Home

Design and implementation of a second prototype of the intelligent

Contents

1. t iint Report ity TE ehnology oe fn EEE BULL Bra a Design and implemenlell n of a second prototype of the intelligent alarm system in anesthesia by J A Nederstigt EUT Report 90 E 233 ISBN 90 6144 233 8 January 1990 Eindhoven University of Technology Research Reports EINDHOVEN UNIVERSITY OF TECHNOLOGY Faculty of Electrical Engineering Eindhoven The Netherlands ISSN 0167 9708 Coden TEUEDE DESIGN AND IMPLEMENTATION OF A SECOND PROTOTYPE OF THE INTELLIGENT ALARM SYSTEM IN ANESTHESIA by J A Nederstigt EUT Report 90 E 233 ISBN 90 6144 233 8 Eindhoven January 1990 This report was submitted in partial fulfillment of the requirements for the degree of Master of Electrical Engineering at the Eindhoven University of Technology The Netherlands The work was carried out from November 1988 until October 1989 at the Department of Anesthesiology College of Medicine University of Florida Gatnesville Florida under supervision of Professor J E W Beneken Ph D and J J van der Aa M E E CIP GEGEVENS KONINKLIJKE BIBLIOTHEEK DEN HAAG Nederstigt J A Design and implementation of a second prototype of the intelligent alarm system in anesthesia by J A Nederstigt Eindhoven Eindhoven University of Technology Faculty of Electrical Engineering Fig tab EUT report ISSN 0167 9708 90 E 233 Met lit opg reg ISBN 90 6144 233 8 SISO 608 1 UDC 616 089 5 NUG
2. The rule set is extended with rules that perform extra signal validation checks before the malfunction detection starts An Intelligent Alarm System in Ancsthesia 30 CHAPTER 4 TESTING PROTOTYPE I The system described in chapter 3 has been tested in a clinical environment as well as on an anesthesia simulator The protocol and the results of these tests are presented in the next sections At the end of the chapter the limitations of Prototype I and the improvements to be made in a second prototype system are summarized 4 1 Simulator Testing Before going into the operating room OR with the system its performance was tested on the Gainesville Anesthesia Simulator developed by Good et al at the University of Florida Gainesville U S A 27 The simulator consists of an Ohmeda Modulus Il Anesthesia Machine together with an Ohmeda 7800 series ventilator and the standard monitoring equipment used in the OR Using a standard breathing circuit and endotracheal E T tube a mechanical lung is ventilated Carbon dioxide CO is fed to the lung to simulate the CO production of the patient The signals we are interested in airway pressure in the inspiratory hose airway flow through the expiratory hose and partial CO pressure at the Y piece are obtained with exactly the same monitors as they would be in the OR A number of sensors and actuators makes it possible to introduce mechanical malfunctions in the breathing circuit or anest
3. Nederstigt J A DESTON AND IMPLEMENTATION OF A SECOND PROTOTYPE OF THE INTELLIGENT ALARM SYSTEM IN ANESTHESIA EUT Report 90 E 233 1990 ISBN 90 6144 233 8 Philippens E H J DESTONING DEBUGGING TOOLS FOR SIMPLEXYS EXPERT SYSTEMS EUT Report 90 E 234 1990 ISBN 90 6144 234 6 Heffels J J M PATIENT SIMULATOR FOR ANESTHESIA TRAINING A mechanical Jung model and a physiological software model EUT Report 90 E 235 1990 ISBN 909 6144 235 4
4. piratory hose until the pressure in the lungs equals the PEEP value The symbols in figures 5 1 and 5 2 have the following mea ning Figure 5 2 Electrical model for expiration R resistance of the inspiratory respectively the expiratory tubing plus the resistance of the patient s airway C combined compliance of lungs and tubing Fr inspiratory flow Fg expiratory flow Pj the pressure in the lungs An Intelligent Alarm System in Anesthesia 44 Van Oostrom 28 derived the following formulas from this model assuming that the peak pressure in the lungs is equal to Pax and that the resistance of the inspiratory tubing equals the resistance of the expiratory tubing in a no malfunc tion situation Pmin PEEP 5 1 Fmax Vtmea RC 5 2 Pmax R x Fma Pmin 5 3 Psope Vraet Tinsp C 5 4 In this set of formulas Vrae is the tidal volume actually delivered to the lungs via the inspiratory hose whereas VTmea is the tidal volume measured in the expiratory limb of the breathing circle We assume that Vrmea x 5 5 In 5 5 is a constant independent of the settings depends on the tidal volume actually set on the ventilator on the fresh gas flow setting FGF ml sec and on the set PEEP value We assume that the fresh gas flow is completely added to the tidal volume during the inspiratory time and is mixed with the expired gases during expiration S
5. 8 female and 3 male patients were involved Data from the three monitors were recorded and the system s performance was evaluated in the hostile and noisy OR environment Of course no malfunctions could be introduced we only tested the system for false alarms To show how the signals are influenced by external or patient conditions in the OR two examples are given in figures 4 2 and 4 3 In figure 4 2 the CO signal is disturbed by cardiogenic oscillations the heart rhythm influences the CO signal at the end of the expiration phase As a result the CO down slope feature will be unreliable In figure 4 3 the pressure signal is disturbed because the surgeon is pushing on the patient s chest The pressure signal is invalid for one or two breath periods As can be seen from the two figures the signal processing algorithms must be very robust because clinical signals are by far not as smooth as those measured with the simulator m CO mmHg 20 25 Time sec Figure 4 2 Example of a disturbed CO waveform An Intelligent Alarm System in Anesthesia 37 RR Pressure cmH20 Ten 20 25 Time sec Figure 4 3 Example of a disturbed pressure waveform The results of the OR tests can be summarized as Like on the simulator the system could be fooled when the anes thesiologist changed the ventilator settings and or the fresh gas flow and the feature baselines were not subsequently reset to their new running average
6. Sometimes when the anesthesiologist wants to see a real time waveform blood pressure is measured invasively with an arterial catheter Other variables monitored are temperature typically measured with a ther mocouple or thermistor based sensor and the degree of muscle relaxation 2 Despite this abundance of mechanical and electrical devices available to help the anesthesiologist his own eyes ears and sense of feeling are still the most important monitors available 2 7 Thus the alarm system described in this thesis should be considered as an attempt to provide the anesthesiologist with valuable extra information when time an extremely important factor in emergency situations is limited It is certainly not an attempt to replace the anesthesiologist An Intelligent Alarm System in Anesthesia 8 CHAPTER 2 ALARM STRATEGIES IN ANESTHESIA Before describing our approach to design an intelligent integrated alarm system we review the work that already has been done in this area In the next paragraphs an overview of the literature of the last decennium about alarms and integrated alarm systems in anesthesia is presented A brief introduction to our strategy is given at the end of this chapter 2 1 Current State of Alarm Technology The purpose of alarms during anesthesia is to get the clinician s attention whenever a potential hazard is detected regarding the patient or the anesthesia equipment For this purpose the monitors use
7. diagnosed as obstruction E T tube Y piece after the tube was suctioned the alarm disappeared Although the system is not designed to support manual ventilation it should not generate all kinds of alarm messages in those instances During the OR tests the prototype I system only generated CAUTION messages during periods of manual ventilation which means that no alarm was triggered for more than one consecutive breath period Two disconnects in the E T tube were diagnosed correctly as was a not completely inflated E T tube cuff The latter one was recognized as a small leak E T tube Y piece No critical event went by undetected The system could easily keep up real time with pediatric respiratory rates of 20 breaths min or more Also the low pediatric tidal volumes did not present any problems for the breath detection algorithms 4 3 Final Conclusions about the First IASA Prototype As a general conclusion prototype I worked as expected on simulator data as well as on noisy compared to the simulator patient data Tests showed that the complete system was about three times as fast as real time on an 8 MHz IBM AT compatible computer with a coprocessor and an EGA videocard installed An Intelligent Alarm System in Ancsthesia 39 Data was read from a file rather than obtained from the monitors during these speed tests However some improvements stil have to be made The most important limitation of the fir
8. schematically pictured in figure 4 1 1 Figure 4 1 is based partly on a drawing of the anesthesia system by J S Gravenstein MD An Intelligent Alarm System in Anesthesia 32 4 1 1 Test Protocol and Results Using 4 different combinations of ventilator and fresh gas flow FGF settings a number of malfunctions was introduced one at a time During a maximum of 30 seconds or 5 breath periods whichever came first the alarm system was expected to detect a malfunction and generate the correct alarm message After that period the system was brought back into the no malfunction state and subsequently the next critical event was simulated Whenever settings were changed we waited until the system had adapted to the new signals this usually takes about 5 or 6 breath periods and then reset the feature baselines to their new running average value see 3 3 This was necessary in order to avoid false alarms or missed detections due to the fact that the baselines were not adequate for the setting combination The different combinations of ventilator and FGF settings are given in table 4 1 The compliance of the mechanical lung was set to 0 1 l cmH5O which resembles a normal lung compliance value of an adult patient The malfunctions introduced by the simulator were incompetent expiratory valve incompetent inspiratory valve exhausted CO absorber disconnect of the ventilator hose CO canister leak and a leak in the E T tube cuff Man
9. 206 1988 ISBN 90 6144 206 0 Schuurman W and M P H Weenink STABILITY OF A TAYLOR RELAXED CYLINDRICAL PLASMA SEPARATED FROM THE WALL BY A VACUUM LAYER EUT Report 88 E 207 1988 ISBN 90 6144 207 9 Lucassen F H R and H H van de Ven A NOTATION CONVENTION IN RIGID ROBOT MODELLING EUT Report 88 E 208 1988 ISBN 90 6144 208 7 J zwiak L MINIMAL REALIZATION QF SEQUENTIAL MACHINES The method of maximal adjacencies EUT Report 88 E 209 1988 ISBN 90 6144 209 5 Lucassen F H R and H H van de Ven OPTIMAL BODY FIXED COORDINATE SYSTEMS IN NEWTON EULER MODELLING EUT Report 88 E 210 1988 ISBN 90 6144 210 9 Boom A J J van den Ho CONTROL An exploratory study EUT Report 88 E 211 1988 ISBN 90 6144 211 7 Zhu Yu Cai ON THE ROBUST STABILITY OF MIMO LINEAR FEEDBACK SYSTEMS EUT Report 88 E 212 1988 ISBN 90 6144 212 5 Zhu Yu Ca Driessen A A H Damen and P Eykhoff A NEW SCHEME FOR IDENTIFICATION AND CONTROL EUT Report 88 E 213 1988 ISBN 90 6144 213 3 Bollen M H J and G A P Jacobs IMPLEMENTATION OF AN ALCORTTHM FOR TRAVELLING WAVE BASED DIRECTIONAL DETECTION EUT Report 89 E 214 1989 SBN 90 6144 215 1 Hoei jmakers M J en J M Vleeshouwers EEN MODEL VAN DE SYNCHRONE MACHINE MET GELIJKRICHTER GESCHIKT VOOR REGELDOELE i NDEN EUT Report 89 E 215 1989 ISBN 90 6144 215 X Pineda de Gyvez J LASER A LAyout Sensitivity ExploreR Report and user s manual EUT Report 89 E 216 1989 ISBN 90 614
10. 3 gives 0 7 lt z k D z k lt 1 for T gt 0 15 sec 3 4 y t Yotart exp t T z 0 Istart Figure 3 4 Graph of an exponentially decreasing time series With T 1 F we get from 3 3 1 T F x In z k 1 z k 3 5 Suppose there are n samples available in the decreasing exponential curve called z 0 up to z n 1 An estimation of the inverse time constant is given by the following average value from 3 5 An Intelligent Alarm System in Ancsthesia 23 VT F x n 1 x 5 z i 1 z i 3 6 Finally an estimation of the time constant is given by Tc estimatea 1 lt 1 T gt 3 7 After every new sample z i the algorithm divides z i by the previous sample 2 1 1 and uses a look up table with the values of In 0 700 In 0 701 In 0 702 1n 0 999 to get the natural logarithm of the result This way an updated estimation of the value of 1 T can be obtained after each sample by updating formula 3 6 If by accident for example due to noise z i gt z i 1 the algorithm waits for z i 1 divides z i 1 by z i 1 and if the result lays in between 0 7 and 1 it looks up the natural logarithm of the result The logarithm is then added to the sum in 3 6 twice because the time between z i 1 and z i 1 is two times T This process goes on until a value z i j is found so that z i j divided by z i 1 lays in between 0 7 and 1 This method implies that when very oft
11. Because the tidal volume loss due to the leak took place entirely downstream of the flow sensor in this case the flow signal indicated no or very little volume loss Therefore no leak message was generated Leaks at that position turned out to be very difficult to detect without extra information This malfunction is not immediately clinically dangerous however Without the leaks at site 2 in the expiratory hose 95 of the malfunctions was detected correctly As a last experiment the speed of the alarm system program was tested by gradually increasing the respiratory rate RR setting on the ventilator It turned out that the software could keep up with RR values as high as 60 breaths min This is much higher than normal rates used during clinical anesthesia up to 30 breaths min for pediatrics Fault detection performance was not tested for higher RR values This will be done with the prototype II system An Intelligent Alarm System in Anesthesia 36 4 2 OR Testing As a second test the system was taken to the OR The CO pressure and flow monitors were connected to the breathing circuit in the same way as pictured in figure 4 1 during 11 surgery cases of different type The anesthesia machine used was an Ohmeda Modulus Il The operations included pediatrics airway nose surgery abdominal surgery heart surgery liver surgery ankle surgery with light anesthesia and eye surgery The ages of the patients varied from 12 months to 75 years
