6th IEEE Conference on Control Technology and Applications (CCTA 2022)
State of the art of Automatic Control of General Anesthesia
Time and Duration:
22 August 2022, Full-day Workshop
Closed-loop control of general anesthesia has raised an increased research interest in the past decades with a significant expansion during the last years because these systems have the potential to bring important benefits in clinical practice. In particular, they can improve the patient safety and the reproducibility of therapy outcomes by avoiding human errors due to distraction and fatigue. In fact, they allow the anesthesiologist to assume the role of supervisor. This implies that, while the automatic control system performs the task of dosing the anesthetic drugs, the anesthesiologist can focus on complex clinical decisions and maneuvers that require medical expertise. The anesthesia process requires administration of different drugs to achieve three main effects related to hypnosis, analgesia and muscular relaxation. There are many issues that should be taken into account when designing the control system: the nonlinearities of the system, the coupling effects between the different drugs, the robustness to inter- and intrapatient variability, the presence of safety constraints, the presence of noise, the estimation of the pharmacokinetic/pharmakodynamic (PK/PD) model, the availability of a reliable simulator for testing the control algorithm, and so on. Because of these challenges to be faced, many different design tools and methodologies have been proposed and exploited in this context.
2 Workshop Scope and Major Contributions
The scope of the workshop is to provide the state-of-the-art of closed-loop control of general anesthesia by presenting some of the most recent methodologies that have been proposed. In particular, during the workshop, the speakers will explain the control problem and how different control and modeling methodologies have been applied in this scenario. They will range from PID-based control, MPC control, intelligent control, advanced modeling techniques, etc. In particular, after an introduction to the control problem, PID control schemes will be explained by especially focusing on the tuning methods. Then, MPC control techniques will be presented, first by considering the control of the Depth-of-Hypnosis only, and then by enlarging the control problem to other hemodynamic variables. An open-source patient simulator that takes into account all the different variables involved in the anesthesia process is then presented and its use in the design of control systems is shown. The fundamental problem of estimating an accurate PK/PD model and to evaluate its variations during a surgery is also addressed. Finally, the use of artificial intelligence and intelligent control is then addressed in the last talk, which also spans outside anesthesia control to show the potentiality of this approach. The attendees will therefore learn how different solutions have been applied for this task and they can be inspired to apply new control methodologies to such a challenging application. The list of potential speakers and a brief abstract of the presented topics are provided in the following section.
3 Workshop Structure and List of Speakers
The workshop will give an overview of the anesthesia control problem and of recently proposed methodologies addressing different issues. Hereafter, the list of speakers that will intervene during the workshop presenting innovative contributions:
1. Title: The anesthesia control problem (1 hour)
Speaker: Antonio Visioli (University of Brescia)
Abstract: The anesthesia control problem and its main challenges will be presented. The three effects (depth of hypnosis, analgesia and neuromuscolar blockade) and three different phases (induction, maintenance and emergence) will be outlined together with the associated drugs. The classical (nonlinear) pharmacokinetic/ pharmadynamic (PK/PD) models will be presented where the synergistic effects of the drugs and the major sources of uncertaintes will be described. Then, issues associated with the sensors will be also explained and the major control constraints will be given in order to justify the need of advanced control methodologies.
2. Title: PID-based control architectures (1 hour)
Speakers: Michele Schiavo (University of Brescia)
Abstract: The design of a PID-based control system for Depth-of-Hypnosis control is presented. Both cases with the administration of only propofol (hypnotic) and the coadministration of propofol and remifentanil (analgesic) will be considered, The process variable is the Bispectral Index Scale (BIS). The tuning of the PID controller will be addressed thoroughly in order to provide the required robustness to the system. Some relevant modifications of the basic control scheme are also presented. Experimental results will be shown and discussed by also making a comparison with manual control.
3. Title: MPC-based control architectures (1 hour)
Speaker: Andrzej Pawlowski (University of Brescia)
Abstract: A model predictive control system for the depth of hypnosis is proposed and analyzed. This approach considers the Bispectral Index Scale (BIS) as the process variable and either the administration of the hypnotic drug only or the simultaneous co-administration of the hypnotic and analgesic drugs as manipulated variable(s). The proposed control scheme is based on an external predictor that, by exploiting the Wiener structure of the pharmacokinetic/pharmacodynamic model, compensates for the process nonlinearity and increases the system robustness by means of an additional filter. Then, it exploits a generalized model predictive control algorithm in order to provide the optimal dosage(s) for the desired BIS level, taking into account the typical constraints of the process.
