Robust health stream processing

Abstract

As the cost of personal health sensors decrease along with improvements in battery life and connectivity, it becomes more feasible to allow patients to leave full-time care environments sooner. Such devices could lead to greater independence for the elderly, as well as for others who would normally require full-time care. It would also allow surgery patients to spend less time in the hospital, both pre- and post-operation, as all data could be gathered via remote sensors in the patients home. While sensor technology is rapidly approaching the point where this is a feasible option, we still lack in processing frameworks which would make such a leap not only feasible but safe. This work focuses on developing a framework which is robust to both failures of processing elements as well as interference from other computations processing health sensor data. We work with 3 disparate data streams and accompanying computations: electroencephalogram (EEG) data gathered for a brain-computer interface (BCI) application, electrocardiogram (ECG) data gathered for arrhythmia detection, and thorax data gathered from monitoring patient sleep status.