IIMB-Lite: Lightweight Misbehavior Detection Approach for Insulin Infusion System

Abstract

Wireless-networked body sensors, actuators, and related technologies have innovatively grown to prominence in reliably providing essential support for medical care delivery. The goal of these devices or systems is to improve, if not extended, the quality of life of patients, especially those that are under the most challenging illness. Though each patient group requires unique support, our current work only focuses on medical systems intended for subcultural group of patients suffering from Type 1 Diabetes Mellitus. Nevertheless, our proposed approach introduced in this study is employable to medical systems for another patient group. Insulin infusion system (IIS), also known as artificial pancreas system, has shown effectiveness in glucose control, maintaining blood glucose level of patients in the normal range. However, as more advanced features (e.g., wireless interaction) are innovatively integrated in this system, safety concerns have extended to cybersecurity threats rather than just faulty system functionality [1]. Even more scary, ethical hackers in [2] and [3] proved in laboratory setting the security breaches of some commercially available devices, gaining full control over the medical system. Thus, it is of paramount importance to establish mitigation strategies against security issues, while considering the resource-limited nature of the integrated medical components. Intrusion detection system (IDS), along with other techniques, has showed to be an effective security solution in various fields. Among the representative approaches, namely signature-based, anomaly-based, and specification-based method, the last method is considered as the most suitable technique to this resource-constrained environment. In this talk, we will introduce the adopted framework for strategic formulation of specification-based rules that are particular to IIS, highlight the feasibility of our proposed IDS method, and its edge over the contemporary machine learning-based IDS according to our experimental results.