-Task Neural Networkdependent Patient Information Hiding for ECG authentication system (MTNN-PIH-ECG) respectively.