Dynamic Modeling and Predictive Control of Cardiovascular System Using Vagal Nerve Stimulation

Vagal nerve stimulation has shown beneficial effects in treating cardiovascular diseases. However, the lack of clinical efficacy, as well as differences in stimulation parameters due to patient variability, indicates the necessity to integrate an automatic closed-loop control method, enabling subject-specific, optimal VNS parameter updates in real time. A mathematical model to predict subject-specific cardiovascular response to vagal nerve stimulation is required for validating the efficacy and safety of the closed-loop VNS device, as well as to explore more advanced control algorithms.In this work, a model-based control technique for regulating the heart rate and blood pressure using VNS technology is presented. The closed-loop framework is based on an in silico model of the rat cardiovascular system for the simulation of the hemodynamic response to multi-location vagal nerve stimulation. The in silico model is derived by compart- mentalizing the various physiological components involved in the closed-loop cardiovascular system with intrinsic baroreflex regulation to virtually generate nominal and hypertension-related heart dynamics of rats in rest and exercise states. A nonlinear model predictive controller (NMPC), using a reduced cycle-averaged model, is firstly proposed to continuously monitor the outputs from the in silico model, estimates the current state of the reduced model, and computes the optimum stimulation locations and the corresponding stimulation parameters. However, there is significant difficulty in the application of an NMPC, including the development and validation of a predictive cardiac model, as well as the high computational cost. To overcome the challenges associated with the NMPC, a multiple model predictive (MMPC) is then proposed, which uses a predictive framework with multiple local models identified from a detailed rat cardiac model to encompass the entire anticipated arterial pressure dynamics. The computational expense of the proposed method is verified with rigorous hardware-in-the-loop implementation, which provides an assessment of the efficacy of the algorithm for future implementation of our MMPC for pre-clinical and clinical studies. The robustness of both control algorithms is demonstrated with respect to their ability to handle setpoint tracking and disturbance rejection in different simulation scenarios.