A semantically-adaptive strategy for energy-efficiency in wireless medical monitoring devices

We propose a novel strategy for energy-efficient operation of wireless monitoring devices under the premise that medical experts are primarily interested in atypical observations - For epilepsy monitoring, EEG data is most valuable at epileptic activity onset. Or, a gait-stability monitoring application is most interested in unusual footsteps. Observations are atypical if application-specific medical metrics and biosignal features are statistical outliers. Our strategy admits energy-efficient early-detection of such observations, leading to: (i) an increase in medical information quality by sampling aggressively over semantically important behaviors, and (ii) a savings in energy by precluding communication of typical measurements. From experimentally collected plantar pressure datasets, we show that this can yield up to a 62% improvement in gait-stability metric evaluation for atypical footsteps and a 10% energy cost reduction compared to a recently proposed non-adaptive compressive sensing technique.