Data-Driven Detection of Sensor-Hijacking Attacks on Electrocardiogram Sensors

Abstract

In this chapter, we build a detector for identifying sensor-hijacking attacks that alter sensor readings in a Wearable Medical Internet of Things (WMIoT). A sensor-hijacking attack targets medical devices in a WMIoT and makes them generate arbitrary user health state information. A sensor-hijacking attack is very dangerous because it can be mounted stealthily on unsuspecting WMIoT users. We focus on sensor-hijacking attack on one of the most commonly collected vital signs from a person, the electrocardiogram (ECG) sensor. Using sensor-hijacking attack to alter ECG measurements can have profound consequences for the user, as an adversary can easily make a person who is experiencing cardiac arrhythmia appear to be normal and thus cause immediate or long-term harm to their health. To detect sensor-hijacking-based alterations of the ECG measurements, our approach leverages the idea that multiple physiological signals based on the same underlying physiological process (e.g., cardiac process) are inherently related to each other, i.e., have common features. Any surreptitious alteration of one of the signals will not be reflected in the other reference signal (s) in the group. We describe an ECG alteration detector that uses arterial blood pressure (ABP) measurements as reference. The advantages of using a distinct reference signals are: (1) It does not require redundant ECG sensors to operate, (2) it adapts to the changing physiology of the user and does not depend on historical trends, and (3) it does not require instrumentation on other hardware modifications of the devices to work. In this work, we describe a temporal ECG alteration detector. Analysis of our detector shows promising results with over $\sim$97% accuracy in detecting even subtle ECG alterations for both healthy users and those with cardiac conditions, within 5 s of the occurrence of the sensor-hijacking attack.