Home

Validation of muscle relaxation measurements

1. 5 3 3 Delay between minimum and maximum peaks During the manual validation see chapter 6 the time delay between the minimum and maximum peaks did not seem to vary much so it could be a good indicator for the validity of the signal Figure 5 16 shows the results and table 5 14 shows more statistical data By definition the mean of Lui Lmax is equal to the difference of their respective means The mean value of this parameter shows a slightly decreasing trend over the four ECAPs just like the Lain and Lmax values themselves It seemed that the mean ratio of Lmax Lmm was constant for all four ECAPs 0 667 standard deviation 0 023 so the response seemed to be scaled in time No further explanation for this could be found Lmin Lmax 1 Lmin Lmax 2 600 600 400 400 200 200 o amna nn ol amen n i 0 02 0 01 o 0 01 0 02 0 02 0 01 o 0 01 0 02 Lmin Lmax 3 Lmin Lmax 4 600 4 600 400 1 400 200 4 200 o anmann iimmm o mnnon a 0 02 0 01 oO 0 01 0 02 0 02 0 01 o 0 01 0 02 Figure 5 16 Histograms of Lum Luax values in s The average value is 3 2 ms and the standard deviation is 5 7 ms Table 5 13 More statistical data on Lam Luax values in ms Mean Minimum Minimum of TOFs scored valid Maximum Maximum of TOFs scored valid 45 5 3 4 Latencies of zero crossings A zero crossing detection algorithm was designed to detect if and where zero crossings occurred
2. For the serial data link wiring and protocol the interested reader may refer to Smans 1993 The following extra features that were not described in the manual were noted When turned on the Relaxograph sends a character FFh and when switched off the Relaxograph sends a character 00h This feature has been utilized in the software see next chapter 2 5 Interfacing to AS 3 ADU There are several ways to receive measured data from the AS 3 ADU There is a high speed serial link 119 200 baud that uses an advanced protocol for the interfacing to many physiological parameters It also provides on line access to several types of digitized raw waveforms like the ECG and capnogram Unfortunately the NMT s EMG waveforms are not among these types Another possibility to receive raw data including the NMT measurements is via the so called UPI board connector The ADU can be configured to export several waveforms via this connector In this way up to 16 different signals are available in analog form with voltages between 5 and 5V In a first test it was noted that the signals are not the actual analog signals but internally D A converted versions of A D converted measurements The signal used as a NMT trigger runs from the collector of transistor V26 to the gates of FETs V2 and V25 and is marked D on the printed circuit board It was connected to pen 8 of the analog output via a 1 KQ resistor The modification was approved by the Ca
3. 800 600 600 400 400 200 200 Mish it thin in 0 1 0 05 Oo 0 05 0 1 0 1 0 05 Oo 0 05 0 1 Figure 5 3 Histograms of average DC voltage in V of ECAPs 1 to 4 Average value 7 2 mV standard deviation 38 0 mV Table 5 2 More statistical data on Voc values in 0 000211 0 00847 Average value Minimum 0 2433 0 3659 Min of TOFs scored valid 0 2300 0 2279 Maximum 0 7672 2 087 Max of TOFs scored valid 0 7672 0 2864 Vde1 1 Vdc1 2 800 r 800 600 600 400 400 200 j 200 ol den alli ts BEE Mh 0 1 0 05 o 0 05 0 1 0 1 0 05 oO 0 05 0 1 Vde1 3 Vdc1 4 800 a 800 as 600 600 400 400 200 200 gba ke ohanaa 0 1 eo Wil 0 1 0 05 o 0 05 0 1 0 05 o 0 05 0 1 Figure 5 4 Histograms of average voltage in V of ECAPs 1 to 4 in the 0 4 ms interval Average value 30 5 mV standard deviation 52 6 mV 36 The part of the signal between O and 4 ms does have a positive DC component as can be seen in figure 5 4 Peaks occur at circa 15 mV and at 80 mV Table 5 3 More statistical data on Voc values in Mean Minimum Minimum of TOFs scored valid Maximum Maximum of TOFs scored valid Vdc2 1 Vdc2 2 600 400 200 A ORS 0 1 0 05 o 0 05 0 1 Vdc2 4 600 4 a 400 400 200 200 0 1 0 05 oO 0 05 0 1 0 1 0 05 oO 0 05 0 1 Figure 5 5 Histograms of the average voltage of ECAPs 1 to 4 in the 4 15 ms int
4. A low level communications driver contained in the RS232 unit serves to receive and send bytes 25 out to two RS232 ports The RS232 unit contains character level serial interface routines to communicate with the Relaxograph and the pump via COM1 and COM2 Since MS DOS does not support serial communications with no handshaking using a three wire cable standard DOS interrupt service routines could not be used A new interrupt service routine is installed that stores the incoming characters in a rotating local buffer A flag is set to indicate if data is available to the rest of the program This unit has no function in the AS 3 version Main program Control mm Screen Ee __ EMG_ Validation Processing Relaxograph Pump Timer AD_Routines RS232 FilelO L General italic to be constructed dotted line link only present in Relaxograph version Figure 3 1 Unit hierarchy proposed for the final controller program Blocks with a gray background are hardware The Relaxograph unit serves as a shell around the RS232 unit that handles the Relaxograph s serial communications protocol and keeps track of its current state It can return the serial data in a data structure of type RE_RelaxogrType as well as in a formatted string that is suitable for screen output It also keeps track of the operating mode of the Relaxograph as well as possible In the AS 3 version t
5. Frans de Kok van de medisch fysische instrumentatie dienst werkte mee aan het praktische gereedmaken van het meetsysteem Verder dank ik Hans Blom voor de goede begeleiding en idee n en ook alle andere medewerkers van de sectie E M E voor de praktische ondersteuning en vooral de prettige sfeer Ron van der Zwaluw van de firma Datex Medical Electronics was behulpzaam met het oplossen van een aantal technische vragen Tot slot wil ik vrienden bekenden en bovenal mijn ouders bedanken voor de interesse en grote steun tijdens de afstudeerperiode Marco van Steen Table of contents 1 Introduction 1 1 Backgrounds 1 2 Control of muscle relaxation 1 3 Data acquisition 1 4 Validation of the measurements 1 5 Formulation of the project 1 6 Contents of this report 2 Hardware 2 1 Measurement of muscle relaxation 2 1 1 Train of four response 2 1 2 Signal processing by the NMT monitor 2 2 Interfacing to Relaxograph and to AS 3 ADU 2 3 A D conversion board 2 3 1 Selection of a data acquisition board 2 3 2 Characteristics of the DAS 1402 board 2 4 Interfacing to the Relaxograph 2 4 1 Relaxograph trigger signal 2 4 2 Analog EMG output 2 4 3 Serial data link 2 5 Interfacing to AS 3 ADU 2 5 1 AS 3 ADU data acquisition chain for NMT signals 2 5 2 AS 3 ADU NMT trigger signal 3 Software of the measurement system 3 1 Design method 3 2 Survey of the units 4 Validation methods for TOF signals 4 1 Demands to a validation al
6. In this paragraph we will focus on the change in ECAP parameters within one train of four 5 4 1 Change of T ina TOF From visual inspection it was clear that the shape of undisturbed ECAPs of all four twitches was almost equal except for the amplitude The amplitude of the four twitches decreases exponentially due to muscle fatigue With low levels of relaxation the fade is minimal It increases with higher levels of relaxation The difference ECAPa ECAP1 is therefore expected to be negative most of the time T2 T1 T3 T2 2500 2500 2000 2000 1500 1500 1000 1000 500 500 DET y dl 0 02 0 01 oO 0 01 0 02 0 02 0 01 Oo 0 01 0 02 T4 T3 T4 T1 2500 2500 al 2000 1500 1500 1 S 1000 500 500 oO LS6 ee ee en 0 aL ae LET te ee a A 0 02 0 01 Oo 0 01 0 02 0 02 0 01 o 0 01 0 02 Figure 5 22 Histograms of the difference between the T parameter between two ECAPSs of the same TOF Overall mean 1 3 mVs standard deviation 2 9m Vs Table 5 18 More statistical data on Tn Tn 1 values in mVs he TE Be Te Mean Minimum Minimum of TOFs scored valid Maximum Maximum of TOFs scored valid 50 5 4 2 Change of No in a TOF The change in the number of zero crossings within normal TOFs is only small Although valid measurements exist with NOn NOm over the whole range from minimum to maximum the larger values were in many cases associated with artefacts Therefore this parameter seem
7. NMT trigger signal 24 3 Software of the measurement system 3 1 Design method The software has been designed in a modular fashion using structured programming techniques in Borland Pascal based on program modules called units This lead to a number of units corresponding to the various tasks and physical parts of the system A very important design task was to choose a logical and consistent structure of units After that the units were designed and tested separately In order to produce a readable and maintainable program the following programming rules have been obeyed 1 All variables in units are invisible outside of the unit To get data out of or into a unit the user calls procedures or functions that return the data via their parameters 2 Inside the units variables may be shared This helps to keep the number of parameters low because the procedures and functions may access this data directly 3 Datatypes are defined in the units that produce data of these types 4 Some naming conventions are obeyed in the whole program e All types procedures and functions that a unit shares with the outside world have a two letter prefix indicating the unit followed by an underscore For example the SC_MsgBox function is in the screen unit It displays a message in a rectangle and asks the user for input It returns an SC_MsgBoxType variable to indicate the user s choice e Constants have a prefix indicating either the unit
8. Performance of the data acquisition systems 6 1 1 Accuracy of Relaxograph Labmaster system The learning set was measured using the Relaxograph NMT 100 combined with a Labmaster A D board The T parameter values as calculated by the PC have been compared to the Relaxograph s own measurements sent over the serial link Results are shown in figure 6 1 and figure 6 2 100 T T T Tt T 8 A D board software 80 Relaxograph Tt Tr 24 10 20 30 40 50 60 70 80 time min Figure 6 1 Ti Try calculated by the Relaxograph and by the PC measurement system When comparing the Ti Tre produced by the Relaxograph and by the PC a minor difference is noted Furthermore the PC s measurements seem to react differently to disturbances for example at the start and after 58 minutes This may be due to some averaging algorithm in the Relaxograph s software Figure 6 2 shows the correlation between the two calculations For this plot the quantisation error of the Relaxograph may be noted at low T Tret levels The mean absolute error was 3 38 of Tre Since the serial port of the AS 3 NMT was not used this comparison cannot be made for the system which acquired the test set 57 Figure 6 2 Correlation between Relaxograph s and PC relaxation measurements uncalibrated measurements were excluded Approximation y ax b with a 0 9534 and b 1 2915 Vertical axis T Tre according to R
9. Values greater than 100 or smaller than 100 were often due to artefacts This parameter seems to be suitable for validation purposes Note that normal values for Vmax T range from 50 to 200 s so the relative change within each TOF is small Table 5 21 More statistical data on Vnax T Vuax T Values are in s tax T N Vmax T Vatax T s Mean Minimum Minimum of TOFs scored valid Maximum Maximum of TOFs scored valid 52 5 5 Change of parameters in successive TOFs The change of several TOF parameters over time was studied The parameters of a TOF were compared to those of the previous TOF that was considered valid by the algorithm The difference of each parameter e g Vmax that was calculated in this way was assigned to a new parameter e g AVmax The new parameters have been analyzed like the previous parameters Since in steady state the controller should warn the clinician after circa 5 subsequent invalid measurements the previous valid measurement will never be older than circa 2 minutes For this analysis however valid measurements older than 2 minutes were also compared to 5 5 1 Change of T in successive TOFs The histograms show that AT has a normal distribution with the exception that there are a little more positive than negative values because the increase of T in the recovery phase takes more time and is present in more measurements The rapid decrease of T during the onset phase i
10. be 100 These twitches showed no fade the 4th of a TOF was as strong as the 1st They occurred most clearly in the MO9 TRF file The EMG response shape changes gradually as relaxation deepens Especially the number of peaks seemed to increase when relaxation increased The EMG signals during deeper levels of relaxation show relatively large variations in amplitude but the number and place of the small maxima and minima seem to stay constant The validation of these small signals proved to be more difficult and a little more arbritrary than the larger and clearer signals Table 6 2 Results of the validation by eye for the learning set and for the test set EER 6297 91 5 of learning set Test set 4975 97 0 of test set Total 11272 735 6 2 4 Results of automatic validation The limits mentioned in chapter 5 were applied to the signals of the learning set The results were compared to the golden standard set by manual validation by eye table 6 3 Table 6 3 Performance of the validation algorithm on the learning set compared to the manual inspection by eye In the columns the results of the validation by the algorithm are shown while the rows show the outcome of the validation by eye Automatic Valid Invalid Total By eye Valid e The algorithm detected 86 9 of all artefacts e The algorithm considered 64 8 of the measurements valid e In 66 7 of the case