12. Fukui Y n expert alarm system Ibid p 203 209 Beneken J E W and J S Gravenstein Sophisticated alarms in patient monitoring Ibid p 211 228 Philip J H verview creating practical alarms for the future J Clin Monit Vol 5 1989 p 194 195 Fukui Y and T Masuzawa Knowledge based approach to intelligent alarms Ibid p 211 216 An Intelligent Alarm System in Anesthesia 63 E15 16 17 18 19 20 211 22 23 241 25 26 27 28 Beneken J E W and J J van der Aa Alarms and their limits in monitoring Ibid p 205 210 Schreiber P J and J Schreiber Structured alarm systems for the operating room Ibid p 201 204 Meijler A P Automation in anesthesia a relief A systematic approach to computers in patient monitoring Ph D thesis Eindhoven University of Technology 1986 Berlin New York Springer 1987 Brunner J X and D R Westenskow P Zelenkov Prototype ventilator and alarm algorithm for the NASA space station J Clin Monit Vol 5 1989 p 90 99 McEwen J A and L C Jenkins Complications of and improvements to breathing circuit monitors for anesthesia ventilators Med Instrum Vol 17 1983 p 70 74 McEwen J A and C F Small L C Jenkins Detection of interruptions in the breathing gas of ventilated anaesthetized patients Can J Anaesth Vol 35 1988 p 549 561 Saunders R J and W R Jewett System
13. Monitor AD board Anesthesia machine Ventilator Flow signal calculated from Pressure signal from CO signal from Fresh gas flow calculated from 2 Software Size of complete executable program IBM AT compatible 8 MHz coprocessor installed EGA or VGA High Resolution Graphics Monitor Data Translation 2811 Ohmeda Modulus II Ohmeda 7810 Mechanical Ventilator Ohmeda 5410 Volume Monitor Ohmeda 5500 Airway Pressure Monitor Ohmeda 5200 CO Monitor Ohmeda 5410 Volume Monitor 118 kB 83 kB code 35 kB data stack and others Sampling frequency for real time signals 20 Hz Program speed Compiler Expert system tool An Intelligent Alarm System in Anesthesia 3 times as fast as real time on the 8 Mhz computer Microsoft C Optimizing Compiler version 5 1 SIMPLEXYS Expert System Language 72 Eindhoven University of Technology Research Reports ISSN 0167 9708 Faculty of Electrical Engineering Coden TEUEDE 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 Butterweck H J and J H F Ritzerfeld M J Werter FINITE WORDLENGTH EFFECTS IN DIGITAL FILTERS A review EUT Report 88 E 205 1988 ISBN 90 6144 205 2 Bollen M H J and G A P Jacobs EXTENSIVE TESTING OF AN ALGORITHM FOR TRAVELLING WAVE BASED DIRECTIONAL DETECTION AND PHASE SELECTION BY USING TWONFIL AND EMTP EUT Report 88 E
14. Utah by Brunner et al 18 already incorporates some An Intelligent Alarm System in Anesthesia 12 intelligence A large number of transducers is placed at different sites of the breathing circuit and ventilator in order to identify several malfunctions during mechanical ventilation Specific alarm messages can be generated by combining the signals coming from the different monitors via if then else rules Early results indicate 98 correct identification of mechanical faults The large number of sensors prevents the system from being commercially attractive at this moment Furthermore an implementation of a breathing gas interruption monitor for use in the breathing circuit during mechanical ventilation is described by McEwen et al 19 20 This monitor can detect different hazardous situations in the breathing circuit by measuring only the pressure waveform The waveform is compared to a learned correct waveform and when differences are detected an alarm message is generated The alarms are not specific however and no multisignal analysis is performed Other papers present only general conceptional or philosophical descriptions of system integration and use of computers for alarms in anesthesia 21 22 23 2 3 The Gainesville Approach As demonstrated in the beginning of this chapter there is a clear need for a system that is able to detect and identify abnormalities in the patient machine combination during anesthesia This has t
15. X Obstr E T tube XXXX X X X Obstr exp hose XXXX X X X Obs r vent hose XXXX X X X Smali leak E T tube FXXX X X X Small leak Y piece XXXX X X X Smali leak insp hose XXXX X X X Small leak 1 exp hose XXXX X X X Small leak 2 exp hose NNNN N N X Small leak vent hose NNNX N X X Large leak E T tube FXXX X X X Large leak Y piece NXXX X X X Large leak insp hose NXXX X X X Large leak 1 exp hose NXXX X X X Large leak 2 exp hose NNNN N N N Large leak vent hose XXXX X X X E T tube cuff leak XXXX X X X CO canister leak XXXX X X X Disc FGF hose XXXX X X X Disc vent hose XXXX X X X Disc insp hose XXXX X X X Disc exp hose XXXX X X X Disc Y piece XXXX X X X Disc E T tube XXXX X X X An Intelligent Alarm System in Ancsthesia 35 Table 4 3 Expert system rules for thrce malfunctions INCOMP INS VALVE A stuck inspiratory valve is detected FLW EXP VOL DOWN AND FLW MAX DOWN AND CO2 DO STR DOWN AND PRS MAX NORMAL SMALL LEAK A small leak is detected FLW EXP VOL DOWN AND NOT MAX DOWN AND MAX NORMAL LARGE LEAK A large leak is detected FLW EXP VOL DOWN AND PRS MAX DOWN malfunction messages are generated incompetent inspiratory valve and small leak This can be done by removing the NOT FLW MAX DOWN part from the SMALL LEAK rule see table 4 3 Of the not detected malfunctions 13 were leaks introduced directly downstream of the flow monitor but still upstream of the expiratory valve
16. high PEEP values Currently research is conducted to improve the system regarding the latter two issues Finally the system was able to generate sensible messages when multiple malfunctions were introduced at the same time A disadvantage of the current system is that faults present before the start of anesthesia can be accepted as normal if undetected by the clinician An Intelligent Alarm System in Anesthesia 59 Regarding program speed tests showed that the system runs about 3 times as fast as real time on an 8 MHz IBM AT compatible computer with coprocessor installed The system could keep up with respiratory rates RR up to 60 breaths min on the simulator which is twice as high as the highest RR values regularly used in clinical practice All tests indicated that the second IASA prototype has after further refinement and testing the potential to be the platform for the development of a highly integrated intelligent alarm system for use during anesthesia The real time expert system approach proved relatively easy to implement and offers many expansion possibilities This future alarm system can be a very useful help to the anesthesiologist during the course of anesthesia since it suppresses unnecessary alarms and in case of mishaps it can shorten the time he needs to come to a diagnosis 7 2 Recommendations for Future Research Currently we are developing a third prototype which will also detect some malfunctions occurri
17. in the E T tube is detected PRS MAX UP AND PRS SLOPE UP AND NOT FLW MAX UP AND FLW T CONST UP THEN DO write alarm Obstruction Y piece or E T tube THEN FA OBST INSP HOSE OBST EXP HOSE PRS MAX UP The maximum pressure feature is above the normal band BTEST Maxpres UP valves INC VALVE or obstructions OBSTRUCTION somewhere in the circle The OBSTRUCTION rule is again composed of subrules that look at the site of the obstruction like OBST ET TUBE Finally at this level the system looks at the symbolic feature values that are fed to the expert system to come to a conclusion Negatives in the rules are added to make them unique or to prevent unnecessary alarms For example when the pressure goes slightly up but the expired maximum flow value is also higher than normal the patient is still ventilated well so no alarms need to be generated yet The THEN DO section of OBST ET TUBE provides a hook to C if the rule gets assigned a value TR the C code on the rest of the line is An Intelligent Alarm System in Anesthesia 28 executed In this case this means that an alarm message is put on the screen The rules after THEN FA are immediately set to FA without evaluation The word BTEST provides a second hook to C the C code on the remainder of the line is executed and the rule is set to the result of this boolean test So the rule PRS UP will get the value TR when the C variable Maxpres one of the sym
18. method turned out to be reliable and stable and were practically the same as the slope values acquired with the old least squares algorithm whereas the speed of the new method is considerably higher In the following the technique used for estimating the time constant of the exponential down stroke in the pressure and flow signals is explained 3 2 2 1 How to Estimate the Time Constant We assume that the analog pressure signal y t enters state 3 see figure 3 3 at t 0 sec and exits this state at t Teng for the flow signal a similar state is implemented So the following formula is valid y t Ysrar expCUT for O lt t lt Tend 3 1 The time constant T of this signal has to be estimated from z k a sampled version of y t z k y kT for k 0 1 2 Tong Ts 3 2 In 3 2 T is the known sample time which is equal to 1 F with F the sample frequency In our case F 20 Hz In figure 3 4 a graph of y t and its samples y kT is pictured From 3 1 and 3 2 we see that when two consecutive samples z k and z k 1 are divided the result is a constant independent of k z k 1 z k exp T T 3 3 Tests on pressure and flow signals recorded in the OR showed that T almost always lays in between 0 3 and 1 5 sec In our algorithm we take an extra An Intelligent Alarm System in Anesthesia 22 safety margin of 50 of the lower boundary value and assume that T gt 0 15 sec Filling this in in 3
19. software requests these data Measured values come from several sensors connected to the ventilator an oxygen Os sensor near the fresh gas outlet a tidal volume sensor in the expiratory limb and a pressure sensor in the inspiratory limb of the breathing circle The ventilator is connected to serial port COMI on the computer used in our prototype Interrupt code is written that requests new data from the ventilator at the start of each breath period The incoming string of characters is buffered and after breath detection converted to numerical values An Intelligent Alarm System in Anesthesia 46 This way an updated set of ventilator settings is available before each expert system run In order to measure FGF a second volume monitor is used Ohmeda 5410 The sensor is inserted into the fresh gas hose see figure 4 1 Similar to the expiratory flow signal this monitor generates a pulse for approximately every 3 ml of gas that passes the sensor The FGF pulse and direction signals are connected to two unused pins at serial port COM2 in the computer the expiratory flow signal also comes in at COM2 The interrupt code is modified so that whenever COM2 generates an interrupt the system first looks from which flow monitor the pulse originated Thereafter the respective counter is increased or decreased depending on the directional signal After breath detection the number of FGF pulses is multiplied by 3 ml and divided by the breath time
20. to an analog to digital AD board inside an IBM AT compatible computer With the AD board sampled versions of the signals are produced The samples serve as input to the signal processing software Van Oostrom 1 describes how a real time flow signal is created by counting pulses generated by the Ohmeda 5410 Volume Monitor Tests showed that a sampling frequency of 20 Hz is enough to calculate all signal features even at the highest respiratory rates sometimes an RR of up to 30 breaths min is used in pediatric or neonatal anesthesia A typical example of the three real time waveforms in a no malfunction situation during mechanical ventilation of an adult patient is given in figure 3 2 During patient inspiration the pressure signal will go up linearly due to the constant flow the ventilator forces into the lungs There is no flow in the expiratory limb of the circuit during inspiration When the positive pressure from the ventilator is removed the expiratory valve opens and the flow through the expiratory limb suddenly starts This marks the start of patient expiration see figure 3 2 At this moment CO rich gas from the lungs starts passing the CO Sensor An Intelligent Alarm System in Anesthesia 18 Delay time CO2 mmHg Time sec Time sec 5 Time sec Start of patient expiration Figure 3 2 Example of 3 real time waveforms in an adult patient Since the lungs empty passively the flow and pressure signals sho
21. was changed to 350 ml RR to 20 breaths min and EE to 1 2 5 setting Ic The results of this test sequence are given in table 6 1 As in chapter 4 an X in table 6 1 means that the correct message was generated within An Intelligent Alarm System in Anesthesia 53 30 seconds or 5 breath periods an F indicates that only false alarm messages were generated and an N means that no malfunction message was triggered at all Table 6 1 Results of test sequence 1 for Prototype II at the simulator Malfunction Setting 1a Setting 1b Setting 1c Obstr E T tube X X X Obstr insp hose X X X Obstr exp hose X X X Obs r vent hose X X X Stuck insp valve X X F Stuck exp valve X X X Exh CO absorber X X X Disc FGF hose X X X Disc Y piece X X X Disc vent hose X X X Small leak insp hose X X X Small leak exp hose X X X Disc CO sampling line X X X When after a disconnection of the CO sampling line a Small leak or Large leak message was generated the reaction of the alarm system was considered correct The only false alarm recorded was a Small leak message when the inspiratory valve was stuck At a high RR value the increase in the down slope of the CO signal was not sufficient to trigger the Incompetent inspiratory valve message see also 4 1 1 and table 4 3 The automatic baseline reset worked as expected during this first test The setting combinations used during the second test sequence are given in tab
22. 4 216 8 Duarte J L MINAS An algorithm for systematic state assignment of sequential machines computational aspects and results EUT Report 89 E 217 1989 ISBN 90 6144 217 6 Kamp M M J L van de SOFTWARE SET UP FOR DATA PROCESSING OF DEPOLARIZATION DUE TO RAIN AND ICE CRYSTALS IN THE OLYMPUS PROJECT EUT Report 89 E 218 1989 ISBN 90 6144 218 4 Koster G J P and L Stok FROM NETWORK TO ARTWORK utomatic schematic diagram generation EUT Report 89 E 219 1989 ISBN 90 6144 219 2 Willems F M J CONVERSES FOR WRITE UNIDIRECTIONAL MEMORIES EUT Report 89 E 220 1989 ISBN 90 6144 220 6 Kalasek 1 and W M C van den Heuvel L SWITCH A PC program for computing transient voltages and currents during switching off three phase inductances EUT Report 89 E 221 1989 ISBN 90 6144 221 4 Eindhoven University of Technology Research Reports ISSN 0167 9708 Faculty of Electrical Engineering Coden TEUEDE 222 223 224 225 226 227 228 229 230 231 232 233 234 235 J wiak L THE F LL DECOMPOSITION OF SEQUENTIAL MACHINES WITH THE SEPARATE REALIZATION OF THE NEXT STATE AND OUTPUT FUNCTIONS EUT Report 89 E 222 1989 ISBN 90 6144 222 2 J Zwiak L THE BIT FULL DECOMPOSITION OF SEQUENTIAL MACHINES EUT Report 89 E 223 1989 ISBN 90 6144 223 D Book of abstracts of the first Benelux Japan Workshop on Information and Communication Theory Eindhoven T
