MODELLING CYBERATTACK VECTORS AND METHODS FOR ENSURING THE CYBER RESILIENCE OF MULTI-LEVEL CONTINUOUS CARDIAC MONITORING SYSTEMS

Abstract

Background. This study examines the key conflict arising from the rapid increase in targeted cyber-physical attacks on ‘Internet of Medical Things’ (IoMT) systems and the limitations of standard signature-based defence systems. Today, multi-layered systems for continuous cardiac monitoring have become a top priority for healthcare organisations; however, attackers are using semantic manipulation and contradictory interference to create covert threats that endanger patients’ lives. Existing security systems do not provide sufficient protection, as standard hard cryptography causes excessive network delays and consumes too many sensor resources, whilst standard neural network-based classifiers remain vulnerable to significant risks due to membership inference attacks. This study presents a hybrid edge intrusion detection system called Edge-IDS, which protects the latent feature space through mathematical validation of its operations. The system combines the structures of a variational autoencoder (VAE) and a deep convolutional neural network (DCNN) with meta-heuristic hyperparameter optimisation based on the Archimedes algorithm (AOA) to achieve accurate anomaly detection. The theoretical basis of the research is founded on a concept that ensures the protection of distributed infrastructures through continuous cyber resilience, rather than through the use of network isolation methods. The system operates using local stochastic gradient descent, which is performed within the hidden layers of the neural network, whilst utilising differential privacy mechanisms (DP-SGD) to build a ‘zero-trust’ architecture at the micro-level. The results of the study were demonstrated by experimental data collected by the researchers. The architecture shows excellent results when tested on benchmark medical telemetry datasets, including various protocols, as well as during simulations of attacks (FGSM) targeting electrocardiographic signals. The micro-level filter implementation proposed in this study significantly reduces the success rate of de-anonymisation attacks, whilst maintaining the system’s ability to generate accurate results based on authentic cardiological data under optimised ultra-low latency conditions and providing protection against cyber threats thanks to a high cyber resilience index (CRI). Conclusions of the study. The theoretical and algorithmic framework yields results that enable a shift from reactive perimeter defence methods to proactive structural cyber resilience in cognitive models. The methodology developed ensures the confidentiality of medical data whilst maintaining the operational capability of cardiac activity prediction systems in life-threatening situations.