Hybrid optimization-based quantum-driven multi-relational graph attention networks for enhanced cyber attack detection in medical IoT networks

The swift implementation of Internet of Medical Things (IoMT) devices has reshaped the way healthcare is provided. Along with the benefits, the use of IoMT devices has introduced new risks stemming from advanced cyberattacks. The conventional intrusion detection systems (IDSs) have numerous issues, including scalability problems, features that are irrelevant and redundant, and a poor response to evolving attack patterns. All these issues contribute to poor real-time detection performance. To deal with these issues, this paper presents a Hybrid Optimization–Based Quantum–Driven Multi–Relational Graph Attention Network for IoMT Security (MREQGAN-DSA) for IoMT environments. This model addresses the problems arising in IoMT environments for IoMT security and integrates the following three novel techniques: Cat Hunting Optimization with Prism Refraction Search (CHO–PRS) for the refined feature selection, Multi-Relational Equivariant Quantum Graph Attention Network (MREQGAN) for attack classification aware of symmetries, and Duck Swarm Algorithm (DSA) for the adaptive dynamic optimization of hyperparameters. The model is evaluated on NSL-KDD and WUSTL-EHMS-2020 benchmark datasets and obtains an average accuracy of 99.81% on the NSL-KDD dataset and 99.77% on the WUSTL-EHMS-2020 dataset, thereby greatly enhancing the detection metrics of well-known precision, F1-score, and recall, along with the existing literature of IoMT intrusion detection techniques. This study discusses the implementation of advanced quantum-inspired learning and bio-inspired optimization in the MREQGAN-DSA framework to address the IoMT security problem. Thus, the framework is presented as a foundation for smart healthcare systems that are both secure and efficient.