Cocaine Detection using Wearable On-body Sensors


The ability to continuously monitor an individuals physiology, his/her activities, and the environment around them using remote wireless sensors can profoundly improve our understanding of cocaine addiction and our ability to develop, monitor, and target medication (and non-medication) treatments for those addicted. Subjective, retrospective self-reports provide a static cross-sectional picture that is not only vulnerable to recall bias and misrepresentation, but one that may also fail to capture important unconscious aspects of addiction of which the individual is truly unaware of (e.g., conditioned cues/context/behaviors). Continuous remote sensing technologies provide a promising new window into patients daily lives one through which an individuals clinical course might be more holistically, accurately, and dynamically viewed as it evolves in real-time. We posit that such a fine-grained approach has the potential not only to provide sensitive and specific signatures of cocaine use/intoxication, but also to shed clinically informative light on unique aspects of an individuals use, the relative roles of specific environmental factors (people, places, and things) in their personalized pathway to relapse (e.g., drug-, cue-, and/or stress-induced), and in turn, highly individualized and dynamic markers of treatment response. This project aims to detect and understand the contexts relating to drug use in highly variable, uncontrolled, real-world circumstances.


We leverage the fact that cocaine causes morphological changes in electrocardiogram signals. In our experiments we use Zephyr BioHarness chestband sensors to gather ECG data during cocaine and non-cocaine sessions. We separately extract ECG features from cocaine and non-cocaine sessions. We train a model to detect cocaine from non-cocaine sessions using the extracted ECG features only. The goal is given a window of ECG data we would like to output the probability of cocaine in that window. For more information refer to our publications below.


  1. Annamalai Natarajan, Gustavo Angarita, Edward Gaiser, Robert Malison, Deepak Ganesan, and Benjamin Marlin, Domain Adaptation Methods for Improving Lab-to-field Generalization of Cocaine Detection using Wearable ECG, Proceedings of ACM Ubicomp 2016.
  2. Natarajan, A., Gaiser, E., Angarita, G., Malison, R., Ganesan, D., & Marlin, B. (2014, September). Conditional random fields for morphological analysis of wireless ECG signals. In Proceedings of the 5th ACM Conference on Bioinformatics, Computational Biology, and Health Informatics (pp. 370-379). ACM.
  3. Natarajan, A., Parate, A., Gaiser, E., Angarita, G., Malison, R., Marlin, B., & Ganesan, D. (2013). Detecting Cocaine Use with Wearable Electrocardiogram Sensors. Proceedings of the 2013 ACM international joint conference on Pervasive and ubiquitous computing (ACM Ubicomp) (Honorable Mention)


  1. Angarita, G. A., Nararajan, A., Gaiser, E., Parate, A., Marlin, B., Gueorguieva, R. R., Lampert, R., Ganesan D., & Malison R. T., A Remote Wireless Sensor Network (RWSN) / Electrocardiographic (ECG) Approach to Discriminating Cocaine Use, Poster presented at the Annual Meeting of the College on Problems of Drug Dependence, Puerto Rico, (June, 2014). PDF
  2. Natarajan, A., Parate, P., Gaiser, E., Angarita, G., Malison, R., Marlin, B., & Ganesan, D., Detecting Signatures of Cocaine Using On-Body Sensors. Poster presented at the Annual Meeting of the American Medical Informatics Association, Washington D.C. (Nov, 2013). PDF


  • Annamalai Natarajan (UMass Amherst)
  • Abhinav Parate (UMass Amherst)
  • Gustavo Angarita (Yale University)
  • Edward Gaiser (Yale University)
  • Robert Malison (Yale University)
  • Benjamin Marlin (UMass Amherst)
  • Deepak Ganesan (UMass Amherst)
  • Funding

    This work was supported in part by National Institute of Drug Abuse grant #149135