23. 52 aen er vEt AE he Manu Rohre SUR TRIN ae x 1 CHAPTER 1 A SHORT INTRODUCTION TO ANESTHESIA 3 1 1 The Anesthesia System 4 4 4 3 1 2 The Most Important Monitoring Equipment 7 CHAPTER 2 ALARM STRATEGIES IN ANESTHESIA 9 2 1 Current State of Alarm Technology 9 2 2 Survey of Modern Alarm Strategies 10 2 2 1 Overview of Implemented Integrated Systems 12 2 3 The Gainesville Approach 13 CHAPTER 3 IASA THE FIRST PROTOTYPE 16 3 1 The Data Flow through the System 16 3 2 Signals and Signal Processing 17 3 2 1 From Monitors to Sampled Signal Waveforms 18 3 2 2 From Signals to Signal Features 20 3 2 2 1 How to Estimate the Time Constant 22 3 2 3 Signal Validation 25 An Intelligent Alarm System in Anesthesia iv 33 symbole Data 5 223 yn an aa era 25 3 4 The Real Time Expert System Approach 26 3 2 Software Upgrades uo oe es Vea ete be em ee an E 29 CHAPTER 4 TESTING PROTOTYPE I 31 4 1 Simulator POSING sa arn ae ee wars ew ix E ER e 31 4 1 1 Test Protocol and Results 52 nr ne 33 AD OR Testing u were ara Sinne a 37 4 3 Final Conclusions about the First IASA Prototype 39 CHAPTER 5 IASA THE SECOND PROTOTYPE 41 5 1 The Automat
24. E INITIALLY TR THEN GOAL SIGNALS OK VENTILATOR OK BREATHING SYSTEM OK LEVEL 1 SIGNALS OK The signals are all valid CO2 SIG OK UCAND PRS SIG OK UCAND FLW_SIG_OK VENTILATOR OK The ventilator is on and running MUST NOT POWER OFF OR VENTILATOR OFF THEN GOAL SETTING CHANGE ACCEPT PEEP CHANGE BREATHING SYSTEM OK No malfunctions in the breathing circuit are detected MUST NOT BS INC VALVE UCOR BS OBSTRUCTIONS UCOR BS LEAKS UCOR BS CO2 ABSORBER UCOR BS_ DISCONNECTS UCOR VENT _ ALARM An Intelligent Alarm System in Ancsthesia 65 APPENDIX 2 DATA FLOW IN PROTOTYPE II In figure I the data flow in the IASA Prototype II software package is somewhat simplified schematically pictured MAIN MODULE Numerical analysis Breath detected Initialization 1 0 initialization Get sample Signal processing update screen Y Update numerical signal features NO Numerical analysis Feature extraction Rule evaluation Selling changed gt M rl Qutput alarms lo screen Reset baselines update constants Figure I Flow chart of the Prototype II software An Intelligent Alarm System in Anesthesia 66 APPENDIX 3 FORMAL DESCRIPTION INTERFACE ROUTINES FUNCTION int ia init io void DESCRIPTION ia init io performs initializing tasks ie the initialization of the serial communication and the user interface screen and the opening of datalogging files This function is
25. I 742 Trefw anesthesie pati nthewaking SUMMARY In today s operating room OR a number of monitors is connected to both patient and equipment Each individual monitor is equipped with its own usually rather unspecific alarms Therefore there is a clear need for an integrated alarm system that by combining data from all monitors generates useful and intelligent alarm messages about possible abnormalities occurring in patient and or equipment A first prototype of such a system combining data from three different monitors to detect mechanical malfunctions in the breathing circuit was tested on an anesthesia simulator and in the OR During tests with the simulator the system identified 88 of the introduced malfunctions correctly within 30 seconds No false alarms were recorded in the OR although the signals were often disturbed A second prototype was designed to adapt automatically to adjustments in one or more of the control settings on the anesthesia machine or the anesthesia ventilator during surgery Furthermore a user friendly interface screen was designed for this system During tests on the simulator 95 5 of the malfunctions was detected correctly The system could keep up with high pediatric respiratory rates After some additional refinements and improvements the system can be the platform for the development of a completely integrated intelligent alarm system for patient as well as equipment during anesthesia A
26. ION ia get breath by breath data fills the structure pointed to by inpointer with the latest complete set of ventilator and fresh gas flow data and gives an indication about the validity of the data set in the return value Also the screen and the logfile are updated with the new settings The function is called before each rule evaluation CALLING CONVENTION status ja get breath by breath data amp breath data in RETURN VALUE 0 when the structure is filled with a new and complete data set 1 when data are invalid or no new data are available yet and 1 when a fatal error occurs during data acquisition An Intelligent Alarm System in Anesthesia 70 FUNCTION int put breath by breath data struct breath by breath data out outpointer DESCRIPTION ia put breath by breath data fills the structure pointed to by outpointer with all the statuses and alarm message strings that were triggered during the last expert system run The highest priority alarms are printed on the user interface screen and the messages are written in a datalogging file The function is called directly after each rule evaluation CALLING CONVENTION status ia put breath by breath data amp expert results RETURN VALUE 1 when a fatal error occurs 1 when one or more non fatal internal errors occur otherwise 0 An Intelligent Alarm System in Anesthesia 71 APPENDIX 4 TECHNICAL DATA PROTOTYPE II 1 Hardware Computer
27. a 42 situations and considerably less steep when for example the inspiratory valve is stuck In the latter case expiration takes partly place through the inspiratory hose and as a result expired CO rich gas is reinspired at the next breath This causes the CO down slope to be less steep than normal The end tidal CO value Peyco2 depends on gas composition minute ventilation and demographic patient data In our first approximation however we assume Peicoz is independent of the settings When the concept of the baseline reset described below is proven a simple patient model may be incorporated in the alarm system This should lead to the automatic calculation of new baselines for P coz and for the slope values after for example the gas composition is changed The expected value for minimum expiratory flow F nin and for inspired Pico will always be zero independent of the setting values The value of the time constant of the pressure and flow down stroke depends on the resistance of the tubing and airway the compliance of the tubing and the compliance of the lungs The first two factors are equipment related and thus constant and completely independent of the settings The compliance of the lungs may change during surgery for example after the chest is opened In our first approximation we elected the automatic baseline reset only to be performed after setting changes not when the patient s condition changes This mean
28. all the numerical feature values are available after breath detection or time out on all three signals an extra cross check is performed to test whether the timing variables of the signals are in accordance with each other For all the valid signals the inspiration expiration and breath times are compared If the data from one signal vary more than 20 from the variables coming from the other two the signal is declared invalid In this way the redundancy that is available is used as good as possible for signal validation 3 3 Symbolic Data Since the expert system needs symbolic data rather than numerical data as its input the calculated numerical features are translated into symbolic format Features that are declared invalid see 3 2 3 get the symbolic value NV Not Valid As Van Oostrom 1 describes the valid feature values are compared to a value that is considered to be normal for that feature in a no malfunction situation This reference value will be called the feature baseline from now on At the beginning of an operation the anesthesiologist currently has to push a RESET BASELINES button when the signals are stationary and he accepts An Intelligent Alarm System in Anesthesia 25 them as normal At that moment all baselines will be reset to the current running average of the feature value thereby assuming that no malfunction is present Resetting the baselines is necessary because the set of initial default baselin
29. an identical way The general result of the preliminary tests of the system behavior during multiple malfunctions is satisfactory More testing has to be performed in this area in the near future however 6 3 Future Testing Although the performance of the Prototype II system during simulator tests was very good it has not been exposed to the hostile OR environment yet Especially the performance of the automatic baseline reset lias to be tested carefully in the presence of noisy signals Currently a third prototype in which some patient related malfunctions will be incorporated is being developed see chapter 7 Since not all patient related mishaps we are interested in can be tested with the anesthesia simulator we decided to test the Prototype III alarm system on anesthetized dogs before taking it to the OR for false alarm tests However the test sequences with dogs and the OR experiments are very time consuming whereas the test protocols for prototypes II and IH will have many aspects in common Therefore we decided to combine the OR test of Prototype II especially the performance of the automatic baseline reset with the dog and OR tests of Prototype III detection performance of patient related malfunctions An Intelligent Alarm System in Anesthesia 58 CHAPTER 7 CONCLUSIONS AND RECOMMENDATIONS 7 1 Conclusions Tests of the first prototype of the Intelligent Alarm System in Anesthesia IASA showed that the system can d
30. bolic features has a value UP This means that the maximum pressure feature is higher than the upper boundary of the normal zone see 3 3 In the same way the whole rule base is constructed as can be seen from table 3 1 this setup is much like a tree structure In the real implementation an alarm message is only put on the screen when the same alarm has been triggered during two consecutive runs When an alarm is detected for the first time a general CAUTION message is generated This is a protection against false alarms due to motion artifacts of the patient or other external disturbances 3 5 Software Upgrades Compared to the preliminary prototype I system that Van Oostrom 1 describes some important changes have been implemented before testing the prototype Some of them have already been mentioned like the new algorithms for slope and time constant estimation the use of an AD board to sample the pressure and CO waveforms 3 2 and the possibility to set the threshold values for every feature separately instead of using a general margin of 2096 8 3 3 Other important changes are mentioned below The flow pressure and CO signal processing is done using fixed point rather than floating point calculation where possible in order to increase program speed An Intelligent Alarm System in Anesthesia 29 A new C version of SIMPLEXYS has become available Since the signal processing as well as the feature extraction sof
31. called only once directly after program start up CALLING CONVENTION status ia init io RETURN VALUE 1 when error s during initialization otherwise 0 An Intelligent Alarm System in Anesthesia 9r FUNCTION int ia exit io void DESCRIPTION ja exit io resets the serial ports and the user interface screen and closes all open files before exiting to DOS The function is called only once before aborting the expert system CALLING CONVENTION status ia exit io RETURN VALUE 0 when normal exit 1 when exit due to error s An Intelligent Alarm System in Anesthesia 68 FUNCTION int ia get real time sample struct real time data inpointer DESCRIPTION ia get real time sample fills the structure pointed to by inpointer with the next unprocessed raw sample from the CO pressure and flow signal The function also updates the real time graphs on the user interface screen and writes the new data into datalogging files if necessary CALLING CONVENTION status ja get real time sample amp sample RETURN VALUE 0 when no errors occur during data acquisition 1 when data is invalid or when one or more monitors are not connected and 1 when a fatal error occurs during data acquisition failure of the AD board or data overrun An Intelligent Alarm System in Ancsthesia 69 FUNCTION int ia get breath by breath data struct breath by breath data in inpointer DESCRIPT
32. change in applied PEEP will influence some of the feature baselines like those for minimum and maximum pressure Other parameters the clinician can regulate on the anesthesia machine are related to gas composition However a change in gas composition will only affect patient related variables and not those associated with the breathing circuit or the ventilator Thus in order to know when to recalculate the feature baselines changes in RR EE FGF and PEEP need to be detected Since a rule evaluation is performed once during each breath period see chapter 3 it is sufficient to collect the updated setting values after each breath detection 5 1 2 Which Feature Baselines Change In this section we determine which feature baselines actually should be recalculated after setting changes As a result of some assumptions we make not all of the normal values for the signal features mentioned in 3 2 2 need to be recalculated In a first approximation we assume that the carbon dioxide CO up slope down slope and end tidal plateau value are not influenced by a change in V4 FGF or PEEP The up slope and the down slope will change slightly when the gas flow in and out of the lungs is altered CO transport is faster with higher flows but tests showed that these changes usually lay within the tolerance zone of our slope calculation The slopes should always be very steep in no malfunction An Intelligent Alarm System in Anesthesi
33. cribes how different feature values change when this malfunction occurs In order to make every rule unique some features that remain unchanged UC can also be added to the rule Van Oostrom 1 gives the complete set of rules in the prototype I system This preliminary rule set was the result of extensive research and discussions with anesthesiologists the experts In order to make clear how the rule evaluation concept works a simplified version of some of the SIMPLEXYS rules is given in table 3 1 Every SIMPLEXYS run starts with looking at one or more rules of the STATE type Each rule of this type has one or more GOALS The rules to be evaluated next are the GOALS belonging to the STATE rule s with value TR True So in table 3 1 the rule that checks whether there is no malfunction in the breathing circle BREATHING SYSTEM OK is always evaluated first This rule is composed of different subrules like the rules that test for incompetent An Intelligent Alarm System in Ancsthesia 27 Table 3 1 Example of some SIMPLEXYS rules RUNNING The breathing circuit expert is up and running STATE INITIALLY TR THEN GOAL BREATHING SYSTEM OK BREATHING SYSTEM OK No malfunctions in the breathing circle NOT INC VALVE OR OBSTRUCTION OR LEAK OR DISCONNECT OR EXH CO2 ABSORBER OBSTRUCTION An obstruct on s detected OBST ET TUBE OR OBST INSP HOSE OR OBST EXP HOSE OR OBST VENT HOSE OBST ET TUBE An obstruction
