International Journal of Psychophysiology, 13 ( 1992) 45-49 0 1992 Elsevier Science Publishers B.V. All rights reserved 0167-8760/92/$05.00
45
INTPSY 00392
ase N&a SziGgyi, Eszter Lhg Department of Comparatiw
and L&z16
Physiology, and Psychophysiology Research Group of Hungarian Academy of Science, Etitriis Lorcind Unirersily, Budapest (Hungary)
(Accepted 12 March 1992)
Kq words: Systolic time interval: Impedance cardiography; Computerized measurement; Waveform recognition
A computerized system has been developed based on impedance cardiography for processing systolic time intervals (STI) in psychophysiological experiments. The conventional method of ST1 determination is based on transducing three signals: electrocardiogram, phonocardiogram and pulse tracing (carotid pulse). Because of its negligible pulse transmission time the first derivative impedance cardiogram (dZ/dt) is more reliable than c:her pulse signals. Since the sharply demarcated points of the dZ/dt waveform occur simultaneously with the cardiac events the dZ/dt makes phonocardiogram processing unnecessary. Computer algorithms for ST1 assessment are based on processing two signals: the ECG and dZ/dt. The pre-ejection period (PEP), the left ventricular ejection time (LVET) and electromechanical systole (QS2) are derived by recognizing the B and X points on the dZ/dt signal as the endpoints of PEP and LVET respectively. X point identification consists of two steps: (i) the estimation of the QS2 based on the regression relationship between QS2 and heart rate (HR) and (ii) a filtering procedure for exact localization of the X point. The B point is determined by calculating the curvature function of the dZ/dt and employing a clustering procedure. The accuracy and reliability of the software were tested by processing data from 40 subjects under stress corldition (cold pressor and mental arithmetic).
INTRODUCTION Cardiologists recognized long ago the importance of systolic time intervals (ST11 as contractility-based measures in the assessment of cardiac performance (see Lewis et al., 1977). In the field of psychophysiology STIs are useful non-invasive indicators of sympathetic influences reflecting inotropic changes of myocardial activity (Newlin and Levenson, 1979; Light, 1985; L5ng and Sziligyi, 1988). The three basic STIs are the electromechanical systole (QS2), the left ventricular ejection
Correspondence to: N. Szilrigyi, Eiitvijs Lorind University, Department of Comparallve Physiology, 1088 Budapest Muzeum krt. 4/A, Hungary.
time (LVET) and the pre-ejection period (PEP). The PEP is the time interval between the Q wave of the ECG and the aortic valve opening. The LVET is the period of time between the opening and closing of the aortic valve. The QS2 consists 1 \IFT of the PEP anu-I +ha Lllcl-. - =. Trnditinnsl a I__________rnpacIIre ____-____ment of STIs requires processing of three signals: the electrocardiogram (ECG), the phonocardiogram (PCG) and the carotid pulse tracing (CPT), to time the opening and closing of the zortic valve. Measurements of aortic root pressure were compared with simultaneously recorded external carotid pulse waves. Phase analysis of the pulses showed a definite change in phase angle between the two pulses - the phase shift being most prominent at higher frequencies. This phase shift could affect the validity of the carotid pulse since landmarks of STIs could be influenced (Lewis et al.,
1977). A more useful noninvasive method which provides direct measures of STIs is impedance cardiography, originally developed by Kubicek ( 1966) for NASA. In this technique a low-intensity, high-frequency alternating current (4 mA at 100 MHz) is caused to flow longitudinally through the thorax and to oscillate between two ribbon electrodes applied around the neck and abdomen. The amount of electrical impedance which varies with the heartbeat can be detected by another two electrodes attached around the thorax above and below heart level. The first derivative of the impedance signal (dZ/dt 1 has several advantages over other pulse tracings. The two most important advantages are that dZjdt is a central pulse with negligible pulse transmission time and that it is less prone ttr noise interfercncc. dZ/dt can be used for direct timing of STls, replacing the CPT and eliminating the need for the PCC. (It is a very difficult task to guarantee high quality of PCGs for processing. 1
dZ /dt
----3
QS* I
0
1
1
02
Fig. I. Ikfini~ion tion pxioc1: LVET.
04
of rhc sysUlic
I
06
t
00
lirnc intewuls.
Icfr vcnlricular
I
1
PEP.
time(s) pre-cjec-
cjcclion tinic: OS?. Ax-
~ronl~chi~nici~l >~\~oIc.
COMPUTERIZED MEASUREMENT SYSTEM PROVIDING STls BASED ON ECG AN dZ/d t All the information concorning the above mcntioncd physioioglcai inrcrvals (US?. LVET. PEP) is inciuded in the dZ/dt signal together with the ‘iIVC.
