A Fuzzy Logic Based Decision Support System for Low-flow Closed-loop Anaesthesia

Embed Size (px)

Citation preview

  • 8/3/2019 A Fuzzy Logic Based Decision Support System for Low-flow Closed-loop Anaesthesia

    1/6

    A Fuzzy Logic based Decision Support Systemfor Low-flow / Closed-loop A naesthesia

    Belinda HooperXiheng HuDepartment of Electrical EngineeringUniversity of SydneyNew South Wales, 2006, AustraliaAbstract

    Low flo w / closed loop anaesthesia (LFCL.4) is thecontrolled administration of the minimum amount of res-piratory and anaesthetic gases required to anaesthetisethe patient. At present, the control procedure is manual,with the anaesthetist visually assessitzg the volume of thereservoir bag and the inspiratory and expiratory gas con-centrations. Considering this, the application of f u z ylogic to clinical anasethesia seems ideal. The design ofsuch a system is in two stages. The first stage is of a sin-gle gas (oxyge n) system, operating in an open decisionsupport mode, based otz the ex pen knowledge of ananaesthetist. Once the design is clinically optimised, thefeedback loop will be closed to provide the automaticcontrol of oxygen only. The second stage of the prO]eCtwill be the extension of the system to handle other an-aesthetic gases. A graphical user interface, intevacedwith a software-driven simulation of the controller, wasdesigned to aid the clinical evaluation and to display thesystem components.

    1. IntroductionLow-flow/closed-loop anaesthesia (LFCLA) is amethod of anaesthesia delivery which recycles the ex-haled inhalation gases back to the patient aftereliminating carbon dioxide. To maintain the correctdepth of anae sthesia, to control pain sufficiently nad tomaintain adequate oxygen supply to the patient, a con-

    trolled quantity of anaesthetic ag ents and gases a re addedto the recycled exhaled gases. Th is quantity is dependenton the amount absorbed by the body which in turn is de-pendent on the bodys uptake and metabolism. Thisdelivery method is in contrast to the present practice ofhigh-flow or open-loop anaesthesia in which the patientreceives a constant quantity of gas well above the mini-mum required level.The advantages of LFCLA are economic, environ-mental and physiological. Th e reduction in volume of

    Gycsrgy JarosBarry BakerDepartment of AnaesthesiaUniversity of Sydn eyNew South W,ales,2006, Australia

    anaesthetic gases used can offer savings of up to 75% ofthe price of anaesthetic agents and lessen environmentalpollution. It also increases the ability to monitor thephysiologic condition of the patient noninvasively, thus,increasing the educational value and clinical understand-ing of the patients state during anaesthesia. Thesenoninvasive physiological and pharmacological monitor-ing capabilities provided by L FCLA ar e considered by theenthusiasts to be the most compelling.Unfortunately, the majority of clinicians have not hadsufficient exposure in LFC LA to be comfortable in its useor to fully appreciate its merit. In ad dition, it requiresconstant monitoring, manual control and quick decisionmaking by the anaesthetist. However, with improvingtechnologies in monitoring equipment, and increasinglyimportant environmental and ho spital cost considerations,there has been a renewal of interes t in the use of low flow,minimal flow, and completely closed system s.On-line decision support system would make the deliv-ery of anaesthesia easier and safer. It would free theanaesthetist from routine mo nitoring of gas levels and al-low concentration on the anaesthetic procedure . LFCLAis a system that would accept complete com puter assistedautomation with much greater benefits to the patient thanany other delivery system [Droh & Spintage, 19861.There have been several attempts [Boaden and Hutton,1986; Boaden et al, 1989; Morris, 1983: Vishnoi a nd Roy,19911 to design a Control System for L FCL A but no proj-ect has been adopted by the industry as an acceptablesolution. These above methods utilise classical controlalgorithms, which rely on ex act and continuous mea sure-ments of the control variables. It is ob vious thay they arenot capable of modelling the nonlinearities and the com-plexity of the an aesthetists decision-makingprocess.The potential of applying Fuzzy Logic to anaestheticcontrol is discussed in the ;uticle by Asbury and Tzabar,1995 [l]. It is considered that Fuzzy Logic can model theanaesthetists cognitive processes and although the use offuzzy logic in medicine is in its infancy, fuzzy logic tech-niques are already present in many consum er products.

