Mitigating cyber-intrusions in medical devices with agent-based self-healing

Medical device security has become a critical focus in the healthcare sector, as increasing connectivity introduces challenges related to patient safety, data confidentiality, and system reliability. To address these concerns, various strategies have been developed, including risk identification, mitigation techniques, and autonomic recovery mechanisms. In this paper, we propose a novel conceptual framework that leverages reinforcement learning for self-healing in implanted medical devices (IMDs). This approach integrates automated recovery actions with real-time risk identification, providing a robust mechanism to maintain system functionality and safeguard patient well-being in the face of adversarial threats. By using a behavioral abstraction model of an insulin pump as a case study, our framework demonstrates the ability to maintain continuous system functionality under a variety of attack scenarios, achieving the maximum simulated survival time of 20,165 minutes for all cases. In comparison, without the self-healing mechanism, survival times drop significantly, particularly under attacks on critical components, such as glucose sensors and meters. These results highlight the effectiveness of the proposed approach in mitigating the impact of system failures and ensuring reliable operation of IMDs in adversarial environments.