34. d The statuses of the 3 signals are also displayed in a traffic light fashion When the signal is OK the message is given in green an invalid signal is represented by a yellow message and a red notice means that there is no signal detected at all signal FLAT The ventilator status word indicates whether mechanical ventilation is present or not When the system status is ALARM an alphanumerical alarm message indicates the most probable malfunction s together with its their most likely site An Intelligent Alarm System in Anesthesia 49 Other information including the system setting values RR EE FGF and PEEP and important measured quantities like the inspired O percentage F O and the expired minute volume MV is presented in numerical fashion Finally program control messages about possible user inputs are given on the lower line of the screen Currently the user can control the system by pushing buttons on the keyboard He can push the RESET BASELINES button as mentioned earlier which will overrule any automatic baseline reset and accept the current running average value of all features as new baselines Furthermore the user can start and stop recording the raw signal samples on hard disk suspend the program temporarily when he wants to take a detailed look at waveforms or alarm messages and abort the program to go back to DOS 5 3 Software Upgrades Since one of the final goals of the Intelligent Alarm P
35. d during anesthesia are equipped with their own alarm limits Currently the available alarm technology relies on the threshold check For each variable measured by a particular monitor the clinician has to define what he considers the normal band before the start of anesthesia This means that an alarm message will be generated when a variable exceeds either its upper or its lower limit Usually the alarm is a tone of a certain frequency sometimes combined with a text message on the monitor s display When for example one only considers a blood pressure monitor this strategy will certainly be appropriate An alarm message is generated whenever e g the systolic blood pressure exceeds its set upper limit In this case the audio alarm will be specific and easy to interpret for the anesthesiologist However the monitoring situation in today s operating room OR is more complicated In paragraph 1 2 the minimal array of monitors used during anesthesia was briefly described Each of these monitors has its own alarms independent of all the others Considering the fact that an alarm is usually accompanied by all sorts of other alarms some of the problems of the current strategy already become clear An Intelligent Alarm System in Ancsthesia 9 In an emergency situation when many measured variables can exceed their set limits the clinician s first reaction will be to silence the abundance of audio alarms In these situations it is often ver
36. e breathing circuit during mechanical ventilation This is performed by measuring and combining a carefully chosen set of signals in or close to the circle breathing circuit An Intelligent Alarm System in Anesthesia 14 The goal for the second version is to incorporate ventilator related malfunctions in the alarm system The target parts of prototypes I and II are illustrated in figure 2 1 The design and testing of these first two prototypes are described in the next chapters Once proven these designs will be the platform to incorporate other patient variables and signals to come to a truly integrated alarm scheme An Intelligent Alarm System in Anesthesia 15 CHAPTER 3 IASA THE FIRST PROTOTYPE The goal for the first prototype is the automated detection of the malfunctions that can occur in the circle breathing circuit during mechanical ventilation in the operating room OR In this chapter a detailed description of the software and hardware that composes this prototype alarm system is presented 3 1 The Data Flow through the System Before we look in more detail into the configuration of the system a schematic of the data acquisition signal processing feature extraction and rule evaluation performed before an alarm message appears on the screen is pictured in figure 3 1 This design was proposed and first implemented by Van Oostrom 1 although the implementation of the signal processing part was already performed earlie
37. e system is the most common breathing circuit used in the United States today and is the main focus of our research Other breathing circuits include the Bain and Mapleson systems 3 in which no rebreathing of anesthetic gases takes place These circuits are not considered in our research TO VENTILATOR FRESH GAS IN SCAVENGING INSPIRATORY VALVE EXPIRATORY INSPIRATORY HOSE PATIENT Figure 1 2 Detailed schematic of the circle breathing circuit The arrows indicate gas flows An Intelligent Alarm System in Ancsthesia 5 A detailed schematic of the circle breathing circuit is given in figure 1 2 Important components include the inspiratory and expiratory hoses two unidirec tional valves the CO absorber and the endotracheal E T tube All these parts together compose a system through which the patient is ventilated The unidirectional valves one is placed in the inspiratory one in the expiratory hose accomplish that gas inhalation and gas expiration take place through different hoses In this way the expired O poor CO rich gases cannot be reinspired at the next inhalation without having passed through the CO absorber and along the fresh gas inlet The E T tube is placed into the patient s airway by the anesthesiologist at the beginning of general anesthesia The tube is connected with a Y piece to the inspiratory and expiratory hoses At the fresh gas inlet new O rich gas from the anesthesia machine is mixed with t
38. ected to an anesthesia machine via a breathing circuit and to all kinds of monitoring equipment In the next paragraphs a short functional description of the anesthesia system and the most common monitoring equipment is given 1 1 The Anesthesia System An important part of the anesthesia system is the anesthesia machine It helps the anesthesiologist by preparing a gas mixture with precisely known but variable composition which is administered to the patient In the United States the gas combination most frequently used consists of oxygen O3 nitrous oxide N O and an anesthetic agent halothane isoflurane or enflurane The anesthesiologist can control the relative volume of each of these composites He delivers the gas mixture to the patient by manually squeezing a breathing bag or by using a mechanical ventilator usually mounted on the anesthesia machine The anesthetic system is generally divided into 5 major parts 1 1 High pressure system Low pressure system 2 3 Breathing circuit 4 Ventilator system 5 Scavenging system An Intelligent Alarm System in Anesthesia 3 A schematic of the entire anesthesia system is given in figure 1 1 OXYGEN IN NITROUS OXIDE IN HIGH PRESSURE SYSTEM FLOWMETER 1 FLOWMETER 2 i LOW PRESSURE VAPORIZER SYSTEM x a ER d VENTILATOR nora iz Bag SCAVEN GING SYSTEM BREATHING CIRCUIT PATIENT Figure 1 1 Schematic of the anesthesia system The ar
39. en z i gt z i 1 the calculated time constant may be slightly lower than the actual time constant because some samples with a high value are skipped If for example when T lt 0 14 z i lt 0 7 x z i 1 the new sample will be rounded to exactly 0 7 x z i 1 and the algorithm goes on to the next sample So when T 0 14 the result of the time constant estimation will be exactly 0 14 However as was mentioned earlier tests showed that the time constant was never lower than 0 3 in clinical situations With this algorithm adequate results are obtained very fast since the time consuming logarithm calculations after each incoming sample are avoided An Intelligent Alarm System in Anesthesia 24 3 2 3 Signal Validation Slope values and time constants are only considered valid if they are based on a minimum number of samples in the respective curve parts At this moment a minimum of six samples is required for both features to be valid Furthermore the breath time is stored for every signal after breath detection The signal processing waits a maximum of 120 of this previous breath time for the next breath detection If no new breath is detected within this time span the signal is declared invalid and a time out flag is set When a signal is invalid all of its features are considered invalid for that particular breath period By setting a time out flag the system is prevented from waiting forever until a breath is detected When
40. ero flow period whereas a new breath period for the CO signal starts at the beginning of the up slope see figure 3 2 This means that breath detection is performed at the start of patient expiration for every signal Thus as soon as a new breath has been detected on every signal a complete set of numerical features representing one breath period of all signals is available These features are used in the fault detection scheme For the CO signal the features include the inspired CO level the end tidal CO level Percoz the values of the linear expiratory up slope and inspiratory down slope and the plateau time The flow features are next to timing variables maximum flow Fax minimum flow Fin the time constant of the expiratory down stroke Tp and tidal volume VT The latter variable is calculated by integrating the positive part of the flow curve during one breath period Van Oostrom 1 describes the algorithms used in the initial implementation for calculating maximum and minimum values slopes and time constants However the time constant and slope algorithms have been updated to speed up the signal processing The derivative of the pressure waveform is updated and filtered after each incoming sample The value of the filtered derivative at the moment the An Intelligent Alarm System in Ancsthesia 21 algorithm detects a transition from state 1 to state 2 see figure 3 3 equals P jope The results of this
41. es based on an average adult patient is often inadequate The normal values depend heavily on the ventilator settings used demographic patient data age weight sex and type of operation Tidal volume for example can vary from 100 ml for neonatals to over 1 liter for young athletes For every feature a particular low and high threshold is defined with the baseline of that feature as a reference in a look up table The zone between upper and lower threshold is called the normal band for that feature For example assume a default baseline value for the maximum flow of 600 ml sec When the low and high thresholds for the maximum flow feature are set to 30 and 2096 respectively the normal zone will be the band between 420 and 720 ml sec Also for each feature a minimum value is defined When the baseline is below this minimum the thresholds are absolute rather than relative to the baseline to prevent the normal band from becoming too small This is particularly necessary for features that have a natural baseline close to zero like minimum flow or inspired CO At the end of each breath period each valid feature gets assigned a symbolic value These values range from UC UnChanged when the feature value lays within the normal band UP Up when the feature value lays above the upper threshold to DN Down when the feature value is smaller than the lower threshold The set of symbolic values UP DN UC NV serves as inpu
42. es of IASA prototype II are presented in this chapter 5 1 The Automatic Baseline Reset First of all the requirements for the automatic baseline reset routine need to be defined This is done in the next sections Subsequently the implementation of the automatic reset in the intelligent alarm software is discussed 5 1 1 When to Reset the Feature Baselines The feature baselines have to be recalculated after a change in one or more of the settings On the ventilator a change in the set tidal volume V4 the set respiratory rate RR or the set ratio of inspiration to expiration time EE will have effects on the normal values of certain signal features A change in the fresh gas flow FGF setting on the anesthesia machine also results in baseline changes since it affects the delivered tidal volume and therefore the flow and pressure related features An Intelligent Alarm System in Anesthesia 41 The anesthesiologist may for therapeutic reasons also apply a certain amount of positive end expiratory pressure PEEP to the breathing circuit This has the effect that the pressure does not go to zero after expiration but remains at the higher PEEP level much like an offset voltage in an electrical circuit As a result the patient s lungs remain partially inflated after expiration which will improve gas exchange In most modern anesthesia systems the PEEP value can be regulated by a control knob mounted on the expiratory valve A
43. etect and locate the most important malfunctions that can occur in the circle breathing circuit during general anesthesia These malfunctions include small and large leaks at various sites obstructions of different hoses an exhausted CO absorber incompetent unidirectional valves and disconnections at various locations Detection and identification of malfunctions introduced with the Gainesville Anesthesia Simulator occurred correctly within 30 seconds in 88 of the cases Furthermore the system was taken to the operating room OR and tested during 11 cases No false alarms were recorded while no critical event went by undetected Also the signal processing routines were robust enough to handle the noisy and sometimes disturbed signals in the OR The second prototype designed to adapt automatically to changes in the anesthesia system control settings had a 95 5 correct detection performance during tests with the simulator All faults that were not detected were not immediately clinically dangerous however No errors due to wrong adaptation to a setting change were recorded Only the detection of an incompetent inspiratory valve turned out to be unreliable sometimes a wrong alarm message was generated Also the system s detection performance was degraded slightly with high values of positive end expiratory pressure PEEP in the breathing circuit This is due to the fact that especially the expiratory flow signal is affected drastically by
44. h issue we will come back in later chapters Apart from the Fukui system very few implemented integrated alarm systems are described in the literature Most of them are outlined in the next paragraph 2 2 1 Overview of Implemented Integrated Systems One of the best structured alarm systems that is already commercially available is implemented in the Narkomed II anesthesia machine manufactured by North American Dr ger It is described by Schreiber et al 16 The system uses several sensors distributed over anesthesia machine breathing circuit and patient together with a centralized display In this way the time the clinician needs to identify and correct a problem is minimized A priority scheme is used to divide the alarm messages into warning caution and advisory messages Although the alarms are still threshold based and no multivariable analysis is performed this is a first step toward smarter alarm systems Another implementation although only on a prototype basis is the Data Acquisition and Display System DADS developed at the Eindhoven University of Technology and described by Meijler 17 This system incorporates a centralized display automatic record keeping capabilities and a threshold and trend detection scheme The problem of many superfluous alarms is not solved however basically because no multivariable analysis is performed A ventilator alarm system for use at the NASA Space Station developed at the University of