    0-7803-3796-4/97/$10.0001997IEEE 1615Authorized licensed use limited to: UNIVERSIDADE FEDERAL DO RIO DE JANEIRO. Downloaded on December 25, 2008 at 17:58 from IEEE Xplore. Restrictions apply.

  • 8/3/2019 A Fuzzy Logic Based Decision Support System for Low-flow Closed-loop Anaesthesia

    2/6

    F U Z Z - I E E E 9 7

    Fuzzy logic has been suggested for application in sev-eral areas of anaesthesia. It has been applied to thetreatment of hypertension during anaesthesia, [Oshita etal, 19933 and to the control of inspired oxygen in venti-lated newborn infants [Sun et al, 19941. To date there isno recorded attempt to con trol oxygen flow by fuzzy logic,although it has been suggested as feasible and ideallysuited for the technique [Martin, 19941.The present project is aimed at designing a decisionsupport system to control the supply of oxygen to the pa-tient during LFCLA, based on estimation of the oxygenvolume used by the patient. For the first stage, it was as-sumed that only oxygen is to be supplied to the patient.This neglects the other gases, such as nitrous oxide whichis frequently administered to the patient. Other gases willbe incorporated into the system at a later stage. It is en-visaged, that after the system has been successful in thetraining of anaesthetic registrars and has gained theirconfidence in LFCLA, it will be extended to control allthe anaesthetic gases and be fully closed.2. The principles of LFCLA

    During LFCLA, the gases exhaled by the patient arerecycled back to th e patient and the volume of fresh ga s isdelivered to the patient to matched the volume of gas ab-sorbed by the body. The ch anges in oxygen uptake can beseen visually by changes in the reservoir bag [Morris,19941. The bag reflects the condition of the patient andcan provide warnings of potential hazardo us events.In an oxygen limited system, the oxygen used by thebody (OXUSE) leaves the circuit and be replaced by theoxygen supplied to the circuit (OXSUP). Thus, the oxy-gen concentration remains constant, ie. OXSUP =OXU SE. Otherw ise, the difference between the two val-ues equals the rate at which the bag volume is changing(DEL TAV OL). During inspiration the gas moves out ofthe bag and into the lungs while during expiration themovement is reversed. If the volume in the circuit is un-changed, the volume of the bag will return to the samevalue each time. However, when OXSUP > OXUSE, thetotal gas in the circuit will increase and the end-expiratory bag volume (BAGVO L) will increase. On theother hand, when OXSUP < OXUSE, the reverse will oc-cur and BAGVO L will decrease. The changes inBAGVOL and DELTAVOL are estimated by the anaes-thetist, visually assessing the reservoir bag, and utilisedto adjust the OXSUP. It is done in an approximate man-ner without know ledge of the exact relationship betweenthe changes in BAGVOL, DELTAVOL and OXSUP.Thus, by experienc e, the a naesthetist gives a little moreor a little less oxygen, according to the values of the twoinput variables. It is obvious from the approximate way

    control is achieved, that fuzzy logic is a good candidatefor the control of oxygen supply during LFCL A.The oxygen concentrations in the inspired and expiredgas also help the anaesthetist control the system. Mon i-toring these, allows the anaesthetist to differentiatebetween a leak , an ove rfill, an obstruction, entrainment ofair,a metabolic change in oxygen usage, cardiovascularsystem changes and other miscellaneous problems withthe patient or in the anaesthetic circuit.3. Fuzzy control System design3.1 The Design Objectives & Requirements

    There were three key objectives in the design:-1.2.

    To regulate the oxygen flow to the anaesthetics cir-cuit according to the patients oxygen consumption.To detect faults within the system and generatewarning alarms when the oxygen supplement valuesare inappropriately high or low parameters values.To display diagnostic messages for possible faultswithin the system, inferred from the warning alarms.3.

    3.2 Data AcquisitionIt was decided to use the knowled ge of one experiencedThe acquisition was performed in steps :-anaesthetist in the design the first version of the system .1.

    2.3.

