SIFT : Multi Physiological Signal Feature Correlation-based Sensor Compromise Detection in Body Sensor Networks

A Body Sensor Network (BSN) consists of a set of sensing devices deployed on a user (patient) typically for health monitoring purposes.The fact that BSNs provide sensitive information to caregivers about the user’s health makes them attractive targets for tech-criminals to exploit. In this work, we focus on BSN sensor compromise detection especially in the cases physiological signal monitoring sensors. A compromised sensor may generate erroneous data which may cause the incorrect interpretation of user health leading to wrong diagnosis and treatment. Detecting sensor compromise is a non-trivial task in BSNs. Traditional sensor compromise detection relies on one of two general techniques: (1) using redundant sensors and some form of voting to determine sensor accuracy, or (2) using past sensor values for predicting the expected current sensor value range. However, in a BSN due to usability limitations we cannot expect the presence of redundant physiological sensors of the same kind. Further, the dynamic nature of the human body precludes the reliance on historical sensor values for predicting the current range of values. In this paper, we present SIFT, a novel methodology to address the problem of sensor compromise without relying on either redundant sensors or historical sensor values. SIFT leverages the fact that in BSNs physiological signals based on the same underlying physiological process (e.g., cardiac process) are inherently correlated i.e., they share similar features with each other. Given this group of correlated signals, any unnatural alteration in one of the correlated signal will not be reflected in the other signals in the group. As SIFT uses signals from multiple sensors it does not require node redundant sensors. Further, as the correlated signals are measured in a synchronous fashion, the current state of the patient’s physiology is automatically taken into account, thus overcoming a typical problem with relying on techniques that use past sensor values. We illustrate the operation of SIFT through a case study where we detect the compromise of a electrocardiogram (ECG) sensor using two related signals arterial blood pressure (ABP) and respiration (RESP) sensors as reference. Analysis of our case study demonstrates promising results with over 98% accuracy in detecting even subtle ECG signal alterations for both healthy and unhealthy patients.