The morphology of the dZ/dt is shown in Fig. 1: the nomenclature is adopted from Lamberts et al. (1984). The dZ/dt signal has characteristic points occurring simultaneously with the cardiac events in question (Lahabidi et al., 1970). The B point of dZ/dt corresponds to the opening of the aortic valve, while the X point indicates its closure. (The Y point indicates pulmonary valve closure.) The Q wave of the ECG curve is often absent or not differentiated. The d&y bctwcen the R and Q waves does not affect the observed changes in STI unless the conduction characteristics of the heart are drastically altered, a situation rare in psychophysiological measurements. Therefore.
many investigators USC‘the R wavt’ instead of the Q wave as the onset of QS2. An abbreviated PEP. calculated as the interval between the ECG point. may also be R wave’ and the dZ/dt sufficient ;IS a mcasurc of myocardial contractility. In order to accomplish continuous SWanalysis we have developed a computerized system based on an IBM/PC AT computer. Data acquisition is ital interface board (UAM 510 performed by a ungary) which contains A/D Unielectrotcrv. and D/A converters, 16 I/O channels and a special event-input, e.g., for capturing scores of the ECG QRS complex. (These are only its main functions. 1 The impedance cardiogram was obtained using 3M electrode tape No. M6001 (3M Company. Minnesota) and I PG-24 1 Dual Channcl Tctrapolar Impcdancc Plethysmograph (Hungary). The dZ/dt and mean body impedance (Z(l) arc sampled at 500 Hz and 2 Hz, respectively. The R wave is scored on the EGG by a Beckman R411 dynograph. R to R intervals were measured
47
dZ’/dt
(,X1
WAVESHAPE
Ohm/s) SUbJeCt
1400
(‘4) Algorithm for X ware iderrtificatim
600.
; 100.
300
i
i
5cO
;
;
700
t
900 ttme (ms)
( ,001
T
Ohm/s
1 subject
ts30
cold
pressor
1400 t
78.
600
HEART PERIOD
RR PEP LVE T
=656 = 914 -88 = 306
Cl52 PEPILVET
‘394 z
HR
.2876 time
Fig. 2. Upper
half: the plot of the first derivative
cardiogram (dZ/dt beginning
(rns.1
impedance
). One cardiac cycle of one subject. Bottom
half: the smoothed dZ/dt The
SOFTWARE
cold pressor
ff30
i
dZ/dt
RECOGNITION
curve superimposed
on the original.
and the end of the left ventricular
ejection
time as indicated by the computer.
by the real time clock, controlled by the computer, with 2 ms accuracy. The measurement was managed by the control program, which included sample processing routines, file maintenance, peripheral device handling, preliminary data analysis and conversion of the processed data to ASCII for later statistical analysis. The experimenter controlled the program using a menu and could edit his/her own experimental schedule. Waveform recognition was performed by application modules.
Our algorithms for the automatic recognition of impedance waveforms described below come from the software library recently developed in our laboratory for psychophysiological measurements and signal processing. The general idea was that if more specific feature characteristics of the waveforms are used, a more effective pattern recognition can be achieved. The X point is the endpoint of the QS2 and LVET. The assumption that the X point is the nadir (the most positive peak) of the dZ/dt curve serves as a base for many computational strategies. The fact that this is not always the case gave us the idea of developing an identification scheme which discards the above restriction. Our algorithms consisted of two consecutive steps: (8 predicting the probable occurrence of the X point, and (ii) exact localization of the point by a smoothing procedure (Szilagyi et al., 1989). (i) The time of the X wave was estimated for each cardiac cycle by calculating the regression of the QS2 against the heart rate of previous beats. In general, Weissler’s regression (Weissler, 1969) obtained with subjects lying supine was used for estimation. However, in stress situations van der Hoeven’s equation ( 1977), based on exercise experiments, proved to be more adequate. Despite these well established general equations, individual regression coefficients seemed to be more suitable in physiological or pharmacological interventions where the intra-individual variation is great (Mantisaari et al., 1984, 1986; Donders, 1980). Therefore, following the first ten cycles (in which QS2 is calculated using the van der Hoeven or Ueissler equation) the computation of QS2 was based on the subject’s own actual regression line. However, whenever the correlation did not differ significantly from zero, the program reverted to the van der Hoeven or Weissler equation. In order to eliminate artefacts only HPs within 1 S.D. were used. A 50 ms wide search window was placed symmetrically around the predicted endpoint of the QS2 (since the time differ-
48
ence between the X and Y waves of the dZ/dt is about 25 ms (Lababidi et al., 1970). (ii) The exact localization of the X point inside the window is achieved by a frequency discrimination procedure. The frequency components in the original dZ/dt that belong to the X wave (and ail the higher frequencies) were removed by an appropriate low-pass digital filter. The iocation of the maximal difference between the original and the filtered version of the dZ/dt signal is considered as the X point (Fig. 2). For some of the subjects no sharp, definite X wave is apparent (Kubicek et al., 1970): in such cases automatic waveform detection is unlikely to be reliable. Thus, it is recommended, not only for detection of the X wave, but also for detection of point I3 (and for ail the waveforms in general Sherwood et al., 1990) for the program to allow the operator to check the shape of the dZ/dt signal cycle by cycle, and to supervise decisions made by the computer. The interactive graphics editing capability of our semi-automated program provided an opportunity for human intervention each time signal quality was inadequate. (If the impedance signal proves to be quite inappropriate for computer processing, it is advisable to use a phonocardiogram.)
