A robust model predictive control framework for the regulation of anesthesia process with Propofol

Abstract

In this work, we present a robust Model Predictive Control (MPC) strategy based on linear matrix inequalities (LMIs) for the intravenous administration of Propofol, a drug which is used for anesthesia during surgeries. The controller is designed for a population of patients and takes into account constraints on the amount of administered drug and on the drug concentration profile. A detailed compartmental mathematical model available in the literature is adjusted to the available data and provides the future predictions of the process. In the context of the application, only the depth of anesthesia (BIS index) is assumed to be measured-as it is common in practice. The state of the system (drug amount in organs) is estimated in real-time by incorporating a state observer.The derived control scheme, along with the designed state observer are able to deal with major challenges in controlling the depth of anesthesia which are inter-and intra-patient variability, model nonlinearity, and model uncertainty.The controller is able to satisfy the divergent characteristics of the patients of the dataset, while satisfying in parallel all the imposed constraints. Moreover, we show that by considering smaller groups of patients with similar characteristics the corresponding responses are significantly improved.

Linear MPC for anesthesia process with external predictor

“…Automated systems for drug delivery have attracted the attention of the control system community due to the presence of challenging issues such as nonlinearities and variable time delays, variable dynamics for specific drug/agent administration, robustness related to the patient safety [1][2][3]. In this context, a significant effort has been made for systems that use feedback control concepts to develop automatic regulation schemes of physiological variables like blood pressure, blood glucose concentrations, etc.…”

“…The application of MPC techniques to the anaesthesia process has been analyzed in several works. Most of them use a state estimator (such a Kalman filter [1,22]), which might yield a large settling time. In [7,23] a piece-wise linearization of the Hill function is proposed to eliminate the nonlinear component from the control loop.…”

MPC for Propofol Anesthesia: the Noise Issue

“…The application of MPC techniques to the control of depth of hypnosis (DoH) in the anaesthesia process using propofol has been analyzed in several works, e.g. [2][3][4][5][6][7][8][9]. This interest is motivated mainly by the possibility of predicting the patient response to drug administration [4,7].…”

“…This interest is motivated mainly by the possibility of predicting the patient response to drug administration [4,7]. In particular, the methods described in [2,3] are focused on inter-/intra-patient variability, targeting the most vulnerable aspect in MPC approaches, namely, model uncertainties. These approaches usually results in a complex control system with heavy computational requirements and re-tuning or adaptation is not trivial [9].…”

“…These approaches usually results in a complex control system with heavy computational requirements and re-tuning or adaptation is not trivial [9]. For example the MPC based control system proposed in [2] exploits state observers, which provide the drug concentration estimation in body compartments. These estimations are used for predicting the effect of the drug in the control signal computation.…”