Unconstrained and constrained generalised predictive control of depth of anaesthesia during surgery
Introduction
Anaesthesia can be defined as the lack of response and recall to noxious stimuli. It comprises muscle relaxation, analgesia and unconsciousness (i.e. depth of anaesthesia). In other words, depth of anaesthesia (DOA) is one of the components of anaesthesia. Anaesthetists use a variety of observations, such as blood pressure, heart rate, lacrimation, movement, sweating, and pupil response, to make a judgement on DOA levels.
The measure of anaesthetic depth during surgical anaesthesia has always represented a tough challenge and the experience of the anaesthetist is required to control the patient's anaesthetic state. Controlling anaesthesia means that the patient is maintained at a suitable level of sedation in order to allow the surgeon to proceed with the operation without causing awareness in the patient. Hence, a number of drugs have to be administered carefully to achieve an appropriate DOA level without compromising the patient's health. There have been reports of incomplete general anaesthesia (Jessop & Jones, 1991) by patients who were pharmacologically paralysed while under anaesthesia (Kulli & Koch, 1991):
The feeling of helplessness was terrifying. I tried to let the staff know I was conscious but I could not move even a finger nor eyelid. It was like being held in a vice and gradually I realised that I was in a situation from which there was no way out. I began to feel that breathing was impossible and I just resigned myself to dying.
Nowadays, safer drugs are used with patients, and better monitoring machines help the anaesthetist make an informed decision during surgery, and state-of-the-art delivery equipment have also helped him/her maintain a more stable anaesthetic depth. Such clinical measures, however, are still open to the anaesthetist's scrutiny as patients react differently with drugs, and as a result, different patients at the same anaesthetic depth can present different clinical measures.
Several researchers have argued that if automatic control of anaesthetic depth relies on its accurate measurement, then more efforts must be directed towards devising ways of measuring DOA directly. Previous research work by Schwilden, Stoeckel, and Schuttler (1989) used quantitative EEG analysis in humans to give an indication of the anaesthetic state. However, the interpretation of the tracings proved difficult and a subjective task. The information proved unreliable even when interpreted by experienced staff, since the characteristic patterns are often disturbed by factors such as anoxia, surgical stimulations. Recently, auditory evoked potentials (AEP) have been shown to be good indicators of anaesthetic depth (Kenny, Fadzean, Mantzaridis, & Fisher, 1992; Linkens, Abbod, & Backory, 1996); they are said to have a graded response with changing anaesthetic depth, they demonstrate little inter-patient variability, and they measure the underlying anaesthetic depth as opposed to the anaesthetic concentration in the blood. However, the extraction, analysis, and the on-line use of such signals in the operating theatre are not a straightforward task.
An alternative to direct measurement of DOA has been the merger of several clinical signs such as blood pressure, heart, respiration, etc. to obtain the closest possible indication of how well the patient is anaesthetised. Indeed, in a study which was conducted by the second author, anaesthetists were asked to rank the relative importance of ten clinical signs. These signs were ranked on a scale of 1 to 10 based on the mean values provided by these anaesthetists. From these 10 clinical signs, blood pressure (ranked 6- the lowest ranked having been given to non-measurable signs or variables which are not easy to model) has been selected as one variable to give an indication of anaesthetic depth. Hence, the use of arterial blood pressure, monitored via an inflatable cuff using a DINAMAP instrument, has been investigated for feedback control with simple PI strategies (Robb, Asbury, Gray, & Linkens, 1988). In this case, the control actuation was via a stepper motor driving the dial on a gas vaporiser. It has been concluded from this study and others that when no emergency conditions occur, blood pressure can be used to provide a good indication of the patient's anaesthetic state. More recently, a multitasked closed-loop control system consisting of two controllers, was presented by Gentilini et al. (2001). In this work, the authors controlled mean arterial pressure (MAP) and hypnosis through the bi-spectral index (BIS), which is a measure of the EEG, via isoflurane administration.
The Research Group at the University of Sheffield, now led by the first author, has worked in the area of anaesthesia monitoring for a number of years. Analytical as well as hierarchical fuzzy modelling and control techniques of anaesthesia were investigated. From a purely analytical viewpoint, the control theme at the heart of this study is that of Model-Based Predictive Control, particularly generalised predictive control (GPC) (Clarke, Mohtadi, & Tuffs, 1987), which is seen by many as the control strategy that has had the most significant impact on solving a wide range of complex industrial problems, including those within the realm of biomedicine (Mahfouf & Linkens, 1998). In this paper, hard constraints are introduced as part of the optimisation problem. Hence, this paper is organised as follows: Section 2 will review the re-circulatory physiological model relating to the drug isoflurane (Derighetti, 1999), together with our own modification in terms of the control actuation being via a syringe pump rather than a gas vaporiser. In Section 3, the development of constrained GPC is briefly reviewed, while in Section 4 results of the simulation experiments are presented and discussed. In Section 5 the transfer of the overall control system to the operating theatre is described, the subsequent real-time experiments are presented, and the discussions and analyses of the various results obtained are given. Finally, Section 6 gives the conclusions relating to this study.
