Neural Network and Physiological Parameters Based Control of Artificial Pancreas for Improved Patient Safety

Acyber-physical system (CPS) establishes a close interaction between system's computational core and the control of physical process. In case of Diabetes, failure in endogenous insulin production requires exogenous infusion of required drug amount. We have proposed an architecture for artificial pancreas and checked its validity in simulations. The aim is to control blood glucose level (BGL) of a patient suffering from diabetes and to prevent the harmful state of Hypoglycemia. For this, vital signs monitoring is introduced through which hypoglycemic condition can be efficiently detected and avoided. Electrocardiogram, Heart beat rate, Electroencephalography and skin resistance are known to depict an irregularity in blood glucose. Upon detection, a specified amount of Glucagon is infused into patient's body. The system consists of an insulin infusion and glucagon pump, through which insulin/glucagon is entered into the patient's body subcutaneously, based on the current BGL. A neural network predictive controller is designed to keep the glucose level inside the desired 'safe range'. The simulations have shown that patient safety can be improved through this strategy.