4. Title: Control of Anesthesia and Hemodynamics for Major Vascular surgery (1 hour)
Speaker: Clara Mihaela Ionescu (University of Ghent, Belgium)
Abstract: A major challenge in anesthesia is to adapt the drug infusion rates from observed patient response to surgical stimuli. The patient models are based on nominal population characteristic response and lack specific surgical effects. In major surgery (e.g. cardiac, transplant, obese patients) modelling uncertainty stems from significant blood losses, anomalous drug diffusion, drug effect synergy/antagonism, anesthetic-hemodynamic interactions, etc. This complex optimization problem requires superhuman abilities of the anesthesiologist. Computer controlled anesthesia holds the answer to be the game changer for best surgery outcomes. Although few, clinical studies report that computer based anesthesia for one or two drugs outperforms manual management. In reality, clinical practice mitigates a multi-drug optimization problem while accommodating large patient model uncertainty. The anesthesiologist makes decisions based on future surgeon actions and expected patient response. This is a predictive control strategy, a mature methodology in systems and control engineering with potential to faster recovery times and lower risk of complications. The goal of this talk is to advance the scope and clinical use of computer based constrained optimization of multi-drug infusion rates for anesthesia with strong effects on hemodynamics. A solution is proposed to identify multivariable models and minimize the large uncertainties in patient response. With adaptation mechanisms from nominal to individual patient models, it designs a multivariable optimal predictive control methodologies to manage strongly coupled dynamics. To maximize performance of the closed loop, it models the surgical stimulus as a known disturbance signal and additional bolus infusions from anesthesiologist as known inputs.
5. Title: Patient simulator for design and evaluation of computer based multiple drug dosing control for anesthetic and hemodynamic variables (1 hour)
Speaker: Dana Copot (University of Ghent, Belgium)
Abstract: We are witnessing a notable rise in the translational use of information technology and control systems engineering tools in clinical practice. This talk empowers the computer based drug dosing optimization of general anesthesia management by means of multiple variables for patient state stabilization. A patient simulator platform is designed through an interdisciplinary combination of medical, clinical practice and systems engineering expertise gathered in the last decades by our team. The result is an open source patient simulator in Matlab/Simulink from Mathworks(R). Simulator features include complex synergic and antagonistic interaction aspects between general anesthesia and hemodynamic stabilization variables. The anesthetic system includes the hypnosis, analgesia and neuromuscular blockade states, while the hemodynamic system includes the cardiac output and mean arterial pressure. Nociceptor stimulation is also described and acts as a disturbance together with predefined surgery profiles from a translation into signal form of most commonly encountered events in clinical practice. A broad population set of pharmacokinetic and pharmacodynamic (PKPD) variables are available for the user to describe both intra- and inter-patient variability. This simulator has some unique features, such as: i) additional bolus administration from anesthesiologist, ii) variable time delays introduced by data window averaging when poor signal quality is detected, iii) drug trapping from heterogeneous tissue diffusion in high body mass index patients. We successfully reproduced the clinical expected effects of various drugs interacting among the anesthetic and hemodynamic states. Our work is uniquely defined in current state of the art and first of its kind for this application of dose management problem in anesthesia. This simulator provides the research community with accessible tools to allow a systematic design, evaluation and comparison of various control algorithms for multi-drug dosing optimization objectives in anesthesia.
6. Title: Model structures, uncertainty and pharmacometric covariate modeling (1 hour)
Speaker: Kristian Soltesz (University of Lund, Sweden)
Abstract: The conventional series interconnection of a mammillary compartment model for pharmacodynamics (PK) and a Hill sigmoid for pharmacodynamics (PD) forms the basis for a vast majority of model-based feedback controller designs for automatic delivery of anesthetic drugs. Considered, and in some cases clinically evaluated, controllers range from PID to MPC. While to some extent physiologically motivated, and well-established within the pharmacology literature, using the models in a clinical setting comes with several challenges. One challenge lies in data available for system identification being limited in excitation, for patient safety and ethical reasons. Another challenge is variability in response dynamics between patients. This is typically approached through pharmacometric covariate modeling, that aims to model the PKPD parameters as explicit functions of known covariates such as patient age, weight, etc. Despite numerous published covariate models for popular anesthetic drugs, there are no models that fit data as closely as implicitly assumed by many proposed control synthesis strategies. There are several plausible contributors to this situation: the pre-determined search space of covariate functions might limit model quality; EEGmonitor artifacts and unmeasured disturbances; inter-patient variability of response dynamics caused by for example change in hemodynamic state; structural mismatch between actual response dynamics and assumed PK-PD model. In this talk we look at some published clinical data and argue that closed-loop control of anesthesia is much more a data-driven modeling challenge, than a controller synthesis challenge. Worthwhile questions to ask are what model structures provide an adequate balance between under-fitting of data, identifiability of parameters form the same data, and adequacy for model-based controller synthesis. There is unlikely one simple answer, but a large and increasing amount of modeling data is available to investigate these questions.
7. Title: Artificial Intelligence and Intelligent Control in Anesthesia (1 hour)
Speaker: Juan Albino Mendez (University of La Laguna, Spain)
Abstract: The advance of automation and artificial intelligence techniques is having an important impact in different areas of society. In particular, many of these applications have focused in the medicine field. Although the range of applications is large, many of them share common methodologies. The basic knowledge of these methods can help the clinician both in the clinical practice and the research. This talk will describe the role of artificial intelligence methods and intelligent control in medicine. Their potential for applications in different fields, from medicine robotics to computer aided decision systems for clinicians, will be presented. Then, the talk will concentrate on biomedical control. It is well known that developing control systems for medical applications has a great difficulty due to the complexity of the processes involved. One example of biomedical control in the field of anaesthesiology will be shown. Main challenges and results obtained in the research group headed by Dr. Mendez will be presented.