    The com ponents of the decision-mak ing process andtheir relative importance,The categorisation of inputs and output parameters,including their data ranges and attributes,The establishment of the series of linguistic rulesused to evaluate each output parameter evaluation.Due to the rule of thumb me thod used and the processbeing evaluated in the mind of the anaesthetist the acqui-sition was very challenging and time-consuming due tobeing a difficult to formulate on paper.

    3.3 The Design StructureThe design of the controller is based upon the decision-making process of the anaesthetist. Presently the oxygensupplemen t is controlled e mpirically by the m onitoring ofseveral variables. The g as supply is then manually regu-lated, according to changes in these variables.From repeated evaluation sessions with experts, it was-clear that the visual observation of the reservoir bagcharacteristics were the most important part of the an-aesthetists evaluation process in determ ining the requiredoxygen supplement (see Figure.1). Other parameterssuch as oxygen concentration in inspired and expired air

    1616Authorized licensed use limited to: UNIVERSIDADE FEDERAL DO RIO DE JANEIRO. Downloaded on December 25, 2008 at 17:58 from IEEE Xplore. Restrictions apply.

  • 8/3/2019 A Fuzzy Logic Based Decision Support System for Low-flow Closed-loop Anaesthesia

    3/6

    F U Z Z - I E E E ' 9 7

    are importan t but only play a secondary role when 1 00%oxygen is used withou t nitrous oxide.

    Figure 1. The co ntrol methodTo reflect this graded decision-making process, theControl System is divided into two blocks or modulesshown in Figure.2. The first module ( 2 odule 1) de-

    termines the required oxygen supplement and possible'alarming' problem s and the second module (FZModule2 ) determines (if necessary) the cause of the possible'alarming' problems in the circuit.

    O X T 'WGVOL*

    OBSOFILLLEAKENTRMETABcvs

    OTHER

    Figure 2. Fuzzy Logic Control SystemThe main characteristics of the reservoir bag are thereservoir bag volume (BAGVOL) and the rate of changein the reservoir bag volume (DELTAVOL). These signalsare the two input signals to FZModule 1. From thesesignals the required oxygen supplement (OXSUP) is in-ferred. In addition to this output signal, are the two

    'alarm' signals (Alarm1 & Alarm2). These alarms indi-cate the detection of faults within the system. Alarms aredetected by the monitoring of the relationship between thetwo input parameters. If a divergence from the appropri-ate response occurs in the parameters then a fault isdetected and an alarm is generated. Alarm1 and Alarm2indicate inappropriately high or low values, respectivelyin the input parameters.The FZ Module 2 is designed to provide beneficial di-agnostic messages to the anaesthetist, informing him/herof possible faults detected in the system. The precise na-

    ture of the 'alarms' are, thus, determined . This isachieved by correlating the output signal alarms from FZModule 2 against the oxygen concentration values in in-spired (INSO) and expired (EXPO) air, and the bagvolume (BAGV OL). Th e diagnostic messages includeobstructions (OBS), overfilling (OFILL), leakage(LEAK), and entrainm ent (ENTR) in the system and me-tabolism (ME TAB) and cardiovascular (CVS) and other(OTHER) problem s with patient.Several assumptions were made in the design evalua-tion. These were that the ciircuit is leaka ge free, the lime-Soda Canister totally absorbs all the carbon dioxide fromthe circuit, the inhalation agents have negligible effectson oxygen concentrations and circuit volumes, and thereis no noise or disturba nces in the system.3.4 Design Completion

    The design and implementation of the Fuzzy LogicControl System (see Figure 2) was executed usingO'INC A Design Framew ork. The advantages of O'INCADesign Framework were:- no limit on the complexity ofthe application; easy to use g raphical user interface; sup-port of pure fuzzy logic: built-in simulation anddebugging facilities; automatic design validation and er-ror location; available DDE applications for the entireproject or individual modules and ease of understandingthe hierarchal design and representation. 0' INCA soft-ware is designed and supplied by Intelligent Machines,Inc. 11 53 Bordeaux Drive, Sunny vale, CA 94089, USA.1.2.3.4.

