Estimation of glucose and insulin concentration using nonlinear Gaussian filters

Abstract

In this ongoing work, three non-linear Gaussian filters viz. the unscented Kalman filter (UKF), the cubature qudrature Kalman filter (CQKF) and the Gauss-Hermite filter (GHF) are designed to track blood glucose and insulin concentrations, as well as interstitial insulin level with the help of the `Bergman's minimal model of glucose-insulin homeostasis'. All the filters successfully track the plasma glucose and insulin level, even without the declaration of meal intake. We evaluate the filters' performances in terms of root mean square error (RMSE) which shows all the three filters are equally capable of tracking plasma glucose and insulin from noisy blood glucose measurements.