Sensor selection for energy-efficient ambulatory medical monitoring

• Main Authors: Shih, Eugene Inghaw (Contributor), Shoeb, Ali H. (Contributor), Guttag, John V. (Contributor)
• Other Authors: Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory (Contributor), Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science (Contributor)
• Format: Article
• Language: English
• Published: Association for Computing Machinery, 2010-02-10T19:59:38Z.
• Subjects: Article
• Online Access: Get fulltext

 Epilepsy affects over three million Americans of all ages. Despite recent advances, more than 20% of individuals with epilepsy never achieve adequate control of their seizures. The use of a small, portable, non-invasive seizure monitor could benefit these individuals tremendously. However, in order for such a device to be suitable for long-term wear, it must be both comfortable and lightweight. Typical state-of-the-art non-invasive seizure onset detection algorithms require 21 scalp electrodes to be placed on the head. These electrodes are used to generate 18 data streams, called channels. The large number of electrodes is inconvenient for the patient and processing 18 channels can consume a considerable amount of energy, a problem for a battery-powered device. In this paper, we describe an automated way to construct detectors that use fewer channels, and thus fewer electrodes. Starting from an existing technique for constructing 18 channel patient-specific detectors, we use machine learning to automatically construct reduced channel detectors. We evaluate our algorithm on data from 16 patients used in an earlier study. On average, our algorithm reduced the number of channels from 18 to 4.6 while decreasing the mean fraction of seizure onsets detected from 99% to 97%. For 12 out of the 16 patients, there was no degradation in the detection rate. While the average detection latency increased from 7.8 s to 11.2 s, the average rate of false alarms per hour decreased from 0.35 to 0.19. We also describe a prototype implementation of a single channel EEG monitoring device built using off-the-shelf components, and use this implementation to derive an energy consumption model. Using fewer channels reduced the average energy consumption by 69%, which amounts to a 3.3x increase in battery lifetime. Finally, we show how additional energy savings can be realized by using a low-power screening detector to rule out segments of data that are obviously not seizures. Though this technique does not reduce the number of electrodes needed, it does reduce the energy consumption by an additional 16%.