11. by artefacts or not The basic assumption in this is that measured EMG waveforms contain enough information to judge their validity This assumption seems reasonable because as was seen in EMG data previously recorded by Joost Smans most sources of artefacts cause visible distortions in the EMG signals Although the waveform varies greatly between patients and during operations in general the variation between two successive valid measurements is limited In case of deep relaxation when the signal level is low validation will probably be more difficult because the signal is noise like So by qualitative and quantitative analysis of EMG signals combined with knowledge about the electrophysiology of nerves and muscles we may gather knowledge about the shape of correct EMG signals This knowledge may be expressed in simple rules that can be implemented in a computer program The performance of this program should be tested by comparing it to some golden standard Since experts on the visual interpretation of muscle relaxation signals are hard to find I decided to judge the signals by myself 1 5 Formulation of the project As pointed out in the above paragraphs two main goals were identified e Develop a real time measurement system for muscle relaxation Study the usefulness of the existing software and develop software for a real time measurement system which reads in the neuromuscular transmission monitor and presents the musc
12. in each ECAP The algorithm scans all samples of the ECAP in time order A zero crossing was then defined to occur if the sign of the current sample is opposite to the sign of the last non zero sample The index nz of the sample to which a crossing was attributed was calculated as follows n nb round ne na 2 where ne is the index of the last non zero element before the crossing na is the first non zero element after the crossing and round x is the integer number nearest to x In this way it was possible to study the number and latencies of the zero crossings The latency of the first zero crossing is shown in 5 17 LO1 1 Lo1 2 1000 1000 500 500 o ahd al il ee aa ede Ae o Dre TA O o 5 10 15 20 o 5 10 15 20 x 10 x 10 L01 3 LO1 4 1000 1000 500 500 o Sheer o a o 5 10 15 20 o 5 10 15 20 x 10 x 10 Figure 5 17 Histograms of the latency of the first zero crossing Loi in s The value 2 ms was assigned if no zero crossing was found Table 5 14 More statistical data on Lor values in ms The value 2 ms was assigned if no zero crossing was found PC CAP ECAP2 Mean 6 0 5 2 Minimum Minimum of TOFs scored valid Maximum Maximum of TOFs scored valid 4 7 4 8 2 0 46 Lo2 1 Lo2 2 2500 2500 2000 2000 1500 1500 1000 1000 500 500 KE 5 10 15 20 Oi 5 10 15 20 x 10 x 10 L02 3 L02 4 2500r 2500 2000 2000 1500 150
13. in histograms 4 Suitable bounds for the parameters were determined 5 The criteria were applied to the learning set and the results were compared to the visual inspection 6 The algorithm was verified with a test set of measurements that is independent of the learning set 7 If necessary the algorithm should be optimized by repeating steps 2 through 6 until the results are satisfactory To assure the independency of the test set a new test set should be acquired and used in the iteration Steps 1 through 6 were carried out Without the optimization step the algorithm was able to detect circa 85 of all artefacts A large number of measurements was incorrectly considered invalid and this number was just on the limits posed by the controller s needs in the steady state phase and below the demands during the onset phase Ways to optimize the algorithm are re evaluation of the visual inspection finding parameters that are still more independent of the level of muscle relaxation and tuning the threshold values Voorwoord Vanaf deze plaats wil ik graag al degenen bedanken die op welke manier dan ook hebben meegewerkt aan het afstudeerwerk dat in deze scriptie wordt beschreven Ten eerste dank ik dr Erik Korsten voor het mogelijk maken van de metingen in het Catharina Ziekenhuis voor zijn enthousiasme en de niet aflatende stroom idee n De anesthesie assistenten toonden interesse en een waardevolle kritische blik tijdens de operaties
14. mms 0 0 05 0 1 0 15 0 2 0 25 0 3 0 35 0 4 0 45 0 5 traction of sample frequency Figure 2 5 Transfer function of the 41 point moving average filter In the Relaxograph version the sample frequency before filtering is 50 kHz while in the AS 3 version the sample frequency before filtering is 75 kHz 2 5 2 AS 3 ADU NMT trigger signal A second output on the UPI board was configured to output an NMT trigger signal This signal is normally OV and changes to 5V at the start of each TOF It stays high during 1 510 seconds The trigger goes low before the last response has faded 23 Trigger sequence in case of medium supramaximal stimulus w T t gt 1E i trigger o l simulated EMG response tp y Ji J 2 a_l 1 ni a_l 0 5 10 15 20 25 30 time s Figure 2 6 Calibration cycle of NMT in AS 3 ADU with NMT simulator connected The simulator s response duration is 11 ms in real time In this case it takes five stimuli to find the supramaximal stimulation level The second group of four twitches is the reference measurement During calibration the trigger goes high for every stimulation In normal operation there is one trigger for every train of four The calibration cycle of the AS 3 ADU differs slightly from the Relaxograph s calibration cycle It is shown in figure 2 5 In this figure the response of an NMT simulator connected to the NMT input is shown together with the
15. of parameters is relatively independent on the relaxation level and is more sensitive to artefacts Certain values usually occur in valid TOFs and more extreme values occur when artefacts are present To these parameters narrower bounds can be applied As discussed earlier it was expected that parameters would have a more or less gaussian shape with clearly visible outliers caused by artefacts However in most histograms this distinction could not be made that easily and the choice for the parameter bounds was not obvious In these cases the bounds were derived from the invalid percentage diagrams Such a diagram shows the number of measurements that was considered invalid by visual inspection as a function of the parameter p The number of measurements is expressed as a fraction of all measurements with the given parameter value gt jen kel fraction of the TOFs considered invalid by eye invalid aca invalid parameter p Figure 5 28 Relationship between dp and the range in which the algorithm will consider the signal valid If for a certain range of parameter values this fraction is higher than some threshold dp O lt de lt 1 the validation algorithm should consider measurements in that range invalid For the sake of simplicity the criteria should be of the type plower bound lt p lt pupper bound Or Pp lt Pbound or p gt gt Pbound By varying dp
16. relaxation it could be interesting to see if and how this signal can be used to improve the controller performance 64 8 References Blom JA The SIMPLEXYS experiment Real time expert systems in patient monitoring Eindhoven Eindhoven Technical University 1990 Dissertation DAS 1600 1400 Series user s guide Rev B 1996 Keithley Instruments Inc 440 Myles Standish Blvd Taunton MA 02780 DAS 1600 1400 1200 Series Function Call Driver User s Guide Rev B 1995 Keithley Instruments Inc 440 Myles Standish Blvd Taunton MA 02780 de Graaf PMA Datareductie als basis voor validatie van fysiologische signalen Graduation report TUE 1993 Metingen in de geneeskunde I Various authors Reader TUE Feldman S Neuromuscular block Butterworth Heinemann Oxford 1996 Franklin Gene F Powell J David Emami Naeini Abbas Feedback control of dynamic systems Addison Wesley 1994 Hines A E Crago P E Chapman G J Billian C Stimulus artifact removal in EMG from muscles adjacent to stimulated muscles in J NEUROSCI METHODS 1996 64 1 p 55 62 Hoevenaren W M Ontwikkeling van een feedback controller voor spierrelaxatie met behulp van een expert systeem meetaspecten afstudeerverslag TUE 1992 Hoogendoorn Paul The design of a rule based blood pressure controller Graduation report TUE 1989 Kalli Ilka Monitoring of neuromuscular blockade by electromyography with special reference to clinical application in an
17. the map method by de Graaf de Graaf 1993 can be used A parameter that is the result of this method may be added to the existing algorithm e Search for better threshold values By shifting the threshold values or dp the performance of parameters may be tuned Perhaps a machine learning program could be used to find an optimal set of thresholds When the validation algorithm functions properly the design of the controller should be assessed e The use of a model predictive controller Schwilden Olkkola seems promising in terms of accuracy and robustness The advantage of using a model based predictive controller is that it is possible to specify future setpoints This allows for example that the clinician specifies in advance that the patient should recover within 20 minutes The controller may then 63 immediately adapt the infusion scheme in order to reach the target level in time Furthermore if a model of the patient s muscle relaxation is available and the parameters are adapted on line it may be possible to predict the time that is needed for spontaneous recovery at the end of the operation This may help to determine the moment when infusion of muscle relaxants should be stopped During the clinical trials it was observed that the capnogram expiratory CO level as a function of time was used by the clinicians to detect spontaneous breathing activity Since spontaneous breathing may be a sign of insufficient muscle
18. this way a golden standard is determined for the validation algorithm and insight in signal properties and artefacts may be gained Choose a large number of parameters that are based on a single twitch on the rate of change between the twitches of one TOF or on the rate of change between TOFs Calculate the parameters for every measurement in the learning set The results are presented in histograms Determine suitable criteria for the parameters Apply the criteria to the learning set and compare the results to the results of the visual inspection Verify the algorithm with a test set of measurements that is independent of the learning set If necessary the algorithm should be optimized by repeating steps 2 through 6 until the results are satisfactory To assure the independency of the test set a new test set should be acquired and used in the iteration The analysis and selection of parameters and the determination of criteria steps 1 2 3 and 4 are the topic of the next chapter Chapter 6 presents the results of applying these criteria to the learning set and to the test set steps 5 and 6 32 5 Parameter analyses In this chapter an extended analysis of the data in the learning set is presented A brief description of the learning set will be given in paragraph 1 Some parameters for a single ECAP response will be shown in paragraphs 2 and 3 In the fourth paragraph the relationships between the parameters of d
19. 0 1000 1000 500 500 f o 5 10 15 20 ee 5 10 15 x 10 x 10 Figure 5 18 Histograms of the latency of the second zero crossing La in s The value 2 ms was assigned if no zero crossing was found Mean value 2 8 ms standard deviation 2 3 ms Table 5 15 More statistical data on Lo values in ms after stimulus The value 2 ms was assigned if no zero crossing was found Mean Minimum Minimum of TOFs scored valid Maximum Maximum of TOFs scored valid 5 3 5 Number of zero crossings The results of a count of zero crossings in every ECAP over all measured TOFs are shown in figure 5 19 Zero crossing counts of over five did not occur too often although some of them were valid Figure 5 20 shows the percentage of TOFs that was judged invalid by hand as a function of No of the first ECAP of every TOF From this figure it is clear that as the number of zero crossings increases the signal quality decreases This is obvious since noisier signals are small have many fluctuations around zero so they will contain more zero crossings Table 5 16 More statistical data on No in all ECA Ps po CaP 1 ECAP2 ECAP3 ECAP 4 Mean Minimum Minimum of TOFs scored valid Maximum Maximum of TOFs scored valid 47 4000 3000 2000 1000 4000 3000 2000 1000 alee 5 Figure 5 19 Histograms of the number of zero crossings Ave
20. 3 NMT module and PC 6 2 Performance of the validation algorithm 6 2 1 Goal of manual validation 6 2 2 Method for validation by eye 6 2 3 Results of validation by eye 6 2 4 Results of automatic validation 6 3 Discussion 7 Conclusions and recommendations 7 1 Conclusions 7 1 1 Data acquisition system 7 1 2 Validation algorithm 7 2 Recommendations 8 References Appendix A Wiring of PC NMT monitor links Appendix B Validation parameters and their bounds 41 42 42 43 44 45 46 47 48 50 50 51 51 52 53 53 54 55 57 57 57 58 58 59 59 59 60 61 63 63 63 63 63 65 67 69 1 Introduction 1 1 Backgrounds In the servo anaesthesia project of the group of Medical Electrical Engineering E M E at Eindhoven University of Technology research is carried out on the question how computer and information technology may help improve the quality of anaesthesia given to patients in intensive care units and operating theatres One of the directions in this program is the development of automatic closed loop control systems that take over routine tasks from the anaesthetist Such tasks include stabilization of blood pressure and keeping the patient s muscles relaxed to a certain degree It is tried to develop systems that are suitable for clinical use on a routinely basis Benefits of such relatively simple control systems may be various The desired level of effect will be more constant and the patient wil