Section snippets
Associated with isoflurane
There has been an increasing interest in pharmacokinetics modelling in recent years, and many concepts and trial models have been proposed, and some of them validated, as a result. Among the proposed models one can cite empirical models (Wise, 1985), compartmental models (Jacquez, 1972; Hull, 1991), and physiological models (Jarvis, 1994). Tucker (1994) reviewed the evolution of various concepts in pharmacokinetics and pharmacodynamics, but the paragraphs below give a brief summary relating to
The unconstrained case
The long-range predictive controller developed in this research study is based on the GPC strategy (Clarke et al., 1987) whose theoretical background is briefly reviewed here.
Consider the following locally linearised discrete model in the backward shift operator z−1:wherewhere u(t) represents the control input and y(t) is
Identification of linear anaesthesia models
For data generation we applied the input profile (liquid isoflurane) shown in Fig. 2. A total of 400 data points were used for identification.
For off-line model identification the following second-order model structure was used:where d is the offset term due to the initial blood pressure value.
It is worth noting that with reference to , , the output y(t) will represent the MAP (see Eq. (2)) and the input u(t) will represent pair (see Eq.
Real-time experiments
The real-time closed-loop control system which was transferred to the operating theatre comprises (see Fig. 7):
- •
An IBM compatible microcomputer which incorporates the control system.
- •
A BRAUN PERFUSOR SECURA digital pump driving a disposable syringe containing a liquid solution of isoflurane.
- •
A DINAMAP Instrument for measuring the arterial blood pressure.
- •
A CAPNOMAC ULTIMA Device for measuring the inspired and expired isoflurane concentrations.
The links between the syringe pump, the Capnomac
Conclusions
The application of constrained GPC which uses the QP approach to on-line control of anaesthesia via MAP measurements has been successful from several viewpoints: (1) Achieving good anaesthesia by keeping a steady level of mean arterial pressure, which in the eyes of many anaesthetists represents a good indicator of DOA, in other words the patients recovered quicker, they were not aware during the operation, and consequently went home earlier. (2) Administration of a low dosage of isoflurane
Acknowledgements
The authors acknowledge financial support from an EPSRC Research Grant GR/L89051/01 and would like to thank the anonymous reviewers for their comments which helped to improve the quality of this paper.
References (23)
- et al.
Conscious awareness during general anaesthesia—what are we attempting to monitor?
British Journal of Anaesthesia
(1991) - et al.
Does anaesthesia cause loss of unconsciousness?
Tins
(1991) - et al.
Multiple model approach to uptake and distribution of halothaneThe use of an analogue computer
Computers and Biomedical Research
(1972) - et al.
Generalised predictive control, Part I and II
Automatica
(1987) - Derighetti, M. (1999). Feedback control in anaesthesia. Ph.D. thesis, Swiss Federal Institute of Technology,...
- et al.
Multitasked closed-loop control in anaesthesia
IEEE Engineering in Medicine and Biology
(2001) - Hull, C. J. (1991). Pharmacokinetics for anaesthesia. London: Butterworth Heinemann, ISBN 0 7506...
Compartmental analysis in biology and medicine
(1972)Physiological pharmacokinetic models—a review of their principles and development
Anaesthetic Pharmacology Review
(1994)- et al.
Closed-loop control of anaesthesia
Anesthesiology
(1992)
Cited by (33)
Definitions and measurements of the states of vigilance
2021, Methodological Approaches for Sleep and Vigilance ResearchDesign and application of generalized predictive control strategy with closed-loop identification for burn-through point in sintering process
2012, Control Engineering PracticeCitation Excerpt :The GPC is one of the most powerful and useful model-based control methods for a wide class of nonlinear dynamic systems (Sato, 2010; Suichies, Leroux, Dechert, & Trusiak, 2000). This has been widely applied in process control industry for its merits (Mahfouf, Asbury, & Linkens, 2003; Tsai, Lin, Wang, & Teng, 2009), such as the time-domain formulation and constraint-handling capacity. In this section, BTP generalized predictive model is firstly discussed.
Using fuzzy self-organising maps for safety critical systems
2007, Reliability Engineering and System SafetyCitation Excerpt :As described in Section 2, the FSOM has been successfully employed in a wide range of control domain applications. The SCANN can be used for both industrial control applications (where inputs are sensors and outputs are actuators) and medical systems (where inputs are sensors and output is dosage [22]). By exploiting safety constraints, the SCANN approach may be more beneficial than safety ‘monitors’.
Model based parametric control in anesthesia
2005, Computer Aided Chemical EngineeringContinuous drug infusion for diabetes therapy: A closed-loop control system design
2008, Eurasip Journal on Wireless Communications and NetworkingGeneral anesthesia regulation using propofol infusion for adult patient administered by a positive system during induction phase
2020, Russian Journal of Biopharmaceuticals