Rapid automatic identification of parameters of the Bergman Minimal Model in Sprague-Dawley rats with experimental diabetes for adaptive insulin delivery

Abstract

Glucose-Insulin regulation models can be used to individualize insulin therapy. However, the experimental techniques currently used to identify the appropriate parameter sets of an individual are expensive, time consuming, and very unpleasant for the patient. Since there is a wide range of intrapersonal parameter variability, the identified parameters in a laboratory setting (at rest) are not optimal for dynamic conditions of daily activities. In this study we propose a methodology to identify three parameters of Bergman's Minimal Model in streptozotocin-induced diabetic rats from the experimental data of the glucose response to exogenous insulin doses, based on a genetic algorithm (GA). The algorithm requires glucose measurements from a continuous subcutaneous sensor once every 5 min and the amount of injected insulin. The model parameters of 20 in vivo experiments (from 19 rats) were identified with high accuracy and the average root-mean squared (RMS) error between predicted and measured glucose concentration was 17.6 mg/dl. Since the algorithm requires a relatively short (60–120 min) observation time it can be used for real-time parameter identification to optimize insulin infusion systems. Model parameter changes due to experimental settings like drug testing or in natural lifestyle changes should be calculable, on-the-fly, using data from only the glucose sensor and the amount of insulin delivered.