21. 4 3 ms The histograms in figure 5 13 show that most of the maximum peaks occur in the 5 10 ms interval As table 5 12 shows valid TOFs have Lmax values ranging from the absolute minimum 0 ms to the maximum 20 ms so it seems that this parameter is at least not decisive for the validation However when inspecting the percentage of invalid TOFs as a function of Lmax figure 5 14 one finds that very small values almost zero caused the TOF to be invalidated in circa 40 of the cases 1 Table 5 11 More statistical data on Lmax values in ms after stimulus Mean Minimum Minimum of TOFs scored valid Maximum Maximum of TOFs scored valid In table 5 12 it may be seen that the average Lmax decreases for every ECAP of a TOF This is remarkable Although the latency shift is small it is consistent No explanation of this fact has been found yet 43 45m 40 30F 257 207 15F Percentage of invalid TOFs as a function of Lmax 1 Sn EK T Ln Ji I 0 002 0 004 0 006 0 008 0 01 Fema i ea 0 012 0 014 0 016 0 018 0 02 Figure 5 14 Percentage of invalid TOFs as a function of Luax of the first ECAP Figure 5 15 shows the latencies of the minimum The histograms clearly show two distinct peaks a sharp one at 5 ms and a more rounded one at circa 12 ms Inspection of the E
22. 4361 99 9614 94 0276 92 9844 Minimum 371 7472 358 6498 362 1908 358 1662 Minimum of TOFs scored valid 371 7472 358 6498 362 1908 358 1662 Maximum 39 2066 45 5836 36 3306 38 0360 Maximum of TOFs scored valid 36 6894 36 8939 36 3306 38 0360 40 5 2 6 Ratio of peak peak voltage to T Ver T is expected to combine the above two parameters As can be seen in table 5 9 there is a clear maximum to this parameter of circa 530s that can be used for validation Table 5 9 More statistical data on Vre T values in s value 0 also assigned when T 0 PCC ECAP 1 ECAP2 ECAP3 ECAP4 Mean Minimum Minimum of TOFs scored valid Maximum Maximum of TOFs scored valid Vpp T 1 Vpp T 2 400 400 300 300 200 200 oh marie nnn d ld o 100 200 300 400 o 100 200 300 Vpp T 3 Vpp T 4 400 400 300 300 200 200 De ini k hi ok dl ol tll o 100 200 300 400 fo 100 200 300 400 Figure 5 10 Histograms of the ratio of peak to peak voltage to T of all four ECAPs Average value 226 s standard deviation 78 9 s 41 5 2 7 Ratio of DC to peak to peak voltage Vdc Vpp 1 Vdc Vpp 2 800p 800 600 600 400 400 200f 200 9 le 0 5 1 Q 0 5 1 Vdc vpp 3 Vdc Vpp 4 8soo y a 800 T l 600 600 al 400 200 200 oO 0 5 1 oO 0 5 1 Figure 5 11 Histograms of the ratio Voc Ver of ECAPs 1 to 4 Average value 0 139 stand
23. Acquisition is always initiated on command of the PC There is no provision for a hardware trigger that initiates the conversion without intervention of a software routine Each sample is stored in memory as a 16 bit word The digitized data called a count value are stored in the highest 12 bits while the channel number is stored in the lower four bits The lowest voltage in the input voltage range corresponds to a count value of 000h while the highest voltage corresponds to FFFh Therefore to calculate the voltage V corresponding to a given 16 bit word W the following formula should be applied V W SHR 4 AND OFFFh 2048 20 0 G 4096 where SHR denotes logical right shift and G is the gain used This formula is valid for use in bipolar input mode only The data transfer from the board to the PC can take place in one of four different modes which are supported by the driver software e In single mode the board acquires a single sample from an analog input channel with a given gain setting and returns it to the calling progam e In synchronous mode the board acquires a single sample or multiple samples from one or more analog input channels The calling program is halted until the specified number of samples have been acquired e In interrupt mode the board acquires a single sample or multiple samples from one or more analog input channels The device driver initiates the conversion and then retur
24. CAPs with minima at 5 ms showed that these minima appeared in cases of direct stimulation of the muscles and in cases of high relaxation levels At low levels of relaxation the amplitude of the 12 ms negative peak is much larger than the amplitude of the peak at 5 ms so the latency of the minimum is 12 ms When relaxation increases the 12 ms peak practically disappears but the 5 ms peak remains present so the latency of the minimum becomes 5 ms This means that an Luin value of 5 ms may be valid or may indicate direct stimulation The 12 ms minimum most often indicates lower levels of relaxation 5 3 2 Latencies of minimum peaks Lmin 2 ihn 0 01 0 015 0 02 Lmin 4 Lmin 1 1000 1000 800 800 600 600 400 400 200 ii 200 o 0 005 0 01 0 015 0 02 o 0 005 Lmin 3 1000 1000 800 800 600 600 400 400 200 200 o nd fe adil o 0 005 0 01 0 015 0 02 o 0 005 iil 0 01 0 015 0 02 Figure 5 15 Histograms of the latency of the minimum peak in s after stimulus 44 Table 5 12 More statistical data on Lum values in ms ECAP 1 ECAP2 ECAP3 ECAP4 Mean Minimum Minimum of TOFs scored valid Maximum Maximum of TOFs scored valid From the above and from table 5 13 it may be concluded that the only suitable Luin criterium would be that if it is smaller than about 1 ms the ECAP is probably invalid Finally note that the mean Lmin values show the same decreasing trend as the mean Lmax
25. ORS Eindhoven University of Technology aN Faculty of Electrical Engineering Department of Measurement and Control u J Section Medical Electrical Engineering Validation of muscle relaxation measurements M H A van Steen Thesis for the degree of Master in Electrical Engineering Completed in the period May 1997 through March 1998 Project assigned by Prof dr ir P P J van den Bosch Supervisor Dr ir J A Blom In cooperation with Dr H H M Korsten Catharina Ziekenhuis Eindhoven De Faculteit Elektrotechniek van de Technische Universiteit Eindhoven aanvaardt geen aansprakelijkheid voor de inhoud van stage en afstudeerverslagen The Eindhoven University of Technology Department of Electrical Engineering does not accept any liability concerning the contents of traineeship reports and graduate reports Abstract The administration of neuromuscular blocking agents during surgery is directed to suppressing involuntary muscle movements in anaesthetised patients Muscle relaxants are conventionally administered by bolus injections This results in a failure to maintain steady relaxation levels Continuous infusion of muscle relaxants leads to a more stable level of muscle relaxation The work in this paper is aimed at the optimization of an existing measurement system and on validation of measurements of muscle relaxation in order to develop in a later stadium a closed loop feedback controller for muscle relaxation An improved ve
26. action of vecuronium Afer 30 minutes a new smaller bolus dose was given After 55 minutes the measurement was disturbed 4 2 Possible methods for validation From literature several possible strategies arise to meet the given demands 4 2 1 Petri nets A Petri net was used to validate arterial blood pressure signals Extrema significant points of the signal are determined and the slopes of the periods in between It is assumed that in valid signals these points and slopes always occur in a known order This order can be represented by a state diagram called Petri net Any transition of a measured signal that is not within this diagram is to be considered invalid Although this method works well for signals with a well defined shape it is not very useful for validation of muscle relaxation measurements Smans 1993 because the exact EMG waveform depends on too many factors and changes dramatically in function of the level of muscle relaxation 4 2 2 Map method Two different approaches are presented by de Graaf de Graaf 1993 The first is called the map method It checks whether a piece wise linear approximation of a measured waveform lies within the borders drawn around a piece wise linear approximation of an ideal valid reference measurement 30 First significant points are abstracted from the reference signal A simplified version is then generated by linear interpolation between these points Upper and unde
27. aesthetized infants and children Diss Helsinki University 1991 Kirkegaard Nielsen H Helbo Hansen H S Lindholm P et al Double burst monitoring during surgical degrees of neuromuscular blockade A comparison with train of four In INT J CLIN MONIT COMPUT 1995 12 4 p 191 196 Knuttgen D Burgwinkel W Zur Nieden K et al Limited applicability of the Datex Relaxograph in diabetics with peripheral polyneuropathy INT J CLIN MONIT COMPUT 1996 13 1 p 21 25 65 Mason DG Linkens DA Edwards ND Reilly CS Automated delivery of muscle relaxants using fuzzy logic control IEEE Engineering in Medicine and Biology 1994 13 678 686 Mason D G et al Self learning fuzzy control of atracurium induced neuromuscular block during surgery Medical amp Biological Engineering amp Computing 1997 35 p 498 503 Melissen Martin Een algemene methode van extractie en validatie van signaalparameters van biomedische signalen Eindhoven University of Technology 1993 Graduation report Nikkelen A L J M Pharmacokinetic and pharmacodynamic modeling of neuromuscular blocking agents for educational simulation Eindhoven Technische Universiteit Eindhoven 1995 Graduation report NMT 100 Technical Manual No 870378 2 1985 Datex Instrumentarium Corp P O box 357 00101 Helsinki 10 Finland Rowaan C J e a A complete and comprehensive computer controlled neuromuscular transmission measurement system developed fo
28. al is amplified by an amplifier with programmable gain For the DAS1402 board this gain can be set to 1 2 4 and 8 times By increasing the gain for small signals the 4096 steps of the 12 bit A D converter are used for a smaller input voltage range In this way the resolution can be improved The relationship between gain and resolution is given in table 2 2 The gain code is a number supplied to the driver software to set a given gain factor Table 2 2 The relationship between gain gain code input voltage range and resolution of the 12 bit A D converter F V to 9 995 Ta 5 V to 4 9976 V 2 5 V to 2 4988 V 1 25 V to 1 2494 V _ 0 61 mV 19 When using multiple channels these should be connected to successively numbered input channels The first and last channel in a scan can be set via software Since there is only one A D converter ADC present only one channel may be sampled at the same time To sample multiple channels a multiplexer connects them to the ADC one after another The scanning of the channels to sample can be done in two modes that can be set by software In paced mode the sampling of the channels is done at regular intervals After finishing a scan the next scan is started after such an interval In burst mode the ADC samples the channels one after another at very short intervals and then waits until the next scan should be performed In this mode the channels can be sampled at 10 us intervals
29. alse alarm or cause a wrong control action Most of the time the number of valid measurements considered invalid by the algorithm does not have to be very low The need for information of the control algorithm in terms of valid measurements per unit of time will be discussed below A prerequisite constraint to the algorithm is that the algorithm should be fast enough to validate the signals in real time on a PC Since there are 18 seconds between each two successive measurements the maximum time available for the validation is in the order of a few seconds per measurement 4 1 1 Maximum number of subsequent invalid measurements It depends on the control algorithm and on the phase of the relaxation onset steady state or recovery how many measurements may be unusable before the control performance gets in danger The limit to the maximum number of measurements that may successively be missing follows from the Nyquist criterium The sample frequency i e the number of valid TOF measurements per unit of time should be twice the highest frequency in the signal Since the response to the muscle relaxant drug is a strongly non linear process this frequency is different during the different phases of action During the onset of relaxation see figure 4 1 the level of muscle relaxation changes very rapidly After injection the patient s response normally stays at 100 for 1 5 to 3 minutes After that within 1 minute i e 3 TOF measurements
30. ans did not report any problems in our setup this led to loss of triggering It is suspected that the implementation of the NMT response signal in the version of the Relaxograph which Smans used differs from our version In any case from the Relaxograph electrical circuit schemes Datex NMT 100 technical manual monitor 1985 it was clear that a certain signal line in the stimulation circuitry would provide better trigger information The signal is used internally to open a noise gate placed before the input of the EMG amplifier This gate is opened 0 5 seconds before the first as well as 0 5 seconds after the last stimulus of a TOF to measure the noise or HF interference level It is also open from 3 to 18 ms after each stimulus The normal level gate closed is 0 7V and the active level gate open is 12 2 V After switching the Relaxograph on the level is 12 2 V When after that the stimulator is turned on the level changes to 0 7V 2 4 2 Analog EMG output The Relaxograph amplifies the EMG signal and filters it with a bandpass filter which was specified to have 60 Hz and 400 Hz cut off frequencies The EMG signal was sampled by the Labmaster A D board at a 50 kS s rate and low pass filtered with a digital moving average filter with a 540 Hz cut off frequency that will be described in paragraph 2 5 The resulting over all frequency transfer function is a band pass filter with 60 and 400 Hz cut off frequencies 2 4 3 Serial data link