45. he CO down slope feature did not increase sufficiently after introduction of the malfunction so the correct alarm message was not triggered Furthermore when the inspiratory valve was stuck and a high PEEP value was applied setting 2d the expiratory flow returned completely not An Intelligent Alarm System in Anesthesia 55 partly as in situations with low or no PEEP through the inspiratory hose In this case the stuck valve provided a short circuit from the lungs to the part of the breathing circle located on the machine side of the valves where no PEEP was present As a result the measured flow signal was flat no breath detection could be performed on this signal and no intelligent alarm was generated since the pressure signal remained unchanged Currently a discussion is going on about possible improvements in the detection of the inspiratory valve malfunction The current rule uses the assumption that the CO down slope becomes less steep when the valve is stuck The tests showed that this is not always the case during high RR or FGF values the detection was unreliable A possible solution is to make the thresholds that define the normal band for each feature adaptive When the quality of the signal is very high the normal band can become smaller so that very small changes are already detected This way when a smooth CO signal is available an increase of for example 5 instead of 20 in the CO down slope value could al
46. he Netherlands 3 5 September 1989 Ed by Han Vinck EUT Report 89 E 224 1989 ISBN 90 6144 224 9 jmakers POSSTBILITY TO INCORPORATE SATURATION IN THE SIMPLE CLOBAL MODEL OF A SYNCHRONDUS MACHINE WITH RECTIFIER EUT Report 89 E 225 1989 SBN 90 6144 225 7 Dahiya R P and E M van Veldhuizen W R Rutgers L H Th Rietjens MENTS ON iNITIAL BEHAVIOUR OF CORONA TENERATEO WITH ELECTRICAL PULSES SUPERIMPOSED ON DC BIAS EUT Report 89 E 226 1989 ISBN 90 6144 226 5 Bastings R H A TOWARD THE DEVELOPMENT OF AN INTELLIGENT ALARM SYSTEM IN ANESTHESIA EUT Report 89 E 227 1989 ISBN 90 6144 227 3 Hekker J J COMPUTER ANIMATED GRAPHICS AS A TEACHING TOOL FOR THE ANESTHESIA MACHINE SIMULATOR EUT Report 89 E 228 1989 ISBN 90 6144 228 1 Oostrom J H M van INTELLIGENT ALARMS IN ANESTHESIA An implementation EUT Report 89 E 229 1989 ISBN 90 6144 229 X Winter M R M DESIGN OF A UNIVERSAL PROTOCOL SUBSYSTEM ARCHITECTURE Specification of functions and services EUT Report 89 E 230 1989 ISBN 90 6144 230 3 Schemmann M F C and H C Heyker J J M Kwaspen Th C van de Roer MOUNT ING AND DC TO 18 GHz CHARACTERISATION OF DOUBLE BARRIER RESONANT TUNNELING DEVICES EUT Report 89 E 231 1989 ISBN 90 6144 231 1 Sarma A D and M H A J Herben BATA ACQUISITION AND SIGNAL PROCESSING ANALYSIS OF SCINTILLATION EVENTS FOR THE OLYMPUS PROPACATION EXPERIMENT EUT Report 89 E 232 1989 ISBN 90 6144 232 X
47. he gas coming from the CO absorber This mixture is delivered to the patient 1 2 3 4 The mechanical ventilator is the driving force behind the breathing system t is used to move gas into the patient s lungs Usually this is ac complished by periodically applying a positive pressure to a bellows connected to the airway forcing the gas mixture through the inspiratory hose and the E T tube into the patient s lungs When the lungs are filled to a suitable level the positive pressure is removed In this way the lungs can empty passively through the expiratory hose into the ventilator bellows This process is repeated continuously The two types of mechanical ventilators commonly used in the United States are the constant flow generator and the constant pressure generator 4 The constant flow generator delivers as its name states a constant but adjustable inspiratory gas flow to the patient whereas the constant pressure generator maintains a constant airway pressure during inspiration On the ventilator the anesthesiologist can make changes to adjust for type of operation demographic patient data or specific patient conditions 1 4 Adjustments can be made for tidal volume respiratory rate RR and inspiratory to expiratory ratio EE Vr is the gas volume delivered to the patient Ty P Ty T An Intelligent Alarm System Anesthesia 6 during one breath RR the number of respirations in one minute and I E the ratio
48. hesia machine These include a stuck inspiratory or expiratory valve an exhausted CO absorber and leaks at different sites By manipulating circulation related variables and signals like the oxygen O saturation of the blood the electro cardiogram ECG and the blood pressure also patient related mishaps like hypoxia the O saturation of the blood is too low and deep or light anesthesia can be simulated Since blood circulation is not really present in the simulator these signals are generated by software rather than physically measured An Intelligent Alarm System in Anesthesia 31 Ohmeda 5410 volume monitor Fresh gas flow from Modulus Inspiratory i Ohmeda 7810 Mechanical ventilator lung Expiratory valve To scavenging CO cylinder system Figure 4 1 Schematic of the test setup with the Gainesville Anesthesia Simulator The dashed lines represent electrical signals The original purpose of the simulator was to teach anesthesiologists how to react on rare catastrophic events during anesthesia With the simulator the clinician can practice his reactions to mishaps and even make mistakes without the stress of exposing a real patient to a life threatening condition Since the malfunctions we would want to detect never can be introduced during real anesthesia the simulator is the ideal testbench for the Intelligent Alarm System The simulator setup and the locations of the sensors used by our system are
49. ic Baseline Reset 41 5 1 1 When to Reset the Feature Baselines 41 5 1 2 Which Feature Baselines Change 42 5 1 3 Simple Breathing Circuit Modeling 44 5 1 4 Implementation 46 5 2 The User Interface a2 2 4 io ocu doe dob X WE ETUR Ru 48 5 3 Software Upgrades ia re ao dX e Bar 50 CHAPTER 6 TESTING PROTOTYPE 1 53 6 1 Single Malfunctions 53 6 2 Multiple Malfunctions 57 63 Future Desunp SERERE 58 CHAPTER 7 CONCLUSIONS AND RECOMMENDATIONS 59 1 1 CODCIUSIDIS rauco EX Oe Wo ERE ee 59 7 2 Recommendations for Future Research 60 An Intelligent Alarm System in Anesthesia v APPENDIX 1 THE KNOWLEDGE BASE 65 APPENDIX 2 DATA FLOW IN PROTOTYPE IT 66 APPENDIX 3 FORMAL DESCRIPTION INTERFACE ROUTINES 67 APPENDIX 4 TECHNICAL DATA PROTOTYPE 72 An Intelligent Alarm System in Ancsthesia vi BOVI CO ECG E T tube FGF IASA LE Uo RR Table of Abbreviations analog to digital electrocautery carbon dioxide electro cardiogram endotracheal tube fresh gas flow intelligent alarm system in anesthesia ratio of inspiration to expiration time input output nitrous oxide oxygen operating room positive end expiratory pressure res
50. integration the need in future anesthesia delivery systems Med Instrum Vol 17 1983 p 389 392 Arnell W J and D G Schulz Computers in anesthesiology a look ahead Ibid p 393 395 Quinn M L Semipractical alarms a parable J Clin Monit Vol 5 1989 p 196 200 Bastings R H A Toward the development of an intelligent alarm system in anesthesia Faculty of Electrical Engineering Eindhoven University of Technology 1989 TUE Report 89 E 227 Blom J A SIMPLEXYS a real time expert systems tool In Proc IASTED Int Symp on Expert Systems Geneva 16 18 June 1987 Ed by M H Hamza Anaheim Cal Acta Press 1987 P 21 25 Blom J A The SIMPLEXYS experiment real time expert systems in patient monitoring Ph D thesis Eindhoven University of Technology 1990 Good M L and S Lampotang G L Gibby J S Gravenstein Critical events simulation for training in anesthesia J Clin Monit Vol 4 1988 p 140 Oostrom J H M van Flow and pressure modeling in the breathing circle Internal paper Department of Anesthesiology College of Medicine University of Florida Gainesville Florida 1989 An Intelligent Alarm System in Ancsthesia 64 APPENDIX 1 THE KNOWLEDGE BASE In the following the rules at the highest two levels of the knowledge base are given LEVEL 0 EXIT Exit expert system go back to DOS BTEST control A RUNNING Breathing circuit expert up and running STAT
51. is first prototype is described in chapter 3 The results of extensive testing procedures are given in chapter 4 The second goal of this thesis research project is to design and implement a second prototype Intelligent Alarm System able to adapt automatically whenever the clinician changes control settings on the anesthesia machine or ventilator the An Intelligent Alarm System in Anesthesia 1 first prototype is not able to do this In chapter 5 the concepts and algorithms used in this second prototype are described while in chapter 6 the first test results of this system are presented Funding for the research was provided by Ohmeda Medical Products Madison WI U S A a company manufacturing anesthesia equipment An Intelligent Alarm System in Anesthesia 2 CHAPTER 1 A SHORT INTRODUCTION TO ANESTHESIA Before addressing the issues of monitoring and alarms during anesthetic procedures a brief introduction to anesthesiology is given in this chapter During surgery the patient is in a state of narcosis unconsciousness muscle relaxation and analgesia insensitivity to pain usually referred to as general anesthesia 2 This condition is induced by an anesthesiologist who administers a combination of intravenous and inhalation drugs to the patient The task of the anesthesiologist is to maintain the patient s vital organ function while adequate levels of anesthesia are sustained In order to facilitate this task the patient is conn
52. jonge kinderen worden gebruikt geen problemen op Met enkele verbeteringen en uitbreidingen kan het systeem als basis dienen voor de ontwikkeling van een compleet geintegreerd intelligent alarm systeem voor de combinatie patient anesthesiemachine An Intelligent Alarm System in Anesthesia ii ACKNOWLEDGEMENTS First of all I would like to thank Ir Jan van der Aa M E E for all the help advice and useful criticism he provided throughout the entire research period and during the writing of this report Furthermore I would like to thank Ir Hans Blom and Ir Hans van Oostrom for their valuable suggestions for improvement in this thesis Professor Dr Ir J E W Beneken and Professor J S Gravenstein M D Dr h c for making it possible to do my graduate research in the inspiring environment of the Department of Anesthesiology at the University of Florida the cooperating anesthesiologists of the department without whose medical knowledge a project like this would be sure to fail my colleagues for making my stay in Gainesville such a great time and finally Leontien for having patience for nearly 11 months An Intelligent Alarm System in Ancsthesia iii CONTENTS SUMMARY x do A CP abe a a LER Ran i SAMENVATTING u gates orn ERROR er Pe mie Ge Us cm o riot ced ii ACKNOWLEDGEMENTS 4r ea cr iii Table of Abbreviations vus y 23 sam aa RU aka os vii Table of Medical Terms 2 as Se uae x Ew A EE viii INTRODUCTION
53. le 6 2 Again no manual baseline reset was performed after transitions from setting 2a to 2b and 2b to 2c As explained in chapter 5 the feature baselines had to be reset manually after the PEEP change during the transition from setting combination 2c to 2d The test results are presented in table 6 3 An Intelligent Alarm System in Anesthesia 54 Table 6 2 Settings during the second test sequence for Prototype II Setting ml RR br min LE l min PEEP cmH O 2a 300 15 1 2 3 0 2b 500 12 1 2 5 0 2c 750 10 1 2 8 0 24 750 10 1 2 8 7 Table 6 3 Results of test sequence 2 for Prototype II at the simulator Malfunction Setting 2a Setting 2b Setting 2c Setting 2d Large leak insp hose X X X X Large leak exp hose X X X X Disc CO sampling line X X X X Small leak insp hose X X X X Small leak Y piece X X X X Small leak exp hose X X X X Disc FGF hose X X X X Stuck insp valve X X F F Obstr insp hose X X X X Obstr E T tube X X X X Obstr exp hose X X X X Obstr vent hose X X X X The automatic baseline reset again worked without errors the calculated normal values always laid within 1096 of the actual feature values after baseline changes and no false alarm messages due to wrong baseline recalculation were recorded The two false alarm messages were a Small leak message at setting 2c and an Apnea alarm at setting 2d both when the inspiratory valve was stuck At a high FGF value setting 2c t
54. lized integrated alarm system for the patient equipment combination during anesthesia An Intelligent Alarm System in Anesthesia 62 1 2 3 t4 5 6 7 8 9 10 11 12 13 14 REFERENCES Oostrom J H M van Intelligent alarms in anesthesia an implementation Faculty of Electrical Engineering Eindhoven University of Technology 1989 EUT Report 89 E 229 J J L C M van der Intelligent alarms in anesthesia a real time expert system application Ph D thesis Eindhoven University of Technology 1990 Dorsch J A and S E Dorsch Understandig anesthesia equipment construction care and complications Baltimore Md Williams amp Wilkins 1975 Dupuis Y G Ventilators theory and clinical application Saint Louis Missouri Mosby 1986 Standards for basic intra operative monitoring Park Ridge 111 American Society of Anesthesiologists 1986 Gravenstein J S and D A Paulus T J Hayes Capnography in clinical practice Boston Mass Butterworths 1989 Gravenstein J S and M B Weinger Why investigate vigilance Editorial J Clin Monit Vol 3 1986 p 145 147 The automated anesthesia record and alarm systems Ed by J S Gravenstein and R S Newbower A K Ream N T Smith Boston Mass Butterworths 1987 Rampil I J Intelligent detection of artifact Ibid p 175 190 Philip J H Thoughtful alarms Ibid p 191 201
55. n Intelligent Alarm System in Anesthesia i SAMENVATTING In de moderne operatiekamer OK wordt zowel de toestand van de patient als het functioneren van de anesthesiemachine continu in de gaten gehouden met behulp van een steeds groeiend aantal meetapparaten Elke monitor is uitgerust met zijn eigen niet specifieke alarms Daardoor is er een duidelijke behoefte ontstaan aan een geintegreerd alarm systeem dat door het combineren van data van verschillende monitoren in staat is bruikbare en intelligente alarmbood schappen te genereren wanneer er iets misgaat Een eerste prototype van zo n systeem dat data van drie verschillende meetapparaten gebruikt om mechanische defecten in het beademingscircuit op te sporen is getest met behulp van een anesthesiesimulator en in de OK Tijdens de testen met de simulator was het systeem in staat om 8846 van de geintrodu ceerde defecten correct te identificeren binnen 30 seconden In de OK werden geen valse alarms geregistreerd ondanks het feit dat de signalen vaak gestoord werden Een tweede prototype is ontworpen dat in staat is zich automatisch aan te passen wanneer instelwaarden op de anesthesiemachine en of de ventilator tijdens de operatie worden veranderd Verder is een gebruikersvriendelijk gemakkelijk te interpreteren display ontworpen voor dit systeem Tijdens testen met de simulator kon dit systeem 95 5 van de defecten correct opsporen Tevens leverden de hoge beademingsfrequenties die vaak bij
