Adaptive sized windows to improve real-time health monitoring: a case study on heart attack prediction

  • Authors:

  • Morten Lindeberg;Vera Goebel;Thomas Plagemann

  • Affiliations:

  • University of Oslo, Oslo, Norway;University of Oslo, Oslo, Norway;University of Oslo, Oslo, Norway

  • Venue:
  • Proceedings of the international conference on Multimedia information retrieval
  • Year:
  • 2010

Quantified Score

Hi-index 0.00

Visualization

Abstract

Today, data stream management systems (DSMSs) mainly support the processing of data stream windows of static size. In the application domain of health monitoring, processing of data streams from body sensor networks can benefit from adapting to physiological processes of the human body, e.g., the heartbeat. This requires the introduction of time-based sliding windows of adaptive "size" in DSMSs. We have compared an existing algorithm with fixed sized windows for the identification of myocardial ischemia, a precursor to heart attacks, with a new version based on adaptive sized windows. The size of the sliding window over the heart acceleration stream was adapted to events in an electrocardiogram data stream that identifies heartbeats. This enabled that accelerometer readings originating from the same heartbeat were calculated in the same batch, as opposed to fixed size batches with no real relation to the continuous variable durability of each heartbeat. As a result, we obtained results with substantially less variance. We present in this paper the design and implementation of our adaptive window technique using the DSMS Esper. We measure both, the improvements of the medical analysis results and the overhead in terms of extra processing. The evaluation results indicate that the technique leads to considerably better medical analysis results in our case studies, and that the overhead introduced by handling the window technique itself is quite low.