31. ard deviation 0 2342 This ratio was expected to correlate with direct stimulation because in such signals the DC component was not OV However only the Vpoc Ver parameter of ECAP 1 could be used for validation since for values lt 1 5 and gt 1 5 a majority of the measurements was invalid No such boundary values could be found for the Voc Vrr parameter of the other ECAPs Table 5 10 More statistical data on Voc Ver value 0 also assigned when Vrr 0 Mean Minimum Minimum of TOFs scored valid Maximum Maximum of TOFs scored valid 5 3 Latency related parameters of single ECAPs Figure 5 12 Parameters related to the latency of extrema 42 Latencies of several peaks see figure 5 12 were calculated If a peak was not present the latency was attributed the value 2 ms to be able to distinguish this case in the histograms Of course the latencies of the minimum and maximum peaks were found in every ECAP 5 3 1 Latencies of maximum peaks Lmax 1 Lmax 2 800 800 600 600 400 400 200 7 200 oO EE o 4 o 0 005 0 01 0 015 0 02 o 0 005 0 01 0 015 0 02 Lmax 3 Lmax 4 800 800 600 600 400 400 200 ik 200 o ih pistes o R Kd o 0 005 0 01 0 015 0 02 o 0 005 0 01 0 015 0 02 Figure 5 13 Histograms of the latency in s after stimulation of the maximum peak of ECAPs 1 to 4 The maximum bin heights are 684 841 1071 and 1164 respectively The average value 6 4 ms standard deviation
32. ases Because of this correlation between the muscle relaxation and muscle fade the muscle fade is often used as a clinical measure for relaxation It must be noted though that the correlation is weak and depends on many factors When Ti Tre is low Ts Ti becomes unreliable because Ts is very small and noise like The two measures can not be used interchangeably and the most reliable measure is T1 Tref The NMT monitor presents T1 Tref if a calibrated measurement was done as well as T T on a display screen After the electrodes have been placed and the NMT cable has been connected but before the muscle relaxant drug is administered the clinician should have the NMT monitor execute an automatic calibration cycle In this cycle the monitor does the following 1 It sets the gain of the internal EMG amplifier 2 It applies a series of stimuli at 0 5 sec intervals with increasing current up to 70 mA until the EMG response does not increase any further i e all innervated muscle fibres are contracting By adding 15 to that current the supramaximal stimulus current is found that will be applied during the rest of the operation 3 A few seconds after that four supramaximal stimuli are applied at 1 sec intervals The average T value of the responses is calculated and used as Tre If this calibration fails the user can try to recalibrate or continue in uncalibrated mode In this mode only the muscle fade is displayed For
33. d bandpass filtered to eliminate 50 Hz noise Every volt in the histograms corresponds to 1 mV in the EMG signal The operations were done under several different anaesthetic conditions The muscle relaxant used most of the times was vecuronium During manual examination of the measurement files the following minor technical imperfections were found e the four control twitches were triggered 1 0 ms late compared to the subsequent twitches e in a few cases the gain had not been adjusted correctly so the tops of the ECAPs were chopped off e sometimes triggering was incorrect so that in these cases the first twitch was recorded as the second The distorted measurements were included in order to test the ability of the validation algorithm to discern technical errors The only indication of noise or HF disturbance recorded in the measurement files are a noise number and the HF disturbance and electrode off flags given by the Relaxograph But since it was not clear how this information was derived and since the electrodes off information was only updated after every 6th TOF so only once in every 2 minutes it is hard to use for validation purposes 33 5 2 Amplitude related parameters of single ECAPs First second third and fourth ECAPs of every train of four have been analyzed separately Mean value and standard deviation have been calculated over all four ECAPs In order to show more detail only the relevant parts of the histog
34. elaxograph horizontal idem according to PC 6 1 2 Accuracy of AS 3 NMT module and PC The final data acquisition system functioned stabily Measurements were collected shown on screen and stored on disk At the AS 3 output an offset voltage was observed Since this voltage varied slightly over time it had to be compensated for The average of some of the first samples of each twitch was subtracted from the twitch After applying this correction the readings of T Tre and Ts Ti on both PC and AS 3 were comparable Calibration of the AS 3 NMT module was successful in 17 of the 18 cases Compared to the measurements with the Relaxograph the sensitivity of the system to electric interference by diathermia was considerably smaller The amplitudes of artefacts were smaller This is probably due to better grounding and shielding of the A D board and connecting cables 6 2 Performance of the validation algorithm The validation criteria have been determined using a rather dirty set First the performance of the chosen parameters and criteria will be assessed This is done by comparing the outcomes of validation with the algorithm against the ideal outcome or golden standard This golden standard is set by an expert Since it is not common practice to monitor evoked train of four EMG signals such experts are hard to find It was supposed that in this case the validity could be judged by the author In the next paragraphs the man
35. erval Over all mean 5 0 mV standard deviation 66 5 mV Table 5 4 More statistical data on Voc values in Mean Minimum Minimum of TOFs scored valid Maximum Maximum of TOFs scored valid In the period from 4 to 15 ms after stimulation for normal signals this is a biphasic signal the DC value is expected to be circa 0 V The histograms of figure 5 5 show that this is true The histograms are broader larger deviation than those from figures 5 4 This is because in the 4 15ms interval the signal itself deviates stronger from Voc than it does over the complete 0 20ms interval 37 Vde3 1 Vde3 2 800 800 r ja 600 400 400 200 MIN 200 all Mrama 9 0 05 oO 0 0 9 5 1005 o 0 05 0 1 Vde3 3 Vdc3 4 800 aad 800 nm 600 600 400 400 200 200 EE EENES 0 1 0 05 oO 0 05 0 1 0 1 0 05 o 0 05 0 1 Figure 5 6 Histograms of the average voltage in the 15 20 ms interval Over all mean 17mV standard deviation 78 1 mV Vpcs was determined only in the learning set The signals in the test set do not contain information after 15 ms from stimulus Table 5 5 More statistical data on Voos values in Mean Minimum Minimum of TOFs scored valid Maximum Maximum of TOFs scored valid 5 2 3 Peak to peak voltages The peak to peak voltages of the several ECAPs were studied because they might be useful for validation in combination wi
36. eters was comparable to that of Ti These distributions were more or less gaussian Table 5 24 shows a summary of these other parameters and their statistical properties Table 5 23 Statistical data on the change of other parameters Mean Standard deviation Minimum Minimum of TOFs scored valid Maximum Maximum of TOFs scored valid Mean Standard deviation Minimum Minimum of TOFs scored valid Maximum Maximum of TOFs scored valid Mean Standard deviation Minimum Minimum of TOFs scored valid Maximum Maximum of TOFs scored valid Mean Standard deviation Minimum Minimum of TOFs scored valid Maximum Maximum of TOFs scored valid 0 791mV 35 9 mV 0 814 V 0 814 V 0 6765 V 0 4756 V 0 140mV 42 9 mV 0 532 V 0 532 V 2 629 V 0 5488 V 3 163 V 0 8724 V 54 5 6 Selection of parameters and bounds From the above it was clear that certain parameters e g T Vee depend mainly on the relaxation level and are not very sensitive to artefacts These parameters do not seem to be very suitable for validation The only clear bounds that may be posed on them lie far from the mean value and are based on the technical limits of the measurement system Some of these parameters were included however to be able to detect errors that did not occur in the learning set and may not be noted by other parameters An other group
37. gorithm 4 1 1 Maximum number of subsequent invalid measurements 4 2 Possible methods for validation 4 2 1 Petri nets 4 2 2 Map method 4 2 3 Linguistic method 4 2 4 Artificial neural networks 4 2 5 Heuristic method 5 Parameter analyses 5 1 The learning set 5 2 Amplitude related parameters of single ECAPs 5 2 1 T Integrated rectified value 5 2 2 Voc Average voltages 5 2 3 Peak to peak voltages 5 2 4 Ratio of maximum voltage to T 5 2 5 Ratio of minimum voltage to T 11 11 11 12 13 13 14 15 15 16 16 17 18 18 19 20 21 21 21 21 22 23 25 25 25 29 29 29 30 30 30 31 31 31 33 33 34 34 35 38 39 40 5 2 6 Ratio of peak peak voltage to T 5 2 7 Ratio of DC to peak to peak voltage 5 3 Latency related parameters of single ECAPs 5 3 1 Latencies of maximum peaks 5 3 2 Latencies of minimum peaks 5 3 3 Delay between minimum and maximum peaks 5 3 4 Latencies of zero crossings 5 3 5 Number of zero crossings 5 3 6 Irregularity parameter 5 4 Change of parameters within single TOFs 5 4 1 Change of T ina TOF 5 4 2 Change of Noin a TOF 5 4 3 Change of Cir in a TOF 5 4 4 Change of Vmax T ina TOF 5 5 Change of parameters in successive TOFs 5 5 1 Change of T in successive TOFs 5 5 2 Change of other parameters in successive TOFs 5 6 Selection of parameters and bounds 6 Results 6 1 Performance of the data acquisition systems 6 1 1 Accuracy of Relaxograph Labmaster system 6 1 2 Accuracy of AS
38. his unit is only used to convert a recorded RE_RelaxogrType record into a formatted string The Pump unit will implement the infusion pump protocol It also uses the RS232 unit to send and receive information from the computer controlled infusion pump via a second serial link This unit is still to be constructed Probably a unit previously developed for the blood pressure control system can be used The Control unit will calculate the amount of pharmacon to be infused based on the last measurement data It may also adapt the parameters of a pharmacodynamic pharmacokinetic patient model The output of that model can be used when a measurement is invalid Several decision rules should be implemented so that the Control unit can monitor its performance and take action if necesary This unit will also be a topic for further investigation The FilelO unit serves to read and write the analog and serial measurements from and to the hard disk When a previously recorded file should be read first a list of files is displayed of which the user may choose one Before starting measurements the unit asks for a filename If no filename is entered the measurements will not be stored on disk 26 The Screen unit provides a set of routines to display EMG measurements graphically and to show messages to the user It can display a message box an input box text boxes a menu box a large screen title and many sorts of graphs line poi
39. ication of the current position in the measurement file was also added As stated earlier the main idea was that it would be better to discard measurements of which we doubted the quality than to use a probably unreliable parameter as the control value In the latter case the system could give the impression of good control while the controlled value had little or nothing to do with the muscle relaxation itself Since we wanted a rather strict validation the following explicit rules of thumb have been used during the validation to make it more consequent e Measurements in which direct stimulation of the muscles was suspected should be discarded e The first twitch had to look well if not the measurement was considered invalid e If only the second third or fourth twitch did not seem reliable but the first one did the measurement was called valid e The quality of Ts was more important than that of T2 and Ts because the T T1 ratio indicates the clinically important muscle fade which is reliable only if Ts is reliable 6 2 3 Results of validation by eye The quantitative results of the validation by eye are presented in table 6 2 The number of invalid measurements was lower in the test set Because of the correct electrode placement no direct stimulation was observed in the test set 59 Direct stimulation was recognized due to its exponentially shaped unipolar twitches that also showed up at times the relaxation was sure to