    The completion of the design involved :-The determina tion of Fuzzy Para meter data rangesand attributes,The design of the linguistic Fuzzy RulebasesThe design of the Fuzzy M embersh ip function forall inputs and outputs, andThe determination of methods of Fuzzification,Inference and Defuzzification.3.4.1 Parameter data ranges and attributesThe linguistic parame ters and labels are determined fromthe anaesthetists' knowledge base. The parametersBAGVOL, DELTAVOL, INSO, EXPO and OXSUP aredivided into fuzzy categories (labels) which reflect the an-aesthetists classification of the data ranges (see Figure.3).These categories are :-NC = n o c h a n g eDS = slow I small d e c r e a s eDL = l a r g e d e c r e a s eD F = f a s t d e c r e a s eIS -IL = l a r g e i n c r e a s eIF - fas t i ncrease

    slow / s m a l l i n c r e a s e

    1617

    Authorized licensed use limited to: UNIVERSIDADE FEDERAL DO RIO DE JANEIRO. Downloaded on December 25, 2008 at 17:58 from IEEE Xplore. Restrictions apply.

  • 8/3/2019 A Fuzzy Logic Based Decision Support System for Low-flow Closed-loop Anaesthesia

    4/6

    F U Z Z - I E E E 9 7

    Figure 3 - The basic membership functionor examnle the data ranse for B AGVOL is below :Parameter Description :-Unit of Measure :- i percentageData Range :- i -100 (decrease by 100%) o

    R e hange in volume of thei reservoir bag......................................................................................................

    .............................................. ....................................................i 100 (increase by 100%)

    Fuzzy Categories :- i NC , DS, DL , IS, ILCategory Data Ranges :- i N C -10 : 0 : lo( lower :middle : upper ) i DS -20 : -10 : 0

    DL -100 : -20 : -10i IS 0 : 10 : 20i IL 10 : 20 : 100

    The w arning alarms and diagnostic messages are YESand NO outputs and are classified over a range 0 to 1.This is a comm only used categorisation for YES /NO fns.3.4.2 The Linguistic Fuzzy Rulebases

    Th e decision making pro cess of the anaesthetist resem-bles IF...THEN rules formulated by the anaesthetist todetermine the amount of oxygen supplement required andwhether an alarm is necessary. The IF...THE N rules weretabulated and ar e visually represented in FAM form. TheFAM is an Fuzzy Associative Memory Bank which pres-ents a set of IF...HEN control rules for the controller.The FAM is a technique used to visually simplify the de-sign and optimise of the rule base, and is used for thegeneration of a rulebase for both modules.M O DUL E 1The distinct block diagram for the Fuzzy Module 1 isshown in Figure.4.The control rules are viewed as linguistic conditionalstatements and have the protocol where each consists oftw o antecedents with N control rules,IF antecedent, AND antecedent, THEN consequencekIn Module 1 there are three conditional statements :where k = 1,2, ._.,NIFBAGVOLk AND DELTAVOLk THEN OXSUP,IF BAGVOLk AN D DELTAVOLk THE N AlkIF BAGVOLk AN D DELTAVOLk THEN a kThis module includes linguistic rules such as IF a smallincrease in BAGVOL AND a slow increase inDELTA VOL, THEN there is a small decrease in OXSUP.This knowledge base used for Fuzzy Module 1 is sum-marised in the FAM in Table.1.

    Figure. 4 - Th e block diagram of Fuzzy Module 1DELTAVOL

    DF DS NC IS IF1 a 1 IS I NC lDF(AZ)IDF(A2)1DF(A2)

    BAGVOL

    Table 1 . FAM for Fuzzy Module 1.M O DUL E 2shown in FigureS.The distinct block diagram for the Fuzzy Module 2 is

    BAGVOL

    omORLLLEAKENTRMETABcvsOTHER

    Figure. 5 - The block diagram for Fuzzy Module 2This rulebase is more complicated where the protocolIn Modu le 2 there are seven conditional statements :IF A1 AND A2 AND WSO AND EXPO ANDIFA1 AND A2 AND INSO AND EXPO AND

    consists of five anteced ents with N co ntrol rules.