56. nesthetized dogs as well as in the hostile environment of the OR as mentioned in chapter 6 The clinical usefulness of the system should be evaluated after measuring the number of suppressed unnecessary alarms Furthermore more tests with multiple malfunctions occurring at the same time must be performed In order to avoid that the system accepts malfunctions present during system start up as normal Prototype II could be used as a platform to develop an automated pre check procedure for the anesthesia machine Currently the clinician has to test the whole system manually before the start of anesthesia As a next step the system can be expanded to generate simple but useful alarm messages ie evaluate a subset of the current rule set after each breath period during manual or spontaneous ventilation The current system is only designed for mechanical ventilation A trend detection scheme for variables like the blood oxygen saturation and the end tidal CO value needs to be designed and implemented An Intelligent Alarm System in Anesthesia 61 Analyses of the available medical knowledge in preparation for representa tion of the oxygen saturation and the end tidal CO in the knowledge base showed the need for such addition Further on in the future more circulation related signals like blood pressure and the electro cardiogram may be incorporated into the system in order to come closer to the final goal of a centra
57. ng in the patient rather than the equipment The following points summarize possible improvements or additions to the Intelligent Alarm System and suggest extra tests to be performed before a real clinical version can be implemented As mentioned in chapter 6 and 7 1 a more sophisticated PEEP model needs to be incorporated in the system Subsequently the automatic feature baseline reset routines can be fine tuned Also the detection rule for the stuck inspiratory valve needs to be revised A Pulse Oximeter has to be interfaced with our alarm system After that the oxygen saturation of the blood and the inspired oxygen percentage can be incorporated into the knowledge base as extra variables This way the An Intelligent Alarm System in Anesthesia 60 third IASA prototype will be able to evaluate gas exchange in the lungs and to give some patient related alarms in addition to the list of detectable mechanical malfunctions Although the current system can automatically recalculate feature baselines after changes in settings see chapter 5 the initial baselines still have to be reset manually by the anesthesiologist when the situation is stable Some simple patient modeling combined with the implementation of a possibility to enter basic patient data age height weight sex into the system may lead to automatic calculation of a set of initial feature baselines Prototypes II and III must be tested extensively on a
58. ns that can occur in the circle system and thus need to be detected by the alarm program This list includes leaks at different sites obstruction of the various hoses disconnections unidirec tional valves that are stuck in the open position due to moisture in the gas mixture and exhaustion of the CO absorber the absorber can no longer remove CO from the exhaled gas mixture Because the system needs to be clinically usable it was decided to use only signals that are routinely measured in the breathing circuit during anesthesia Furthermore the number of sensors should be kept to a minimum in order not An Intelligent Alarm System in Anesthesia 17 to make the system more complicated than necessary The set of signals chosen for prototype I consists as mentioned in 3 1 of the partial CO pressure measured at the Y piece of the breathing circuit the airway pressure measured downstream at the patient side of the inspiratory unidirectional valve and the airway flow measured upstream at the patient side of the expiratory unidirectio nal valve see also figure 1 2 This is the most common way of placing the sensors in today s clinical practice 3 2 1 From Monitors to Sampled Signal Waveforms Standard monitors manufactured by Ohmeda Madison Wl are used to obtain the waveforms The Ohmeda 5200 CO Monitor and the Ohmeda 5500 Airway Pressure Monitor provide analog outputs for the respective signals The analog signals are fed
59. o the contribution of FGF to the delivered tidal volume is equal to FGF x In order to build up and maintain the PEEP a certain part of the tidal volume is lost Experimental measurements showed a tidal volume loss nearly linearly related to the PEEP setting For a PEEP lower than 10 the loss equaled approximately 25 ml cnH5O x PEEP This results in the following approximation VrTael X Tinsp x 25 0 X PEEP 5 6 An Intelligent Alarm System in Ancsthesia 45 With 5 1 to 5 6 we are able to give all the expected feature values when no malfunction is present as a function of the setting values RR EE FGF and PEEP and the constants R C RC and K see also 28 Ty sec 60 0 RR breaths min 5 7 Tinsp sec EE x T 1 0 EE 5 8 Tap sec Th Tinsp 5 9 Vrmea ml K x FGF x Tis 25 0 x PEEP 5 10 Pin cmH 0 PEEP 5 1 F max m sec Vasa R 5 2 Pmax 00H50 R x Fma Pin 5 3 Pslope cmH5O sec VTmea K x Tinsp x C 5 11 This set of formulas is used to calculate the new feature baselines when one or more settings have changed The implementation is discussed in the next section 5 1 4 Implementation The Ohmeda 7810 ventilator provides a mode in which all the measured values and all the settings dialed on the front panel of the ventilator are transmitted via its serial port after the user i e the intelligent alarm
60. o be done in real time by combining and evaluating several signals derived variables and other patient data The Intelligent Alarms System in Anesthesia IASA project started a few years ago at the University of Florida in Gainesville and the Eindhoven University of Technology with the intention to develop a working prototype of such a system In this paragraph the general method is explained briefly before an extensive description of the first prototype configuration is presented in the next chapter The process the anesthesiologist goes through when he tries to locate and identify a problem during anesthesia can be divided into several phases This An Intelligent Alarm System in Anesthesia 13 subsequent checking of different parts of the patient equipment combination is schematically pictured in figure 2 1 LEVEL 1 LEVEL 2 LEVEL 3 Figure 21 Schematic of the decision process of the anesthesiologist SITUATION STABLE PATIENT EQUIPMENT OK OK VENTI LATOR OK PROTO i TYPE I PROTOTYPE II The final goal is to design a system that can give intelligent messages about abnormalities occurring in the patient as well as in the equipment Because of the enormous complexity of especially the patient system many factors influence the situation of the patient during anesthesia we choose to start cautiously The first prototype incorporates an alarm scheme that detects mechanical malfunctions in th
61. of inspiration to expiration time Depending on the type of ventilator used the fourth important setting is either the inspiratory flow Fj constant flow generator or the inspiratory pressure Pj constant pressure generator The anesthesiologist can also switch to a breathing bag with which he can manually ventilate the patient This is often used during emergency or during critical phases in anesthesia like intubation entering the E T tube at the beginning and extubation taking out the E T tube thereby allowing the patient to breathe by himself after surgery 5 The scavenging system removes excess gases from the breathing circuit At the end of expiration a valve to the scavenging system opens thereby allowing gas to leave the circuit The scavenging system removes the excess gas from the operating room OR to prevent pollution of the clean OR air with anesthetic gases 1 2 The Most Important Monitoring Equipment During surgery a variety of monitors is connected to the patient while other measuring devices are inserted into the breathing system This gives the anesthesiologist the ability to discover changes in the state of the patient or malfunctioning equipment as early as possible so that corrective action can be taken in time A standard for minimal monitoring defined by the American Society of Anesthesiologists ASA in 1986 5 requires that at least the patient s oxygenation ventilation circulation and temperature are monit
62. ored continuously In most cases the oxygen level of the inspired gas is measured with an O analyzer placed in the breathing circuit near the fresh gas inlet Oxygenation of blood is usually measured with a pulse oximeter This monitor uses differences in light absorption characteristics between hemoglobin and oxyhemoglobin to calculate the oxygen saturation of the blood An Intelligent Alarm System in Ancsthesia 7 The ventilation of the patient is typically measured with a capnograph This device monitors the partial pressure of CO in the gas mixture The signal contains a variety of information about the patient and equipment statuses 6 Most of the times the CO content of the gas is measured at the Y piece of the breathing circuit Other common monitors that help the anesthesiologist evaluate the patient s ventilation are an airway pressure monitor usually placed near the inspiratory one way valve and a tidal volume or flow monitor in the expiratory limb of the breathing circuit 2 In order to provide the anesthesiologist with some information about the patient s blood circulation two more variables are measured The first signal the electro cardiogram ECG gives information about the electrica activity of the heart The ECG is measured with three or more electrodes attached to the patient s chest The second variable is the blood pressure Blood pressure is measured continuously usually noninvasively with an inflatable cuff
63. otype II has currently only been tested on the simulator The results of the simulator testing procedures are summarized in the next sections 6 1 Single Malfunctions The test setup with the Gainesville Anesthesia Simulator was the same as described in 4 1 and pictured in figure 4 1 The second flow device was inserted into the fresh gas hose and the data from this Ohmeda 5410 volume monitor and the Ohmeda 7810 ventilator were fed to the computer as described in chapter 5 To test the fault detection performance of the system two test sequences were executed During each sequence a number of malfunctions was introduced at different combinations of tidal volume V4 respiratory rate RR ratio of inspiration to expiration time EE fresh gas flow FGF and positive end expiratory pressure PEEP settings The test protocol was the same as described in 4 1 1 except for the fact that no manual baseline reset was performed after setting changes other than PEEP changes The only other time the baselines were reset to their current mean was after system start up The compliance of the mechanical lung was set to 0 1 cmH O At the beginning of the first sequence the following setting values were present 750 ml RR 10 breaths min EE 1 2 FGF 5 l min and PEEP 0 cmH 0 setting la After introducing a number of malfunctions the FGF setting was lowered to 2 l min and the experiments were repeated setting 1b Finally
64. piratory rate tidal volume An Intelligent Alarm System in Ancsthesia vil abdomen analgesia anesthesia machine artificial nose breathing circuit electrocautery endotracheal tube enflurane halothane hypoxia isoflurane narcosis scavenging system ventilator Table of Medical Terms belly insensitivity to pain machine preparing the gas mixture delivered to the patient during general anesthesia device inserted into the breathing circuit to moisten the inhaled gases network of hoses and valves connecting the patient to the anesthesia machine the process of cutting the patient s skin by means of a high frequency high power electrical signal tube inserted into the patient s airway during general anesthesia connects the patient to the breathing circuit anesthetic agent added to the inhaled gas mixture see enflurane situation in which the oxygen saturation of the patient s blood is too low see enflurane unconsciousness system that removes excess anesthetic gases from the breathing circuit device that forces fresh gas into the patient s lungs via the breathing circuit An Intelligent Alarm System in Anesthesia viii INTRODUCTION During surgery the patient is under anesthesia unconscious and insensitive to pain The anesthesiologist maintains the patient s vital organ function during anesthesia The clinician gets important information about the patient s status from several monitors connected to bo
65. r by Bastings 24 The system first processes three real time signals measured in the breathing circle partial pressure of carbon dioxide CO at the Y piece airway pressure close to the inspiratory valve and airway flow through the expiratory valve The signals are transformed into symbolic feature values and subsequently fed into a rea time expert system at the end of each breath period later on in this chapter we will define breath period and feature Signals and signal features are chosen so that each fault that can occur in the breathing circuit will be reflected as a change in the set of symbolic data The expert system evaluates and combines the feature data derived from the different signals looks for changes in this data set reaches a conclusion about the status of the breathing circle OK or ALARM and gives an intelligent indication about mechanical malfunctions that have occurred together with their most probable site The signal processing feature extraction and expert system approaches are described in detail in the following paragraphs An Intelligent Alarm System in Anesthesia 16 BREATHING CIRCUIT SIGNAL PROCESSING FEATURE oa C SYMBOLIC SYMBOLIC SIGNAL SIGNAL FEM RULE abe EXPERT SYSTEM INTELLIGENT ALARM MESSAGES FE Tq So Figure 3 1 Data flow in prototype 1 of the Intelligent Alarm System 3 2 Signals and Signal Processing Van Oostrom 1 defines a list of malfunctio