40. ifferent ECAPs within a train of four and between one ECAP and the reference ECAP are explored In the fifth paragraph the time course of some parameters is discussed The final paragraph shows the relationship of some signal parameters and the T1 Tre ratio 5 1 The learning set The learning set consisted of circa 6878 train of four measurements containing 27512 EMG responses that were collected during 30 surgical operations in the Eindhoven Catharina Hospital for a previous work of Joost Smans Since the main goal of that work was to determine correct electrode placements the positions of the stimulating as well as of the measuring electrodes were different in each operation and sometimes the electrodes were moved during an operation For that reason the measurement files may contain more unusable TOFs than in normal clinical practice Moreover due to the different electrode positions the shape of the EMGs varies strongly It is expected that this variation will be smaller when the positions are chosen optimally but by training with this varied test set the algorithm will be prepared for non optimal electrode positions too A LabMaster data acquisition board was used for data acquisition Preprocessing consisted of a subsampling and interpolating filter that left 100 samples per twitch as described earlier The voltages presented here are the voltages on the Relaxograph s output On this output the EMG signal was amplified 1000 times an
41. ks Yet another method would be the use of artificial neural networks Since the validation can be seen as a classification problem neural networks might be helpful However this approach has the serious drawback that the decision of the network cannot be reduced to physiological knowledge the network cannot tell why a certain measurement was considered valid or invalid 4 2 5 Heuristic method The method that will be used in this work could be called a heuristic approach to validation It results from the observations that the shape of the TOF signals depends on the level of muscle relaxation that the signal shape varies between patients and that the signals are noise like in case of deep levels of relaxation First recorded TOF measurements are analyzed by eye in order to gain knowledge about the signals and artefacts Then from the learning set many parameters and their probability distributions are derived It is expected that the parameter value distributions are gaussian with outliers caused by artefacts Based on these distributions suitable criteria may be derived A measurement is considered valid only if all parameters satisfy the criteria The algorithm can be optimized by letting it judge the learning set of measurements If the results are satisfactory the final algorithm can be tested on a set of independent test data The algorithm will base its decision on clear criteria and will be able to tell why a measurement was c
42. l only receive the amount of drug that is needed for the desired effect By taking over routinely and time consuming tasks the anaesthetist may have more attention for the patient and be more alert to signs of complications Earlier a controller for blood pressure has been developed and implemented succesfully at E M E Zwart 1992 Now research is focusing on a controller for muscle relaxation Hoevenaren 1992 Scheepers 1992 Smans 1993 The general architecture of this system is shown in figure 1 1 Setpoint entered by clinician PC based control system Computer Data int controlled res Val A 5 acquisition alidation Control drug infusion pump Neuromuscular Transmission Monitor Patient Figure 1 1 Architecture of a control system for muscle relaxation 1 2 Control of muscle relaxation During operations most patients are given muscle relaxant drug in order to suppress unintended movements that might disturb the surgeon s work The relaxant makes all skeletal muscles Insensitive to nerve action potentials Since the ventilatory muscles are also paralyzed these patients are ventilated artificially The heart and the digestive muscles are not affected 11 In normal clinical practice the desired level of muscle relaxation is reached by injecting an initial dose of relaxant drug in a vein and maintained by smaller repeated injections This causes large f
43. le relaxation in to the clinician Test this measurement system on a number of patients and evaluate reliability and accuracy e Develop a method for the design of a validation algorithm implement such an algorithm and test it on a set of measured signals Literature on the above subjects should be studied to identify problems and their possible solutions 13 1 6 Contents of this report The development of a new data acquisition system is described in chapters 2 and 3 Chapter 2 covers the hardware and chapter 3 the software After that we will focus on validation methods Chapter 4 outlines the goals and possible methods for validation Every validation method makes use of a priori knowledge about the signal In chapter 5 analyses of TOF signals are described that should result in the needed knowledge Based on this knowledge criteria for valid signals are derived To test the data acquisition system and to acquire a set of TOF signals to test the validation algorithm a series of measurements have been carried out in the operating room Chapter 6 presents an evaluation of both the data acquisition system and the validation algorithm Finally chapter 7 lists conclusions and suggestions Points of attention for further research will also be presented 14 2 Hardware In this chapter we will describe the hardware used to measure the level of muscle relaxation The main questions are how can the level of muscle relaxation be
44. lt 02 V Voc gt 0 3 V lt 0 3 V Voe33 gt 0 3 V lt 0 3 V Voe3 4 gt 0 3 V lt 0 3 V Voc Vrr lt 1 5 gt 1 5 Vmax T gt 25s lt 300s VmMax T 2 gt 25s lt 300s Vmax T s gt 2557 lt 300s Vmax T s gt 255 lt 300s Vun T lt 36s Vmumn T z lt 36s Vmn T lt 36s Vun T lt 36s Ver T lt 500s Ver T 2 lt 500 s Ver I s lt 500s Ver T s lt 500s Cire gt 1 9 lt 7 Cirrz gt 1 9 lt 7 Cirr gt 1 9 lt 7 Cine gt 19 lt 7 No lt 6 No lt 7 No lt 7 Nox lt 7 69 Parameter NO NO NOs NO NOs NO NO NO T2 Ti Ts T2 T4 Ts Ti Th Cirr 2 Cire Cirr 3 Cirr 2 Cires Cir Each TOF compared to the last valid TOF ACir 1 ACirr z AT AT AT3 ATs AVpci 1 AVoc1 2 AV Dci AVDc1 4 AVpca 1 AVoc2 2 AVpcz3 AVoc2 4 AVpc3 1 AVpe3 2 AVpc3 3 AVpc3 4 A Vmax T 1 A Vmax T 2 A Vmax T 3 A Vmax T 4 AV p 1 AVpp 2 AVpr3 AVpp4 A Ver T A Ver T lt 1 8 lt 3 lt 2 lt 5 lt 5 0 018 Vs lt 0 015 Vs 0 02 Vs lt 0 01 Vs 0 02 Vs lt 0 02 Vs 0 01 Vs lt 0 01 Vs 0 4 V lt 04 V 0 4 V lt 0 4 V 0 4 V 0 4 V 0 4 V 0 4 V 0 4 V 0 3 V 0 3 V 0 3 V 0 3 V 0 3 V 200 s 200 s 200 s 200 s 5 V 4 V 4 V 4 V 300 s 300 s 200 s 300 s VAA MV VMV VVV VMV VMV VVV MVAAANA AV VMV VVV V Vv Vv VA VAA VV VY 70
45. luctuations and in most cases an initial overshoot in the level of relaxation An automatic control system might overcome these problems From literature it is known that existing control systems for muscle relaxation may show good performance in terms of deviation from the target level but often show problems concerning the measurement system Olkkola 1996 As a solution some focus on robust control algorithms while others even used two measurement systems in parallel to increase the reliability Mason 1997 No reports on attempts to automatically validate the muscle relaxation measurements have been found in literature It is also unclear how the control systems react in case of heavily disturbed measurements and if safe behaviour can also be guaranteed in these situations The major causes of problems in case of NMT monitoring by EMG are e incorrect positioning of the stimulating and recording electrodes e unintended direct stimulation of the muscle via the skin surface instead of via the nerve e electrical activity in parts of the muscle that move but don t contract e diathermia use of an electric knife e movements e electrode cables getting loose The influence of these artefacts on the final controller performance may be reduced in several ways for example by preventing their occurence by automatic checking of the signal quality validation and by designing a control algorithm that is robust to noise at its sensor in
46. measured and how can the data be made available for processing with a PC The first paragraph tries to answer the first question while the rest of the chapter is devoted to the second In the second paragraph the reasons for developing two versions of the data acquisition system will first be pointed out After that the A D conversion board that is common to both versions will be described Finally some details of both links will be presented 2 1 Measurement of muscle relaxation First of all it should be noted that this paragraph is only meant as a short introduction to the method of muscle relaxation measurement used in our system Hoevenaren Hoevenaren 1992 has investigated the several methods of measurement and motivated the choice for this method Smans Smans 1993 further optimized the method For the physiological background of neuromuscular block the reader may refer to Feldman 1996 The level of muscle relaxation can be measured by a neuromuscular transmission NMT monitor This monitor applies a pattern of electrical stimuli to a nerve via surface electrodes Depending on the level of muscle relaxation more or less muscle fibres of the muscles that are connected to the nerve will contract in response to stimuli This contraction is then measured by force movement acceleration EMG or other sensors We chose to use EMG sensors When placed on the skin near the belly of the muscle these surface electrodes pick up the superimp
47. ns control to the calling program The board generates an interrupt after each A D conversion The called interrupt routine should transfer the sample from the board into memory e In DMA mode the board acquires a single sample or multiple samples from one or more analog input channels The device driver initiates the conversion and then returns control to the calling program The board writes data directly to memory using the PC s DMA controller DMA mode is faster than interrupt mode because the actual data transfer is not controlled by the CPU Processes on the CPU can continue 2 4 Interfacing to the Relaxograph The interfacing of the Relaxograph to a PC has been described extensively by Smans The Relaxograph has two outputs one analog output for the EMG and triggering signals and one RS 232 serial data output The digital output exports the twitch heights calculated by the Relaxograph and some status information to the PC The status information concerns several internal alarms electrode off and HF disturbance With respect to Smans a few changes were made to the link 20 2 4 1 Relaxograph trigger signal Before the signal that is called NMT response pin 4 was used as a trigger signal It was discovered that this signal does change on the moment of stimulation but its amplitude is proportional to the EMG response This means that it is almost zero when the patient s muscles are relaxed completely Although Sm
48. nt bar with or without axes in a flexible and user friendly way The graphs can be defined using an SC_GraphType record The Timer unit contains time handling functions It can return the current time and uses the PC s timer interrupt for a time out routine This routine is used to monitor the progress of measurements Finally the General unit contains several general purpose functions especially string formatting functions that can be used by all other units 27 4 Validation methods for TOF signals In this chapter a method to develop a statistical validation algorithm for TOF signals will be proposed First we will define the demands to a validation algorithm Then several possible methods for validation will be described and one will be chosen The chapter is concluded with a more detailed description of that method 4 1 Demands to a validation algorithm There are several criteria to be met for the algorithm to be useful in clinical practice de Graaf 1993 First the algorithm should recognize all measurements that an expert for example an anaesthesist would consider invalid Second the number of measurements that is considered invalid by the algorithm while being considered valid by an expert should not be too high The main property of a good validation method is that the number of invalid measurements that is considered valid by the algorithm is minimal This is important because every such measurement may trigger a f
49. o these fuzzy rules could not be implemented in the algorithm e The selection of the parameters may not have been optimal Including or excluding a parameter from the algorithm may change the number of correctly recognized artefacts but also changes the number of incorrectly invalidated measurements A good balance should be found when choosing parameters 61 e Even if a sufficient set of parameters was used the choice of the criteria for these parameters influences the final validation performance This is due to the smooth transition from the range of valid parameter values to the range of invalid values that occurs for most parameters Of course for all these causes remedies may be sought 62 7 Conclusions and recommendations 7 1 Conclusions 7 1 1 Data acquisition system A PC based data acquisition system for the acquiring of train of four EMG measurements has been constructed that is suitable for linking with both the Datex Relaxograph NMT 100 and with the Datex AS 3 anaesthesia depth unit ADU During clinical trials with the ADU the total system performance was satisfactory Because of the use of an optimal electrode positioning and because of better shielding and grounding a reduction of the number of measurements with artefacts compared to a previous study was reduced from 8 5 to 3 0 7 1 2 Validation algorithm A heuristic method for finding a validation algorithm for train of four m