    BAGVOL THENOB&BAGVOL THEN OFILL,F A 1 AND A2AND INSO AND EXPO ANDBAGVOL THEN O T H E hThis module includes linguistics rules such a s IF A1 isyes, AND A2 s no, AND B slow increase in INSO , ANDno change in EXPO, BACV OL is no change THENLEAK is yes. A portion of the rulebase is shown below inTable.2.

    1618

    Authorized licensed use limited to: UNIVERSIDADE FEDERAL DO RIO DE JANEIRO. Downloaded on December 25, 2008 at 17:58 from IEEE Xplore. Restrictions apply.

  • 8/3/2019 A Fuzzy Logic Based Decision Support System for Low-flow Closed-loop Anaesthesia

    5/6

    F U Z Z - I EEE9 7

    Table 2 - A portion of the rulebasefor F u u y Module 23.4.3 The Fuzzy Mem bership FunctionsTh e simple linguistic rules that describe the anaesthetistsactions and membership functions were tested systemati-cally in several runs. By design some of the systemparameters had similar data attributes which meant itcould be simplified by eliminating unnecessary duplica-tion of membership functions. A simplified triangularmembership function was used and then optimised duringtesting. An optimised BAGVO L mem bership function isshow n below in Figure.6.

    Dala.100 Degree 1

    -1W -80 011 -6000 -40.00 -10.00 0.00 20.00 40.00 6000 Bo 00 100change m ihe bag mlume (%)Figure.6 - The membership function for BAGVOL

    Figure 7- The optimisation

    L619

    3.4.4 The Fuzzification, Defuz zification and Infer enceSeveral methods were tested in the o ptimisation. Th efinal fuzzication inference com position selected was sum-product for Module 1 and max-min for Module 2 meth-ods, and the centre of g ravity method for defuzzification.

    3.5 OptimisationThe optimisation technique utilised the trial and errormethod. Parameters to be optimised were the controlrules and membership functions. Th e optimisation proc-ess was accompanied by the constant re-evaluation of the

    expert knowledgebase (see Figure.7).(1) The comp leteness of the control rules :There are four imp ortant aspects of design:-That the controller generated control for any and allinput fuzzy states. The ]possibleomission of a co ntrolrule was eliminated by die use of FAM representationof the formulation and design of the linguistic rules.(2) The consistency of the co ntrol rules :There cannot be an impllemental an do r contradictoryinformation in the control protocols that may lead tounexpected and unsatisfactory results.

    Must ensure that all control rules and the fuzzy setsmust not be too precise and must cover the range ofthe universe of control.(4) The robustness of the coatroller :

    This requires the tolerance of the fuzzy controller fornoise and disturbance in the system. Fuzzy setsforming the fuzzy partiuon for the input variable ex-hibit a certain noise immunity over Booleancounterparts and absorbs a notable portion of noise.

    (3) Interaction of the control rules :

    Authorized licensed use limited to: UNIVERSIDADE FEDERAL DO RIO DE JANEIRO. Downloaded on December 25, 2008 at 17:58 from IEEE Xplore. Restrictions apply.

  • 8/3/2019 A Fuzzy Logic Based Decision Support System for Low-flow Closed-loop Anaesthesia

    6/6

    F U Z Z - I E E E 9 7

    4. The user interfaceI F U M OGIC D E U S f f lCOHTROLLER I

    PATIENT ANAESTHETIST

    Figure.8 - Fuzzy Logic Decision Support System.The optimised OINCA driven fuzzy control system isintegrated with a Visual Basic interface to produce theFuzzy Logic Decision Support System (see Figure. 8).The interface performs of data acquisition, display, c on-trol and storage. It receives input parameters from theuser via the console or datafileand displays these on theInput Display of the PC using dynamic data exchange(DDE ) methods. At the controller they are inferred andreturned to the Output Display of the user interface.The oxygen supplement, the possible acoustic warningalarm s and diagnostic messages are displayed for the user.This system provides a basis for the clinical trial evalua-tion of the controller and adaptation for real-time clinicalusage. Provided with the user interface is a tutorial pro-gram with a number of quizzes which are incorporated inthe education program of trainee anaesthetists.

