Anomaly Detection in the Artificial Pancreas

Abstract

The integration of subcutaneous sensing and insulin delivery technologies with novel control strategies has brought closer the development of the Artificial Pancreas. Nevertheless, thought recent developments are aimed at preventing chronic complications and less patient discomfort, few works have addressed the critical issue of performance monitoring of the artificial pancreas as well as detection of abnormal functioning in any of its components. This work presents an anomaly detection monitoring tool using the widely known Clarke Error-Grid to identify functional degradation in the artificial pancreas components and guarantee safetycritical control of blood glucose levels. The effect of imperfect calibration of glucose sensors, time lag between blood glucose concentration and interstitial glucose readings, and excessive variability in glucose levels are evaluated against an expected behavior of the glucose regulation loop achieved through an optimal control policy. Results obtained evidence the feasibility of this novel use of the Clarke error grid as a comprehensive monitoring tool for the artificial pancreas.