50. onsidered invalid Depending on the reason for invalidation the measurement might simply be rejected the clinician might be advised to correct the cause of the artefact e g in case of direct stimulation of muscles or in some cases the artefact could perhaps be corrected for The parameter set may include continuous signal properties like amplitude and duration as well as discrete properties like the number of zero crossings Boolean parameters that are the result of more complex algorithms may also be used 31 The following groups of parameters are proposed I I parameters based on the shape of a single twitch for example the amplitudes and latencies of peaks in the signal parameters based on the speed of variation of the shape since the muscle relaxation and probably also other parameters do not vary rapidly especially in the steady state phase the rate of change of these parameters should lie within narrow bounds The rate of change can be considered for parameters A within one TOF for example ratios of peak to peak values B between subsequent TOFs In this way we hope to find and make use of constancies and or reproducible features in the signal These constancies and reproducibilities constitute the knowledge about correct TOF signals Thus the Heuristic approach to validation used in this work may be summarized as follows 1 Inspect a learning set and a test set of measurements by eye In
51. onvertor 100 S s CPU memory PC with A D board low pass A D convertor 3 kS s weighted average a subsampling filter Fs out Fs in 10 Figure 2 4 Signal processing chain using AS 3 monitor and PC In this way the sample frequency of the EMG signal at the input of the A D conversion board becomes 3000 x 25 75 kHz This 30 times oversampling was used to avoid distortion of the signal due to timing errors related to the step wise changes in the AS 3 output signal The A D board could not be synchronized with the ADU s D A convertor 22 Finally the signal is low pass filtered and subsampled by the acquisition program on the PC The transfer function of the moving average filter is depicted in figure 2 4 This filter is the same as the one used by Smans Smans 1993 It has a cut off frequency 3dB of 0 0108 sample frequency which results in 810 5 Hz Since the bandwidth of the signal is now only about 2 of the Nyquist frequency samples may be left out without loss of information This is done by the subsampling filter that outputs 1 out of each 10 successive input samples The filtered version is stored on disk The sample frequency of the stored signals is 7500 Hz The implementation of both filters was combined by calculating the response of the moving average filter only for the samples that are output by the subsampling filter 20 30 40 J 50 20 log IHI dB 60 80 90 fi 4 1 L 1 L 1 1
52. or the procedure they are used in For example SC_MsgBox returns MSG_OK if the user selected OK in response to the messagebox 3 2 Survey of the units Figure 3 1 shows the hierarchical organisation of the units of the control system Test programs have been written to show the capabilities and the way to invoke their functions Two versions of the program have been developed The Relaxograph version interfaces to the Relaxograph NMT 100 while the AS 3 version interfaces to the AS 3 ADU The names of the AS 3 version units are preceded by AS3_ A short description of the functionality of each unit now follows The raw EMG available on the NMT s analog output is first sampled by the AD Routines unit This unit returns raw twitch data of type AD_TwitchType It takes care of the communication with the A D board via a driver library The triggering of the measurements is done by software in this unit The main purpose of the EMG_Processing unit is to process the raw signal The signal is filtered and sub sampled using a 41 point moving average and subsampling filter as described in paragraph 2 5 1 Several simple parameters of each twitch rectified integrated EMG maximum minimum and of each TOF T T1 Ti Tre are calculated The train of four data can be exported as EP_TOFType records and as a formatted string that can be put on screen directly The serial data calculated by the Relaxograph are received over a serial communications link
53. osed Nervus ulnaris Abductor digiti minimi stimulus stimulus Adductor pollicis b Figure 2 2 a position of nervus ulnaris and abductor digiti minimi b optimal electrode placement Smans 1996 a electrical activity of a number of contracting muscle fibres The measured EMG waveform is often referred to as an evoked compound action potential ECAP The pattern of stimuli we used is the so called train of four TOF stimulation This means that four stimuli each lasting 100 us are applied at 0 5 second intervals This pattern is repeated every 20 seconds So every 20 seconds the NMT monitor carries out one measurement The ulnar nerve in the forearm and the abductor digiti minimi a muscle on the little finger see figure 2 1a form a convenient nerve muscle combination When this combination is used electrodes should be placed according to figure 2 1b 2 1 1 Train of four response The EMG response to TOF stimulation see figure 2 2 consists of four twitches At the moment of stimulus a stimulus artefact is seen This is caused by conduction over the skin not by muscle contraction Since the internal amplifier is gated only after 3 ms the stimulus artefact is not present in the output signal After 3 ms a more or less biphasic potential can be seen that lasts for circa 25 ms It is caused by the depolarization front in the muscle tissue that moves under the electrode Metingen in de geneeskunde I Note
54. osing a board with a high resolution A D converter more than 8 bits The performance of two selected boards was very similar and sufficient for our purpose Because of practical reasons the Keithley Metrabyte DAS 1402 board was selected eventually With the board comes a driver library that can be linked with Pascal C and Basic programs 2 3 2 Characteristics of the DAS 1402 board Now we will briefly discuss how the board acquires converts and stores samples of the analog input signals The four digital inputs and four digital outputs of the DAS 1402 board are not covered here The analog signals to be measured are connected to one or more of the 16 physical input channels on the board s I O connector The inputs may be operated in differential mode or in single ended mode In differential mode the difference between two physical inputs is measured and mapped to a logical channel In single ended mode the voltage between an input and ground is measured In differential mode there are 8 analog input channels available while in single ended mode there are 16 channels The mode is set by a dip switch on the A D board The inputs may be configured for unipolar or bipolar voltages with an other dip switch Unipolar voltages should always be equal to or greater than OV while bipolar voltages may have positive and negative values We selected bipolar voltage mode because the Relaxograph s output range is 10 10V The incoming analog sign
55. ould be made 1 The existing software did function correctly but its structure could be improved in order to be able to include the validation and control system parts 12 2 Nowadays the Relaxograph has become part of an integrated anaesthesia depth unit ADU which is able to monitor the most important physiological patient data and also contains a ventilation and anaesthetic vapor unit A number of operating rooms in the Eindhoven Catharina Hospital has been equipped with these AS 3 monitors of Datex Engstr m Finland Because of their greater flexibility ease of use and interfacing possibilities and because staff had become familiar with this equipment it would be desireable to interface to these monitors 3 Hoevenaren and Smans both reported serious problems with the Labmaster board Interrupts did not function there was no high level driver software available documentation contained errors and cabling was sensitive to EM interference Although eventually work arounds for these problems were found it was doubted if such hardware was reliable and safe enough for our goal Moreover better hardware had become available in the mean time These three reasons lead us to the reconstruction of the data acquisition system hardware and software It will be discussed in chapters 2 and 3 1 4 Validation of the measurements As pointed out earlier the task of a validation algorithm will be to check if a given measurement is disturbed
56. posed on the signal it can be shown that Cir is equal to 4 a Nie A where a is the peak peak amplitude of the fluctuation Nee is the number of local extrema and A is the peak peak amplitude of the signal This means that for this type of signals Cirr is proportional to the number of turns but multiplied by their significance That is why a number of small turns will not alter the irregularity of the signal Cirr 1 2000 1500 1000 500 Cirr 3 2000 1500 1000 500 i Ol oO 5 10 Cirr 2 2000 1500 1000 500 10 Cirr 4 2000 1500 1000 500 10 Figure 5 21 Histograms of the irregularity parameter Cor for all four ECAPs The mean value is 2 643 with a standard deviation of 0 789 Some signals were not measured correctly and had a Ver equal to zero These were attributed a Cin value of 2 since lim C 2 Figure 5 21 shows the histograms of the resulting G of all four Vop LO PP ECAPs Table 5 17 More statistical data on Cim A value of 1 means that Vrp for the given measurement was equal to 0 ECAP1 ECAP2 ECAP3 ECAP4 Mean Minimum Minimum of TOFs scored valid Maximum Maximum of TOFs scored valid As could be expected from the visual inspection Cir increases in every ECAP of a TOF because the second third and fourth ECAPs decrease in amplitude and become more like a noisy signal 5 4 Change of parameters within single TOFs
57. put At E M E work has been done on the first possibility E g an optimal electrode positioning for reliable monitoring was determined Smans 1996 This may prevent failing calibration procedures and direct stimulation Careful shielding and grounding of cables and equipment may reduce the influence of diathermia Signal processing especially low pass filtering may also reduce spikes caused by diathermia Loose electrode cables are signalized by the NMT monitor itself but do lead to incorrect measurement values But since it is still possible that measurements are disturbed each measurement should be validated before use to make sure that the information supplied to the controller is only correct or missing but not incorrect A method to construct a validation algorithm should be developed This will be discussed in paragraph 1 4 Finally although some recommendations about the control system will be included in this report its actual design is beyond the scope of this work 1 3 Data acquisition In previous work at E M E a data acquisition system for the measurement of muscle relaxation has been set up Hoevenaren 1992 Smans 1993 A Relaxograph type NMT 100 produced by Datex Datex Relaxograph User s Manual was used as a measuring device of which the analog output was connected to a Labmaster data acquisition PC board The PC software was written in Borland Pascal 7 0 A closer look showed that some improvements could and or sh
58. r a trigger signal The output voltage range of the Relaxograph and the ADU are 10 10 V and 5 5 V respectively So the input voltage range of the board should at least be 10 10 V Table 2 1 Demands to a data acquisition board and the performance of two existing comparable boards Feature Required Keithley Metrabyte DAS 1402 and Advantech PCL818L Number of analog input channels 2 differential 8 differential differential single ended mode 2 single ended 16 single ended Number of digital input channels 0 4 Number of digital output channels 0 4 Maximum input voltage range 10 to 10V 10 to 10 V Programmable gain yes yes 1 2 4 8 times Sample frequency gt 600 Hz up to 100 kS s Device driver software supports DOS Pascal DOS Windows C Pascal Visual Basic Data transfer mode DMA DMA I O interrupt Resolution bits gt 8 12 Advantech board has 16 digital inputs and 16 digital outputs 18 Since the EMG signal contains almost no frequencies above 150 Hz the sample frequency of the A D board should be higher than 300 Hz As a margin of safety the required sample frequency should be at least 600 Hz The EMG signal should be digitized with a good accuracy Since the amplitude of the EMG signals may decrease a factor 100 or more quantization errors should be kept to a minimum This can be done by increasing the A D board gain for small signals and by cho
59. r clinical research on muscle relaxants In J CLIN MONIT 1993 9 1 p 38 44 Scheepers F N L H Ontwikkeling van een regelaar voor spierrelaxatie met behulp van een expert systeem regelaspecten afstudeerverslag TUE 1992 Schippers Houkje C Pharmacodynamics of vecuronium bromide in anaesthetized neonates infants and children Proefschrift Erasmus Universiteit Rotterdam Rotterdam 1988 Smans J L A Betrouwbaar meten van spierverslapping afstudeerverslag TUE 1993 Smans J L A Korsten H H M en Blom J A Optimal surface electrode positioning for reliable train of four muscle relaxation monitoring in International Journal of Clinical Monitoring and Computing 13 9 20 1996 Kluwer Academic Publishers Van den Brom R H G Monitoring of neuromuscular transmission with special emphasis on the assesment of intubating conditions Groningen Rijksuniversiteit Groningen 1994 Dissertation Young Shuenn Tsong en Kuang Ning Hsiao A pharmacokinetic model to study administration of intravenous anaesthetic agents in IEEE Engineering in Medicine and Biology 1994 p 263 268 Zwart R M P 66 Implementatie en evaluatie van een robuuste adaptieve bloeddrukregelaar Graduation reporst TUE 1990 Appendix A Wiring of PC NMT monitor links Link from Datex AS 3 ADU to PC with Keithley DAS 1402 A D conversion board Datex AS 3 Monitor UPI board Connector 44 pins D male Kanaal 0 20 brown PC A D board Connec