    5. ConclusionsThis study has demonstrated that the application ofFuzzy Logic to the controlling of LFCLA is an appropri-ate and a viable one. The imprecise nature of the input

    data and the non-linear and complex nature of the anaes-thetists reponses during L FCLA a re ideally suited to themethods of Fuzzy Logic as compared to the classical con-trol systems using PID methods.The resulting controller is elegant and simulates to allspecifications. The most obvious benefit in the develop-ment of a Fuzzy Controller is that the amount of designtime required is significantly reduced. It is not necessaryto produce a complex and explicitly defined mathematicalmodel and a Fuzzy controller can be applied successful toill-defined processes by c oding directly in to the controlprotocol of the fuzzy controller. Thus the knowledge baseof the anaesthetist is much more adaptable to the designof the fuzzy controller rather than to the PID or adaptivecontrollers previously attempted. The adaptive controllerhas been successful but the controller and design processis complex with a set of m ass equations required to solve

    the control process. These mass equations are difficult toestablish and provide no room for noise interference.Thus the fuzz y controllers are generally more rob ust.The extensive clinical testing of the controller willcontinue and the optimisation process will be completedfor the regulation of oxygen flow during LFCL A. Duringclinical testing the Fuzzy controller will be run along sidean anaesthetist during a Closed Circuit Anaesthesia. Theadvised responses to the system given by the controllerand displayed on the user interface will be compared tothe actual responses of the anaesthetist. It will be neces-sary to test the controller using a num ber of anaes thetistsand a wide range of patients in order to evaluate ade-quately allpossible operating conditions.The ultimate extension is for the total integration of thefully optimised Fuzzy control System to the re gulation ofinhalation gases during LFCLA delivery during clinicalanaesthesia. This integration requires the optimisation ofthe present controller using clinical evaluation, the exten-sion of the controller to include anaesthetic agents andnitrous oxide and its optimisation, and then the integra-tion of the controller with an anaesthetic workstation.References

    A.J. Ashbury & Y. Tzabar,Fuzzy Logic - New Ways of thinkingfo r anaesthesia, British Journal of Anaesthesia, 75(1): 1-21995.B.A. Baker, Low Flow and closed circuits, Anaesth I r t l m Care,R.W. Boaden & P. Hutton, The digital control of anaesthetic gasflow, Anaesthesia, 41: 413-418, 1986.R.W. Boaden, P. Hutton, C. Monk, A co mputer controlled anaes-thetic gas mixer, Anaesthesia,44: 665-669, 1989.R. &oh & R. Spintage, Closed Circuit System and other Innov a-tions in Anaesthesia, Springer-Verlag Berlin Hiedelberg, 1986.T.D. East, J.K. Hayes, W.S. Jordan & D.R Westenskow, Com-puter controlled anaesthetic delivery, Med. Intsmment, 18: 224-231, 1984.D.W. Hawes, D.C Ross, D.CWhite& RT Wlcch, Servo control ofclosed circuit anaesthesia,Brit. J. Anaesthesia, 54: 229-30,1982.B.L Hooper, A Fuzzy C ontrol System for the oxyg en flow regula-tion during closed circuit anaesthesia, (1993, Department ofElectrical Engineering, University of Sydney, Undergraduate The-sis ProjectD.A. Linkens, & S.B. Hasnain, Self-organising fuzzy logic controland apllication to mu scle relaxant anaesthesia, IEEproceedings-D , 138(3): 274-284, 1991.J.F. Martin, Fuzzy control in anaesthesia [editorial; comment],Journal of Clinical Monitoring, 10:77-80, 1994.S . Oshita, K. Nakakimura, R. Kaieda, T. Murakawa, H. Tamura,I. Hiraoh [Application of the concept of fuzzy logistic controllerfor treatment of hypertension during anaesthesia]. [Japan ese].Masui - Japanese J oumal ofAnesthesiology 42 : 185-9, 1993.Y. Sun, I. Kohane, A.R Stark Fuzzy logic assisted control of in-spired oxygen in ventilated newborn infants, Proceedings - theAnnual Symposium on Computer Applica tions in Medical Care,R. Vishnoi, R.J.Roy, Ad aptive control of closed-circuit anaesthe-sia, IEEE Transactions on Bionredical Engineering, 38::39-47,1991.

    22::341-342, 1994.

    756-61, 1994.

    1620