Integrated real time bowel sound detector for artificial pancreas systems

• We demonstrated a compact ultra-low power bowel monitoring system which consumes 53 μW of power and can be implemented in a 0.96 mm2 chip space.
• We have adopted an analog feature extractor to relax the requirements of power and computation intensive digital signal processors and optimized the module for low power operation.
• The monitoring system extracts bowel features in real time, can be seamlessly integrated in a portable system and provides a non-invasive way to detect and correlate physiological measures with motility or meal instance.
• The prototype demonstrates 85% accuracy with low false positives, which is sufficient to determine whether a patient has eaten when monitored continuously.

This paper reports an ultra-low power real time bowel sound detector with integrated feature extractor for physiologic measure of meal instances in artificial pancreas devices. The system can aid in improving long term diabetic patient care and consists of a front end detector and signal processing unit. The front end detector transduces the initial bowel sound recorded from a piezoelectric sensor into a voltage signal. The signal processor uses a feature extractor to determine whether a bowel sound is detected. The feature extractor consists of a low noise, low power signal front-end, peak and trough locator, signal slope and width detector, digitizer, and bowel pulse locator. The system was fabricated in a standard 0.18 μm CMOS process, and the bowel sound detection system was characterized and verified with experimentally recorded bowel sounds. The integrated instrument consumes 53 μW of power from a 1 V supply in a 0.96 mm2 area, and is suitable for integration with portable devices.