66. ready trigger the Incompetent inspiratory valve message Problem with this method however is that we have to define and calculate a reliable signal quality index Furthermore an improved PEEP model has to be implemented to accommodate for the changes in the signals during high PEEP The current simple estimation formula 5 10 in which only the measured tidal volume is corrected for PEEP turned out to be insufficient Close attention has to be paid to the complexity of this model however Since the alarm system must be able to give correct messages in many different situations the implemented models must be kept as simple and robust as possible Apart from the issues mentioned above the overall detection performance was good Of 87 malfunctions introduced 84 were detected correctly within the 30 seconds time span This means that 95 5 of the faults were recognized by the system An Intelligent Alarm System in Anesthesia 56 6 2 Multiple Malfunctions Although theoretically the probability of multiple malfunctions occurring at the same time is extremely small we tested the system for a few combinations of malfunctions The goal of these tests was to make sure that the alarm scheme generates reasonable messages during multiple malfunctions and that no nonsense messages are presented The results of these experiments are given in table 6 4 the settings were the same as setting la see 6 1 Table 6 4 Test results for Protot
67. roject is to centralize alarms coming from different monitors the alarm messages generated by the ventilator and transmitted via its serial port are taken into account and displayed when no intelligent message is triggered by the rule set The ventilator alarms include messages about low gas supply pressure low and high airway pressure failing or tidal volume sensors low and internal electrical failures To prevent an enormous number of alarm messages in case of emergency a message shell is implemented Every alarm message is assigned a certain priority At the end of each expert system run only the triggered alarms with the highest priority are selected and put on the screen Every triggered alarm message is written in a file however This makes it possible to track back and evaluate the expert system behavior afterward Furthermore an extra FLAT status is implemented for the three signals This way the expert system will see the difference between no signal at all FL and an invalid signal for example due to artifacts NV The symbolic data that serve as input to the expert system can now have a value of either UP An Intelligent Alarm System in Anesthesia 50 MAIN MODULE LIBRARY i SIGNAL ANALYSIS ROUTINES Global declarstions Symbolic feature extraction routines Routines User that interface interface graphics routines ecquisition routines Figure 5 4 Configuration of the IASA prototype II sof
68. rows indicate gas flows EXCESS GAS OUT The 5 parts of the anesthesia system from figure 1 1 can be described as follows 1 The high pressure system regulates the gases coming from the hospital pipeline system O N O and usually air or in case of pipeline failure from cylinders mounted on the back of the anesthesia machine The highly pressurized gases are converted to low pressure by a network of pipes valves and regulators inside the anesthesia machine 3 before they enter the low pressure system 2 The low pressure system contains separate flowmeters that are calibrated for either N O or air With control knobs the anesthesiologist can adjust the gas flow for each of the components Upon exiting the flow meters the gas An Intelligent Alarm System in Anesthesia 4 components are mixed and driven through a vaporizer in which a controllable volume of anesthetic agent is added to the mixture After passing the vaporizer the gas mixture leaves the anesthesia machine at the fresh gas outlet and enters the breathing circuit 3 The breathing circuit shown in figure 1 1 is the circle breathing circuit in which carbon dioxide CO is removed from the exhaled gas mixture by a CO absorber Most of the times the CO absorber is placed in between the ventilator and the fresh gas inlet In this way exhaled gases can be reused so that as little gas and anesthetic agent as possible is lost to the scavenging system The circl
69. rvey Philip Fukui and Beneken propose three different approaches to sophisticated and useful alarms in patient monitoring An Intelligent Alarm System in Anesthesia 10 Philip 10 defines the patient s state as the set of variables that must be known to manage the patient In a first approximation he looks at only two possible patient states the correct state and the incorrect state For the detection of abnormalities a change detection scheme is used for all the monitored variables Whenever a change is detected the patient will enter the incorrect state At the same time the dimension of the state vector changes from 1 fault or no fault to 4 four different fault categories are distinguished and an alarm message is generated The algorithm determines whether the abnormality occurred in circulation anesthesia respiration or metabolism the 4 subsystems The state dimension increases with growing severity of the situation Circulation for example can be subdivided into medium blood conduits blood vessels and pump heart The advantage of this system is that the clinician only needs to observe a minimum number of variables during stable situations A disadvantage however is the fact that the complexity of the system increases rapidly with a growing number of states Fukui 11 uses an approach borrowed from Artificial Intelligence AI A number of patient variables including blood pressure ECG and temperature are con
70. s of the anesthetic system and the patient as well as real time raw signal waveforms monitored Furthermore the numerical values of the most important measured patient and equipment related variables is presented A schematic of the user interface screen designed for the prototype II system is pictured in figure 5 3 In order to effectively represent the state of the anesthetic system three different types of presentation are implemented These include a traffic light real time graphs and alpha numerical messages and An Intelligent Alarm System in Anesthesia 48 status Traffic light Pressure status Flow status Vent status Pressure Ventilator settings numerical values Measured quantities numerical values Alphanumerical alarm messages Program control messages Figure 5 3 User interface display screen numbers The left half of the screen is used to display waveforms representing the most recent 25 seconds of raw CO pressure and flow data The waveforms serve a quality control purpose when an alarm message would be generated caused by artifacts the clinician will immediately see the artifact in the raw signals The colored box in the right upper corner of the screen gives the clinician a first quick insight in the state of the anesthesia system The box is either green everything OK yellow a CAUTION message is generated by the expert system or red an ALARM condition is detecte
71. s that we consider the time constants of the flow Tp and pressure signal Tp as constant The clinician will still have to reset the feature baselines to their running average value manually by pushing the RESET BASELINES button see 3 3 when due to mutations in the patient s condition changes in the signals occur This leaves the time related variables inspiratory time Tinsp expiratory time Texp and breath time T for each signal and the minimum pressure P nin the pressure slope value Pope the maximum pressure the maximum flow Fmax and the measured expired tidal volume Vy 4 as the features for which new baselines have to be calculated after setting changes An Intelligent Alarm System in Ancsthesia 43 5 1 3 Simple Breathing Circuit Modeling Van Oostrom 28 derived some formulas for the variables mentioned in 5 1 2 The basis for the derivations is a simple first order electrical model for the pressure and flow signals in the breathing circle In figure 5 1 the electrical model for the inspiration is pictured while in figure 5 2 the expiratory model is given During inspiration the ven tilator acts as a current source a constant flow F is forced into the lungs through the inspiratory hose while the expiratory flow Fg is zero At expiration the driving force of the ventilator is removed the flow through the inspiratory hose Fy equals zero and the lungs empty passively through the ex
72. st prototype is the fact that the feature baselines have to be reset after every change in ventilator or FGF settings One of the major goals for the second version is to implement an automatic baseline reset This should prevent false alarms and missed detections due to setting changes Another issue is the fact that the system does not consider a flat signal and an invalid signal as fundamentally different Implementation of an extra signal flat status will also be performed in the second version An Intelligent Alarm System in Anesthesia 40 CHAPTER 5 IASA THE SECOND PROTOTYPE As mentioned in paragraph 4 3 the most important limitation of prototype I is that the user must reset the feature baselines see 3 3 whenever he changes the ventilator or fresh gas flow settings a setting change is performed very often in clinical practice If he fails to do so the alarm system may generate false alarm messages or miss occurring critical events The main goal of prototype II is to implement a feature baseline reset that is performed automatically whenever one or more of the settings in the anesthesia system is changed The prototype II system should detect malfunctions occurring in the breathing circuit or the ventilator system patient related malfunctions will not be incorporated yet The second major goal is to devise a user friendly interface screen for the intelligent alarm system Theory and implementation of all important new attribut
73. t to the expert system after each breath period 3 4 The Real Time Expert System Approach The goal for the first prototype of the Intelligent Alarms System is to provide the user with a conclusion about the integrity of the anesthesia breathing An Intelligent Alarm System in Anesthesia 26 eircle after every breath period Van Oostrom 1 explains the reasons for choosing the SIMPLEXYS Expert System Language developed at the Eindhoven University of Technology Eindhoven The Netherlands by Blom 25 26 as a tool for our implementation The main advantage of SIMPLEXYS is the fact that it provides hooks to a high level programming language currently SIMPLEXYS versions for Pascal and for C are available and therefore data acquisition and graphical user interface routines can easily be interfaced with the expert system body After compilation this provides a fast and efficient program The SIMPLEXYS language was especially designed for real time expert system applications The expert system body consists of a set of rules that contain the expert knowledge in the system Starting with one or more root rules the complete rule set or a part of the rule set is evaluated after every breath Every rule to be evaluated gets assigned a value TR True FA False or PO Possible Depending on the result other rules may be triggered and evaluated as a consequence For every malfunction in the breathing circle to be detected a rule des
74. th patient and equipment However the number of monitors increases steadily in today s operating room OR making the situation more and more complex Each monitor is equipped with its own alarms generating a sound or displaying a message whenever a measured quantity exceeds its user defined threshold During a number of situations many monitors sound an alarm simultaneously making it very difficult for the clinician to diagnose the situation and correct an emerging problem A few years ago the Intelligent Alarms Project was started as a joint venture between the University of Florida Gainesville U S A and the Eindhoven University of Technology Eindhoven the Netherlands with the intention to develop a working prototype of an integrated alarm system This system should combine several signals in order to come up with intelligent alarm messages whenever an abnormality is detected during the course of anesthesia In 1988 Van Oostrom 1 implemented a first version that processes three signals measured in the anesthesia breathing circuit transforms the signals into symbolic feature data and evaluates and combines these symbolic data using a rule based expert system From this first prototype a refined system was developed and tested The system concentrates on mechanical malfunctions in the breathing circuit After an introduction to anesthesia chapter 1 and current alarm strategies chapter 2 the configuration of the final version of th
75. the new setting values into formulas 5 1 to 5 3 and 5 7 to 5 11 At these moments the constants are considered the unknown components in the formulas This way the algorithm will adapt to slight changes in compliance or resistance due to external circumstances like the surgeon opening up the patient s chest Preliminary tests showed that all assumptions are valid except the assumption that an increase in PEEP does not influence the constants R C RC and K Whenever the PEEP value is increased the alarm system is not able to see the difference between a PEEP setting change and an obstruction in the expiratory limb of the breathing circuit This can be explained by the fact that turning up the PEEP valve physically causes the expiratory resistance to increase In case of an obstruction the resistance is also increasing A solution to this problem would be to build a sensor on the PEEP valve that senses when the clinician turns it Then the system will detect the difference between a change in PEEP setting and a change in minimum pressure as a result of a malfunction Since this solution was not feasible the remaining solution is to let the anes thesiologist reset the feature baselines manually after a change in PEEP 5 2 The User Interface The basic purpose of the user interface is to show the status of the anesthetic system to the clinician controlling it The items to be presented on the display screen include the high level state
76. the system was still considered correct From table 4 2 we see that from the total of 189 events simulated 167 were detected correctly while 20 were not detected and 2 false alarms were recorded This means that 8896 of the mishaps was detected correctly within 30 seconds The two false alarms were an incompetent inspiratory valve message at a small and at a large leak in the E T tube To explain the false alarms the essential parts of the rules for incompetent valve large leak and small leak are given in table 4 3 When leaks were introduced at the E T tube we observed that the CO down stroke sometimes went down In that case the only difference between a small leak and an incompetent valve is the maximum flow feature The difference between a large leak and a stuck inspiratory valve is the fact that the pressure should go down in case of a large leak and remain unchanged during a stuck valve Since the maximum pressure did not change the false alarms were generated This problem was solved after the test procedure by changing the rules so that when the pressure is not going down during a leak in the E T tube two possible An Intelligent Alarm System in Anesthesia 34 Table 4 2 Matrix with test results of IASA prototype I on the Gainesville Anesthesia Simulator Malfunction Setting 1 Setting 2 Setting 3 Setting 4 Stuck exp valve XXXX X X X Stuck insp valve XXXX X X X Exh CO absorber XXXX X X X Obstr insp hose XXXX X X