60. r limit borders are calculated that run in parallel to the curve at a given perpendicular distance Of every measured signal the simplified representation is calculated and it is checked to the borders If the representation lies completely within the borders it is considered valid else it is called invalid In case of muscle relaxation measurements the ideal reference measurement should probably be chosen as the last measurement A problem with the application of this method to muscle relaxation signals is that rapid signal changes may be valid 4 2 3 Linguistic method The second method that de Graaf proposes is a linguistic method The measured signal is again simplified into a piece wise linear approximation Each line segment is then characterized by its slope and length Each combination of slope and length is given a letter code so each line segment is assigned a letter Placed one after another these letters form words for each wave A list dictionary can be made of all valid words If the word belonging to a given signal is not in the dictionary that signal is considered invalid However it turned out that rather similar signals could yield different words In our case this would probably yield large problems with the small noise like signals in case of deep levels of muscle relaxation Moreover the words belonging to these signals are likely to be very different in length 4 2 4 Artificial neural networ
61. rage 2 49 crossings standard deviation 1 47 40 100 T T p T gopr 80 60 50 40 Number of invalid TOFs of total TT 1 TT 30 20 10 12 No 1 Figure 5 20 Number of invalid TOFs as a percentage of all TOFs as a function of the number of zero crossings in the first ECAP of the measured TOFS 5 3 6 Irregularity parameter In 1995 Zalewska and Hausmanowa Petrusewicz proposed a new measure to quantify the irregularity of single motor unit action potentials Zalewska 1995 We will apply it to the evoked compound action potential It is defined as follows nl If Ver gt 0 then Cir TENS Yaa Vop ia else Cir 2 where Ver is the peak peak amplitude yi is the sample with index i and ns is the number of samples The measure actually depends on the length of the ECAP curve This length increases 48 as the ECAP becomes more irregular Chr is independent of the amplitude and duration of the peak By definition it has the following properties 1 For a normal biphasic action potential with positive and negative peaks a and b a gt 0 b gt 0 Cor is equal to Cin 2a 2b a b 2 2 For a multiphasic signal with mp positive and nn negative peaks all with the same absolute amplitude a Cir is equal to Cir 2npa 2nna 2a np m n which is the number of phases 3 For a signal with small amplitude fluctuation local extrema super
62. rams are depictured In each case more than 95 of the parameter values are represented in every histogram Figure 5 1 Amplitude related parameters T Voc Voci Voc2 Vics Vmax Vmin and Vpr 5 2 1 T Integrated rectified value The T parameter was computed using the following formula 0 015 0 018 N 0 020 ien lo N n 0 003 N 0 020 T So each ECAP was rectified and summated from 3 to 18 ms grey area in figure 5 1 The sum is made independent on the duration 15 ms and the number of samples This time period was taken because before 3 ms a stimulus artefact may be present and after 18 ms the signal is more or less random The theoretical maximum value of T is 200 mVs 10V 20 ms No T values above 80 5 mVs were found The high maximum T values for ECAP 3 were due to artefacts Since these TOFs did have a well behaved ECAP 1 they were not scored valid during the visual inspection The larger T values belonging to the unrelaxed state in the beginnings of the operations are not visible since their values range from 10 to 81 mVs Since the distribution of this parameter is not a gaussian one no standard deviation has been calculated We must conclude that this parameter depends on the relaxation level and is only useful for validation with very wide bounds The value T 0 Vs is not expected because there is always some background noise present 34 Table 5 1 Mean and maximum of T values all value
63. rsion of the data acquisition part of the measurement system was developed A new digital to analog conversion board was adapted to interfacing to an integrated anaesthesia monitor was established and software was developed to collect present and store the muscle relaxation measurements The measurement method used in this work is the train of four TOF method with EMG sensors The purpose of the validation algorithm is to detect measurements that are disturbed by artefacts If the quality of a measurement is doubted the algorithm should consider it invalid The final goal is first to discard all measurements that contain artefacts and secondly to avoid the discarding of valid measurements Since there is very little knowledge about the correct shape of the signals knowledge was acquired by analyzing many parameters of the EMG signals The heuristic approach to validation used in this work may be summarized as follows 1 A learning set and a test set of measurements were inspected by eye In this way a golden standard was determined for the validation algorithm and insight in signal properties and artefacts was gained 2 A large number of parameters was chosen that are based on a single ECAP evoked compound action potential on the rate of change between the ECAPs of one TOF or on the rate of change between TOFs 3 The parameters were calculated for every measurement in the learning set The results were presented
64. s of all four ECAPs Average value 127 s standard deviation 40 7 The histograms show that the parameter usually lies between 50 and 200 The ratio is a little influenced by the level of muscle relaxation Larger values of the parameter belong to larger T 39 values because of sharper peaks while smaller values belong to smaller T values broader more noise like signals This parameter can be used for validation because the maximum value is clearly limited see table 5 7 Table 5 7 More statistical data on Vuax T values in s value 0 also assigned when T 0 po CAP 1 ECAP2 ECAP3 ECAP 4 Mean 134 8167 126 7744 123 0193 122 2937 Minimum 0 7 3828 15 4343 0 Minimum of TOFs scored valid 0 0 15 4343 0 Maximum 1206 9 507 0 366 8 1250 0 Maximum of TOFs scored valid 291 2558 338 9831 304 0425 304 7356 5 2 5 Ratio of minimum voltage to T The rationale behind this parameter is the same as for Vmax t It was expected to be more or less equivalent to minus Vmax t The histograms show that the range of values is comparable to that of Vmax t but their shape is a little different a dn Tdh 2 En S ee T Figure 5 9 Histograms of the ratio Vum T of all four ECAPs Average value 98 9 s 1 standard deviation 69 7 s Table 5 8 More statistical data on Vum T values in s value 0 also assigned when T 0 PoC ECAP ECAP2 ECAP3 ECAP4 Mean 108
65. s delta T 1 sea 800f 600 400 200 ol en 0 5 o 0 5 1 x 10 delta T 3 1000 800 600 400 200 delta T 2 goo 1 600 400 200 1 ntl 0 5 o 0 5 1 x 10 delta T 4 600 400 200 oO 1 0 5 o Figure 5 26 Histograms of the change in T between two ECAPs of successive TOFs Over all mean 4 762 10 Vs standard deviation 1 5 10 Vs reflected by a few large negative parameter values The high maximum of AT table 5 23 was caused by an artefact Since it occurred in ECAP 3 the concerning TOF had not been scored valid Table 5 22 More statistical data on A Tn All values are in Vs Mean 0 4743 107 Minimum 0 0317 Minimum of TOFs scored valid 0 0232 Maximum 0 0217 Maximum of TOFs scored valid 0 0217 53 0 4806 10 0 3445 10 0 6053 10 0 0314 0 0311 0 0311 0 0211 0 0211 0 0212 0 0284 0 0750 0 0171 0 0143 0 0750 0 0138 90F 80F 70F l sop lt 7 40 20b i 0 01 k L 0 04 0 03 0 02 Number of TOFs considered invalid by eye of total Delta T1 Figure 5 27 Number of TOFs scored invalid by eye as a percentage of all TOFs as a function of the change of T over successive TOFs 5 5 2 Change of other parameters in successive TOFs The distribution of the change in other param
66. s in 10 Vs Average value Max Max of TOFs scored valid 800 600 400 200 0 03 0 04 o 0 01 0 02 0 03 0 04 Figure 5 2 Histograms of integrated rectified values in Vs of all four ECAPs Average value 2 6 mVs standard deviation 4 4 mVs Count per bin limited to 800 5 2 2 Voc Average voltages The reason for looking at DC components was that they might be good indicators of direct stimulation It is supposed that direct stimulation causes unipolar exponential waveforms with a negative DC component Moreover technical failures may lead to higher DC levels The DC component was determined over the complete ECAP and over three different parts of the ECAP see figure 5 1 The three parts correspond to the stimulus artefact biphasic action potential and afterwave time windows The average DC voltage over a whole ECAP from 0 to 20 ms is circa zero as can be seen in the histograms The histograms do show a peak at 20 mV but apart from that Voc seems to be distributed normally with some unproper values at the extremes at 0 08 V and at 0 08 V Examination of the concerning TOFs showed a constant DC offset voltage that was probably due to the amplifier or Labmaster A D board Only a fraction of these TOFs was disturbed by direct stimulation 35 Vdc 1 Vdc 2 800 600 400 200 9 0 1 0 05 i o 0 05 0 1 Vde 3 Vdc 4 800 r
67. s the algorithm agreed with the inspection by eye The same criteria were applied to the test set as to the learning set as much as possible The voltage related criteria were changed to fit the 5 5V voltage range of the AS 3 ADU The calculation of the parameters for the test set was adapted to the other time window and sample frequency The results were again compared to the manual validation The performance of the algorithm on the test set is shown in table 6 4 60 6 3 Discussion Because of small differences between the test set and the learning set the algorithm considers more measurements invalid in the test set than in the learning set This appeared to be mainly due to the Non parameter number of zero crossings of ECAP n in a TOF The measurements of the learning set contained a small offset voltage that was enough to shift low signals completely above the OV level In the test set this offset voltage was compensated for so the number of zero crossings in small signals was higher Table 6 4 Performance of the validation algorithm on the test set compared to the inspection by eye In the columns the results of the validation by the algorithm are shown while the rows show the outcome of the validation by eye Invalid Total The algorithm detected 85 0 of all artefacts e The algorithm considered 28 1 of the measurements valid e In 30 2 of the cases the algorithm decided in accordance with
68. s to be suitable for validation Table 5 19 More statistical data on NO NOn t Mean Minimum Minimum of TOFs scored valid Maximum Maximum of TOFs scored valid No3 NO02 Ar a D A o o 10 8 6 4 2 o 2 NO4 NO03 Figure 5 23 Histograms of the difference in zero crossing counts between two ECAPs of the same TOF Mean 0 0277 standard deviation 1 223 5 4 3 Change of Cir ina TOF The histograms show that Cir more often increases than decreases during a TOF This could already be seen in figure 5 21 Table 5 20 More statistical data on Cirn Cirnn t CeCe Caries Cie Ciera Mean Minimum Minimum of TOFs scored valid Maximum Maximum of TOFs scored valid 51 Cirr2 Cirr1 Figure 5 24 Histograms of the difference in the irregularity parameter between two ECAP of the same TOF Mean 0 0942 standard deviation 0 608 5 4 4 Change of Vmax T in a TOF VmawT 2 Vmax T 1 T T T ERP nn 20 o 20 40 60 80 100 9o00 80 60 40 Vmax T 3 Vmax T 2 600 T T T eee AT T 400 200 nT TT LE Poo 80 60 40 20 oO 20 40 60 80 100 Vmax T 4 Vmax T 3 600 T E T T T prk gi 400 g Mil Moi ida en HN EERI Pos 80 60 40 20 o Figure 5 25 Histograms of the difference in Vuax T between two ECAPs of the same TOF Mean 4 174 standard deviation 31 21
69. sor signal is however not available at the output so it is not suitable for our purpose e the ADU stores all collected physiological data also the NMT data The ADU can show these trends on a display or print them e almost all measured physiological signals ECG blood pressure oxygen saturation ventilatory flows pressures and concentrations administered anaesthetic vapors etc are available on the digital or analog outputs So if the muscle relaxation control algorithm might also need other data it can use the same physical link Disadvantages are that the digital serial interface is more complex than the Relaxograph s serial link and that the ADUs are in permanent use in the operating rooms so the time to test the PC ADU interface is limited A Relaxograph that was no longer in use could be borrowed from the hospital so it was decided to make interfaces to the Relaxograph as well as to the new monitors In case the latter interface would not function well enough the former could be used as a back up 2 3 A D conversion board 2 3 1 Selection of a data acquisition board To digitize the analog EMG signals an A D conversion board is used Because of the problems previously experienced with the Labmaster A D board Smans 1993 a new board was selected The main demands to a suitable board together with two alternatives to the Labmaster are presented in table 2 1 We need one input channel for the EMG signal and one fo
