Sensor Fusion for Glucose Monitoring Systems

A fully automated artificial pancreas (AP) requires accurate blood glucose (BG) readings. However, many factors can affect the accuracy of commercially available sensors. These factors include sensor artifacts due to the pressure on surrounding tissues, connection loss, and poor calibration. The AP may administer an incorrect insulin bolus due to inaccurate sensor data when the patient is not supervising the system. The situation can be even worse in animal experiments because animals are eager to play with the sensor and apply pressure. In this study, we propose and derive a Multi-Model Kalman Filter with Forgetting Factor (MMKFF) for the problem of fusing information from redundant subcutaneous glucose sensors. The performance of the developed MMKFF was assessed by comparing it against other Kalman Filter (KF) strategies on experimental data obtained in two different animals. The developed MMKFF was shown to provide a reliable fused glucose reading. Additionally, compared to the other KF approaches, the MMKFF was shown to be better able to adjust to changes in the accuracy of the glucose sensors.