77. tinuously sampled and translated into one of three symbolic values By comparing them to certain predefined thresholds each variable gets assigned a value high low or just right Pattern recognition techniques are used to identify abnormalities in the patient system Fukui has implemented this algorithm together with a graphical display on which among others the blood pressure trend is displayed Furthermore a sad or happy cartoon face gives an immediate and easy to interpret indication about the patient s condition Beneken et al 12 use systems engineering principles in their approach They consider the measured output quantities as a function of input variables actual outputs a noise component and time They use a patient model together with a library of fault models each possible fault will have a certain unique effect on the measured output quantities The algorithm reduces each measured variable to a three digit number which indicates the static dynamic and stochastic An Intelligent Alarm System in Anesthesia 11 properties of the respective variable Each fault model is stored as a set of these three digit numbers At any moment the algorithm can decide whether the patient is OK and if not which is the most probable fault Recently Philip 13 and Fukui 14 published an update on their ongoing research regarding their alarm algorithms mentioned before Also Beneken et al 15 looked at alarms and their limits on whic
78. tive upper and lower threshold values see figure 3 3 to detect when a signal switches from one state to another 24 For example when the pressure signal crosses the lower threshold in the upper direction and the signal is currently in state 4 a switch from state 4 to state 1 will be performed After every sample the algorithm first checks in which state the signal was and then whether a state transition is detected Depending on the new state one or more variables that characterize the signal are updated An Intelligent Alarm System in Anesthesia 20 thereafter For the pressure signal the variable to be updated in state 1 is the pressure derivative or slope value Psjope In state 2 this is the maximum pressure Pax whereas the time constant of the down stroke Tp has to be estimated in state 3 Finally in state 4 the algorithm looks for the minimum pressure Pin Other variables have to do with the timing and include inspiration time expiration time and breath time The theoretical value of these variables is indicated in figure 3 3 The inspiration time is estimated by adding the time periods of state 1 and 2 the estimated expiration time is the sum of the state 3 and state 4 time breath time is the sum of both For the pressure signal the transition from state 2 to state 3 is considered the start of a new breath period For the flow signal a breath period starts with the detection of the jump in expiratory flow which ends the z
79. to get the average fresh gas flow during one breath period The FGF result is corrected for non linear sensor error by means of a look up table of correction factors as a function of the average FGF value The look up table is created based on the measured inaccuracy of the FGF result at a flow of 100 So at breath detection a new value for RR EE PEEP all coming from the ventilator and FGF coming from the second flow device is available These new setting values are compared to the settings directly after the most recent baseline reset If the relative difference is 1096 or more and if FGF gt 0 ml sec a negative measured FGF value means a disconnect of the fresh gas hose rather than a setting change new feature baselines are calculated with help of formulas 5 1 to 5 3 and 5 7 to 5 11 After that the current setting values are stored as the new reference values However the constants RC R C and K can show slight changes during the course of anesthesia The algorithm will update a moving average value for these four constants after each expert run if 1 a new and valid set of settings is available 2 all numerical features are declared valid and 3 no malfunction is detected in the breathing circle The updated moving average values are obtained by filling in the values of Vime Fmaw P Pis Pstope max and Tins calculated by An Intelligent Alarm System in Anesthesia 47 the signal processing routines and
80. tware package DN NV or FL see also chapter 3 As a last change the whole structure of the software package is altered The new program structure is schematically pictured in figure 5 4 This structure has the advantage that the Intelligent Alarm System software is easily portable from one system to another since the machine dependent data acquisition and user interface routines are separated from the expert system and signal processing routines In future the system can be implemented into an anesthesia machine by only reprogramming the routines that interface the main module with the low level I O routines In appendix 1 all rules at the two highest levels of the rule base tree structure are presented In appendix 2 a flow chart is given that pictures the data flow in the prototype II alarm system software in a somewhat simplified fashion In appendix 3 a short formal description of the five routines that interface the main module with the low level I O routines is presented while An Intelligent Alarm System in Anesthesia 51 in appendix 4 a list of detailed technical data regarding the software and hardware composing Prototype Il is given An Intelligent Alarm System in Anesthesia 52 CHAPTER 6 TESTING PROTOTYPE II The second prototype described in chapter 5 has been tested similar to the first prototype Although tests with dogs and subsequently tests in the operating room OR are planned for the future prot
81. tware was already written in C the Pascal routines in the expert system body were translated into C The use of MultiDos Plus as a multitasking extension to the MS DOS operating system caused a lot of overhead because lots of data had to be sent from one task to another and back Therefore we choose to pause the signal processing during the feature extraction and expert system evaluation that is performed once a breath The incoming samples are stored in a buffer during this short period since the three different programs signal processing feature extraction rule evaluation do not run concurrently anymore Because all the functional parts are now in the same program all pertinent data is automatically accessible for all routines Since the C version of SIMPLEXYS is able to evaluate approximately 2000 rules per second and the prototype I system contains about 60 rules the pause period is very short and provides no problems for the data acquisition The one program approach also overcomes the major drawback imposed by MultiDos Plus no support for high resolution graphics Thus the real time graphs of the CO pressure and flow waveforms can be presented on the same screen as the alarm messages now The raw samples do not have to be sent to a second PC anymore the whole system is implemented on one IBM AT compatible computer Datalogging capabilities for storing the raw flow pressure and CO data on hard disk are implemented
82. ually introduced were leaks in different hoses obstructions of different hoses and disconnections of hoses other than the ventilator hose Table 4 1 Setting combinations used during the tests at the simulator Setting no Vy ml RR breaths min EE l min 1 500 12 1 2 6 2 500 12 1 2 3 3 750 6 1 2 6 4 750 6 1 2 3 An Intelligent Alarm System in Anesthesia 33 Leaks of two different sizes were introduced in the various hoses by connecting an open tube with a length of 7 5 cm and a diameter of either 1 5 mm small leak or 3 mm large leak to the system at the respective locations Obstructions were simulated by pinching the hoses For setting combination 1 the critical events were repeated 4 times to test the consistency of the system The results of all the tests are given in the matrix in table 4 2 In this table two different locations for leaks in the expiratory hose can be distinguished Leak 1 was introduced upstream at the lung side of the flow sensor connection while leak 2 was inserted downstream of the flow sensor In the matrix an X indicates that the correct or best available alarm message was generated within the 30 seconds time span an N means that no alarm was triggered at all and an F means that only false alarm messages were generated When more alarm messages appeared indicating the system detected more than one possible malfunction but the correct message was one of them the reaction of
83. value For example when the V4 setting or the FGF setting was decreased a small leak or large leak message was generated during several operations Our method of using the serial port of the computer for counting pulses coming from the 5410 volume monitor to generate a real time flow signal see 8 3 2 turned out to be sensitive to electrocautery BOVI BOVI sometimes caused high frequency interference on the flow signal When the surgeon was pushing the abdomen or chest of the patient one or more of the signals were sometimes considered invalid by the alarm system see chapter 3 Despite this fact the system most of the times was able to generate the correct decreased compliance message The signals were never invalid longer than two breath periods In general the patient data were smooth enough to allow correct calculation of the signal features during nearly the entire operations An Intelligent Alarm System in Anesthesia 38 When an artificial nose a device inserted between the Y piece and the E T tube to moisten the inhaled gases was used the airway resistance sometimes went up This was due to water partially blocking the airway inside the artificial nose As a result a correct obstruction E T tube message was generated When the anesthesiologist accidentally pushed the E T tube too far in the patient s trachea this was also recognized as an obstruction in the E T tube Finally a plugged E T tube was correctly
84. w an exponential down slope much like the discharging of a capacitor via a resistance whereas the CO signal increases quickly to an end expiratory plateau value This value is approximately equal to the alveolar CO concentration With another inspiration the patient will inhale fresh gas without CO and thus the signal will go down to zero Due to the fact that the CO monitor removes gas from the Y piece via a sampling line a delay is observed in this signal compared to the flow and pressure recordings As can be seen from figure 3 2 gas transport from the Y piece to the monitor takes approximately 2 5 seconds An Intelligent Alarm System in Anesthesia 19 3 2 2 From Signals to Signal Features For all three signals a breath detection and feature extraction algorithm is implemented that divides one breath period of each signal into several states To explain the concepts of these algorithms we take the pressure signal as an example In figure 3 3 the signal is pictured together with some help variables Signal states Upper detection Lower detection threshold i Inspiration Expiration time i time Breath time eee Figure 3 3 The pressure signal divided into several states The pressure signal is divided into 1 a linear inspiratory up slope 2 a maximum pressure state 3 an exponential expiratory down slope and 4 a low end expiratory plateau The algorithm uses adap
85. y difficult for the anesthesiologist to diagnose the problem because the alarms are not specific and do not point out possible causes Other problems that can be identified are 2 t becomes increasingly impractical for the clinician to set all the thresholds for all the variables manually The alarms do not give an early indication when the situation is slow ly deteriorating without variables exceeding thresholds There is no priority scheme for alarms coming from different monitors All triggered alarms will ring the bell simultaneously It is very difficult for a single variable monitor to detect the difference between a potentially hazardous situation and a threshold crossing triggered by a motion or other interference This is the well known artifact detection problem 9 For all these reasons research efforts in the last few years have focused on how to generate helpful and specific alarms Recent publications point out that multivariable analysis and integration is needed in order to come to more intelligent alarm messages A brief survey of recently proposed alarm strategies their problems and examples of implemented and tested prototype systems is presented in the next paragraph 2 2 Survey of Modern Alarm Strategies The first attempt to compile a survey of research activities regarding alarm strategies and implementations other than the threshold check was performed by Gravenstein et al in 1987 8 In this su
86. ype II during multiple malfunctions Malfunctions Triggered alarm message s Small leak exp hose Obstruction exp or vent hose Obstr exp hose Stuck insp valve Incompetent inspiratory valve Obstr exp hose Small leak site unknown or when compiete occlusion Apnea Stuck insp valve Incompetent inspiratory valve Small leak exp hose Small leak site unknown Stuck insp valvc Incompetent expiratory valve Stuck exp valve Obstr exp hose Obstruction E T tube Y piece Obstr insp hose As can be seen from table 6 4 in nearly all cases at least one of the triggered alarm messages was a correct one When a combination of a stuck inspiratory valve and a complete obstruction of the expiratory hose is introduced the incorrect Apnea message is triggered This situation can be compared to the stuck inspiratory valve in combination with a high PEEP setting described in 6 1 Since the flow signal becomes completely flat this alarm is generated by the An Intelligent Alarm System in Anesthesia 57 ventilator Because no other intelligent alarm message is triggered the ventilator message is copied by the system and displayed on the screen When the inspiratory and the expiratory hose are obstructed simultaneously the logical result is an Obstruction E T tube Y piece message In both cases both the inspiratory and the expiratory resistance increases so the pressure and flow signals are influenced in

Download Pdf Manuals

Related Search

Related Contents

Manual de Instruções  Medisana Bloodpressure monitor MTP  LT408 Service Manual A  User Manual - IVB 995-OH_M SD XC TYPE 22  Uterine Cervix Cancer of High-risk HPV Genotype Related  User Manual  PVI-DESKTOP  Monsieur le Pharmacien Responsable SANOFI AVENTIS FRANCE  Patton electronic 1082/I Network Card User Manual  TrueLok - Orthofix  

Copyright © All rights reserved. Failed to retrieve file