70. testing purposes a Datex EMG train of four simulator property of the Catharina hospital was used It is connected to the NMT electrodes and delivers square pulses of circa 11 ms in response to a stimulus The output level as well as the muscle fade can be adjusted By using this simulator the developer does not need to be connected to the NMT monitor himself 2 2 Interfacing to Relaxograph and to AS 3 ADU Since the raw EMG signal is needed for validation purposes a link between PC and NMT monitor should be set up that makes this data available in a digital format For synchronisation purposes a trigger signal is required to note when the monitor stimulates the patient Previously a Datex Relaxograph NMT 100 was used as a measuring device As pointed out in paragraph 1 3 new anaesthetic depth units ADUs with integrated NMT monitors had been purchased by the Catharina Hospital The NMT monitor module of the ADU has several advantages over the older Relaxograph e the user interface is much easier for the NMT module there is only one button to start the calibration cycle and one button to start stop the NMT monitoring 17 e not only train of four TOF but also double burst stimulation DBS post tetanic count PTC and single twitch stimulation modes are supported and the stimulus duration can be configured to be 100 200 or 300 us e as a sensor either EMG sensors or accelerographic sensors may be used The accelerographic sen
71. th other parameters Peak peak voltages larger than 20 V could not be measured The peak peak voltage strongly correlates with the relaxation level As can be seen in table 5 6 TOFs with a large range of peak peak voltages have been considered valid during manual validation so this parameter does not seem to be of use for validation Table 5 6 More statistical data on Vrp values in Mean Minimum Minimum of TOFs scored valid Maximum 10 0098 Maximum of TOFs scored valid 10 0098 38 Vpp 1 Vpp 2 1000 800 600 400 200 1 5 10 Vpp 3 Vpp 4 1000 1000 800 800 600 600 400 400 200 200 5 A0 SD 5 10 Figure 5 7 Histograms of the peak to peak voltage in V of ECAPs 1 to 4 Average value 0 674 V standard deviation 1 258 mV 5 2 4 Ratio of maximum voltage to T The maximum voltage of an ECAP divided by the area under its curve Vmax T is a measure for how narrow and peak like the ECAP is For very steep and narrow ECAPs this parameter will be large while broad and flat ECAPs yield small Vmaxr values The T value was calculated as discussed earlier When the T value was very small the value 0 was assigned the parameter VmaxT 1 VmaxT 2 300 300 200 200 pra T Ta ahd kee 100 150 100 150 200 VmaxT 3 VmaxT 4 300 300 200 200 100 100 or el o Ad o 50 100 150 200 o 50 100 150 200 Figure 5 8 Histograms of the ratio Vuax T in
72. tharina Hospital medical instrumentation service For this purpose two passwords need to be entered in the monitor setup menu 21 2 5 1 AS 3 ADU data acquisition chain for NMT signals It was found that the signals are stepped due to quantization errors and because the monitor s D A converter is not followed by a low pass filter More important the D A conversion takes place at another rate than the A D conversion This means that the signals at the output are scaled in time These signal properties may however be overcome if the PC samples the signal at an adjusted rate and uses a digital low pass filter to round of the stepped signal The internal signal processing chain from EMG electrode to the UPI board output together with the adapted PC acquisition system are shown in figure 2 3 In the AS 3 ADU the EMG signal from 3 to 18 ms after the stimulus is amplified band pass filterered from 60 to 400 Hz and converted at a 2 5 kS s rate for internal storage This signal is then converted back to an analog signal at a 100 S s rate 25 times slower that is sampled by the Keithley A D board at a 3000 S s rate It was noted that after each 375 ms response the sample amp hold circuit of the monitor s D A convertor kept the output fixed at the last encountered voltage Datex AS 3 ADU Stimulus NMT trigger x 15 ms A D converter 2 5 kS lt gt Band pass filter s O 60 Hz 400 Hz 375 ms t Q D A c
73. that the actual movement is a much slower process that lasts hundreds of ms 1 25 t 4 se cra ze Ve eee ee BEE mn Sm cease lake deel CD U AS RS so HT SOF Ad Id IL 0 A A TN OU A 20 msec gain relaxograph 4 Figure 2 3 EMG response to train of four stimulation measured at the output of the NMT in V 2 1 2 Signal processing by the NMT monitor Since the stimulus artefact and the small slow afterwave are irrelevant for this purpose the NMT monitor uses a time window from 3 to 18 ms after each stimulus The signal is amplified circa 1000 times band pass filtered from 60 to 400 Hz rectified and integrated The final integrated voltage is proportional to the surface under the curve between 3 and 18 ms It is referred to as Tn where n 1 2 3 4 for the different twitches in a TOF Tre is the T value in the normal unrelaxed state Two clinically important parameters may be derived from Ti Ts and Tre muscle relaxation and muscle fade 16 1 Muscle relaxation is defined as 100 100 T1 Tre It can only be calculated if a reference measurement Tret has been carried out before injection of muscle relaxant 2 Muscle fade is defined as 100 100 T Ti As can be seen in figure 2 2 the fourth twitch of a TOF is markedly smaller than the first Muscle fade is also called TOF ratio or TOF value One can state that within certain limits the fade increases when the muscle relaxation incre
74. the inspection by eye The percentage of measurements that were considered invalid by eye was smaller in the test set 3 0 than in the learning set 8 5 This improvement is probably due to the better electrode placement and to better shielding of the connecting wires It may also be due to some unknown form of signal processing by the AS 3 monitor The number of measurements that were considered valid by the algorithm 28 1 was on the bounds of the acceptable range a minimum of circa 1 valid measurement every minute for the steady state phase Since the invalid measurements are not evenly distributed sometimes longer periods without valid measurements will occur that may interrupt the controller operation The algorithm correctly identified 85 of the artefacts in the test set which is close to the performance of the algorithm on the learning set However it is not 100 This may be due to several reasons e Inaccurate validation by eye As mentioned before it was especially difficult to judge small noise like signals The noise on the signal made it difficult to discern the muscles response A great deal of the time the signals were small e During the manual validation implicit rules of thumb may have been used However if these rules were a little fuzzy and were not applied very consequently which is typical for human experts it is difficult to reconstruct them by a global statistical analysis of the data S
75. the performance of the parameter may be tuned High values for dp will cause many invalid measurements to be let through but the measurements that are called invalid are sure to be unreliable Low values make the algorithm more restrictive so many valid measurements will be called invalid but artefacts are discovered almost certainly 55 It must be noted that dp relates to the performance of one parameter only and that the over all performance of the validation algorithm will depend on the dp s of all parameters As a compromise between the above extremes it was decided to try dp 0 50 for all parameters The results of this procedure are shown in appendix B which contains a list of all selected parameters and their bounds 56 6 Results The developed data acquisition system was tested in the operating rooms of the Catharina hospital The goals of these measurements were to test the reliability and accuracy of the data acquisition system and to acquire a set of TOF EMG signals to test the validation algorithm After informing the local medical ethical committee NMT measurements of 18 patients were recorded using the NMT module of the AS 3 ADU The EMG electrodes were placed according to figure 2 1 A total of 5129 measurements were collected After discussing the performance of the two data acquisition systems used for measuring the learning and test sets of data the performance of the validation algorithm will be discussed 6 1
76. the response goes from 100 to a value of about 0 In this phase the maximum sample frequency of 3 measurements per minute is actually not high enough So in this phase in theory the controller has only very little control over the patient because it does not have enough data After the onset phase a steady state phase starts In this phase the level of relaxation stays more or less constant Of course when the muscle relaxant control system is used the level of relaxation will be kept as constant as possible In this phase the dominant time constant is related to the duration of action of the muscle relaxant drug used This time constant is in the order of 27 standard deviation 5 0 minutes for vecuronium to circa 10 minutes for mivacurium This means that 1 valid measurement every 5 minutes would be enough However in order to have a 29 margin to improve the controller s performance a limit of one valid measurement per minute is proposed in the steady state phase Recovery of muscle relaxation is a slower process than the onset Here the same limit applies as in steady state Because of these considerations the maximal number of subsequently invalid measurements is 0 in the onset phase and 5 in the steady state phase 120 r T y T T 100 co J onset 60 T1 Tref 40 20 recovery steady state 0 0 10 20 30 40 50 60 time minutes Figure 4 1 Phases in the course of
77. tor 37 pins D male Kanaal 1 14 white 37 Channel 0 HI IN Kanaal 2 19 Ground 7 Link from Datex Relaxograph NMT 100 36 Channel 1 HI IN 35 Channel 2 HI IN 19 LL GND to PC with Keithley DAS 1402 A D conversion board Datex Relaxograph NMT 100 PC Analog output A D board Connector 8 pins DIN Connector 37 pins D maie Ground 6 shield 4 19 LL GND Output EMG amplifier 7 fe l 37 Channel 0 HI IN Triggerpulse 8 ae 36 Channel 1 HI IN Serial O Serial port COM2 Connector 9 pins D female Connector 25 pins D male Data out 8 green 2 Data in Data in 3 blue 3 Data out GND 5 black 7 GND 67 Appendix B Validation parameters and their bounds The TOF signal parameters that were used for validation of the learning and test sets are shown here For the test set all amplitude related criteria with dimension V or Vs were divided by 2 to adapt to the AS 3 ADU s output voltage range Table B 1 Criteria for valid measurements When two criteria are present measurements are considered valid only if both are met Parameter applies to Each ECAP in the TOF gt 0 Vs lt 0 03 Vs 5 gt 0 Vs lt 0 03 Vs Ts gt 0 Vs lt 0 03 Vs Ts gt 0 Vs lt 0 03 Vs Voer gt 0 4 V lt 0 4 V Vpci 2 gt 0 4 V lt 0 4 V Voc1 3 lt 0 4 V Voci 4 lt 04V Voca 1 lt 0 4 V Voc2 2 lt 04V Voc2 3 lt 04V Voc2 4 gt 0 3 V lt 0 3 V Vpcs 1 gt 0 3 V
78. ual validation of the learning and test set will be discussed 58 6 2 1 Goal of manual validation The goal was to score which train of four responses would yield a T1 Tret measurement of a quality that is reliable enough to use for closed loop control By doing this by eye also a better insight might be gained in the shape of the responses and in which parameters could possibly be interesting candidates for the automatic validation algorithm The learning set as well as the test set have to be validated by eye The learning set was recorded using the Relaxograph as an NMT device 6 2 2 Method for validation by eye Smans s measurement program EMG MEAS PAS was adapted to record and display validation information The information was stored in a previously unused bit in the record structure of the measurement files For each TOF the bit was set when the TOF was scored invalid and cleared if the TOF was scored valid The bit used is bit 3 of the 10th byte of the so called serial field in the TRF files This byte is also used to store the three Relaxograph flags The new coding is displayed in table 6 1 Table 6 1 New coding of the 10th byte of the serial field in the TRF files Bit HF disturbance flag electrodes off flag uncalibrated mode flag visual validation flag 0 valid 1 invalid not used The current state of the validation flag is displayed on the screen and the user may toggle the flag on and off An ind
79. uscle relaxation measurements was proposed implemented and tested on clinical data A number of parameters was analyzed Many signal parameters depend strongly on the level of muscle relaxation A validation algorithm should cope with these large variations The change in many parameters seemed to be limited A learning set and a test set of measurements were validated by visual inspection It proved difficult to judge the small noise like signals in case of deep levels of relaxation The algorithm recognized circa 85 of the artefactual measurements Based on the fact that a real optimization of the algorithm has not yet been carried out better results are probably possible 7 2 Recommendations The performance of the validation algorithm may be optimized in several ways e The validation by visual inspection should be verified by reviewing the measurements or by comparing the current validation by eye to a second opininion Was it consistent enough Are there more rules or parameters that can be derived from the visual inspection e Search for better parameters The parameters used in this study were relatively simple without or with only a simple compensation for the influence of the level of muscle relaxation If the complex relationship between signal parameters and the level of muscle relaxation is modelled better this may yield parameters that are better usable for validation e It could be interesting to see if an adapted form of

Validation of muscle relaxation measurements

Contents

Download Pdf Manuals

Related Search

Related Contents