TrMAps: Trust Management in Specification-Based Misbehavior Detection System for IMD-Enabled Artificial Pancreas System
Philip Virgil Astillo, Gaurav Choudhary, Daniel Gerbi Duguma, Jiyoon Kim, Ilsun You
Abstract
Advances of implantable medical devices (IMD) are transforming the traditional method of providing medical treatment, especially those patients under the most challenging condition. Accordingly, the IMD-enabled artificial pancreas system (APS) has now reached global market. It helped many patients suffering from chronic disease, called diabetes mellitus, in monitoring and maintaining blood glucose level conveniently. However, this advancement is accompanied by various security threats that place the life of patients at risk. Hence, protective measures, especially against yet unknown threats, are of paramount importance. This paper proposes a specification-based misbehavior detection system (SMDS) as an alternative solution to effectively mitigate security threats. Moreover, an outlier detection algorithm is also introduced to validate integrity of unprotected data transmitted by the different components. The monitor agent applies a smoothened-trust-based scheme to assess the trustworthiness of the APS. To demonstrate effectiveness of the proposed method, we first extend the UVA/Padova simulator for glucose-insulin data collection and subsequently simulate scenario with well-behave and malicious APS in MATLAB. The results show that there exists an optimal trust threshold that can achieve high specificity and sensitivity rate. Moreover, the proposed technique was compared to contemporary machine learning classifier including decision tree, support vector machine, k-nearest neighbor, and the SMDS called SMDAps. It is shown that our approach can dominate detection performance, especially to malicious behavior that manifests habitually (hidden mode).