(B) Algorithm for B point idcrttificatim
QS2 is automatically divided into LVET by identifying the B point of t cath-holding, end-exnt lies on the zero reference line of the dZ/dt. This is why Rasmussen et al. (1975) used the zero crossing point prior to the most negative peak of the dZ/dt as the onset of LVET. Due to small oscillations in the waveform, Kubicek et al. (1970) suggested using the point on the curve equal to 0.15 dZ/dt/max. This is obtained by going back in time down the dZ/dt wave from the most negative peak. In stress situations, however, certain alterations may occur in tho impedance waveform (Hypoxic stress ( aiasubramanian et al., 1978); and in the mental arithmetic test (Lang et al., 1989a) even if the respiration-related baseline movements are eliminated.
These facts confirm the importance of defining the I3 point as the onset of left ve tion, i.e., as the onset of the rapid upstroke of dZ/dt (Sheps et al., 1982; Sherwood et al.. 1990). Our algorithm identifies the onset of the rapid upstroke in a special way. The shape of the dZjdt is characterized by the curvature function. This measure has its highest values at that part of dZ/dt which corresponds to the beginning of ventricular ejection. These characteristic points of the dZ/dt curve constitute a loosely contiguous cluster which is determined by an agglomerative hierarchical clustering atgorithm. ‘I’he first element of the cluster corresponds to the B point.
EVALUATION OF IMPEDANCE ANALYSIS SOFTWARE
SIGNAL
The advantpges of our dZ/dt processing program were shown in experiments when true STIs of 40 healthy male subjects were studied under stress conditions (cold pressor test, mental arithmetic). The evaluation of the physiological findings of the experiment will be discussed in a separate paper. A service program has been developed which continuously monitors the system’s r each cardiac cycle t curve with the corn mined events (endpoints of PEP and QS2) and allows the operator to reject physiologically unreasonable data (artifacts). To evaluate the accuracy and reliability of the ST1 analysis program the computer processed events were compared to those marked by the experimenter. Of 90 randomly selected periods, 8 were marked erroneously by the computer in the case of the point (correct identification: 91%) and 15 in the case of the I3 point (correct identification: 83%). The missing X point seems to be due to the discrepancy between the estimated and real occurrchce time of the X wave for the actual period. The algorithm might be more effective by adding some monitoring of the sudden changes (e.g., in arrhythmia) of the heart period.
49
REFERENCES Balasubramanian. V.. Mathew, O.P., Behl. A., Tewari, S.C. and Hoon. R.S. (1978) Electrical impedance cardiogram in derivation of systolic time intervals. Br. Heart 1, 40: 26% 275.
Divers,
.T.. Katona. P.G., Dauchot, P.J. and Hung, J.C. tinuous real-time computation and display of systolic time intervals from surgical patients. Camp. Biomed. Res.. 10: 45-59. Donders. J.J.H. (1980) Computer algorithms for automatic determination of systolic time intervals. In W.F. List, J.S. Gravenstein and D.H. Spodic (Eds.1, Systolic rime inrerrals. Springer-Verlag, Berlin, pp. 100. Kubicek. W.G.. Karnegis, J.N., Patterson, R.P.. Witsoe, D.A. and Mattson, R.H. (1966) Development and evaluation of an impedance cardiac output system. Aerospace Med., 37: 1208-1212. Kubicek. W-G.. Patterson, R.P. and Witsoe, D.A. (1970) Impedance cardiography as a noninvasive method of monitoring cardiac function and other parameters of the cardiovascular system. Ann. NY. Acad. Sci., 170: 724-732. Lababidi, 2.. Ehmke. D.A., Durnin, R.E., Leaverton, P.E. and Lauer. R.M. (1970) The first derivative tboracic impedance cardiogram. Circulation, 41: 65 l-658. Lamberts, R., Visser, K.R. and Zijlstra, W.G. (1984) Impedance cardiography, V,an Gorcum, Assen. Lang. E.. Sziligyi. N. and Adam, G. (1989) Simultaneous non-invasive measurements of parasympathetic and sympathetic influences on the heart in stress related reactions in humans. In T. Radil and Z. Bohdanecky (Eds.), froceedijlgs of the Fourtl7 Conference of the fnrernational Organization of Psychophysiology, Prague, September 12- 17,
1988, pp. 158. Lewis, R.P.. Rittgers. S.E.. Forester, W.F. and Boudoulas, H. (1977) A critical review of the systolic time intervals. Circulation. 56: 146-158.
Light. K.C. (1985) Cardiovascular and renal responses to competitive mental challenges. In J.F. Orlebeke, G. Mul-
der and L.J.P. van Doornen (Eds.1, Psychophysiofogy of cardiorascular control, Plenum Press, New York, pp. 683. Mantysaari. M., Antila, X., Halkola. L., Kero, P. rind Lansimies, E. (1986) Individual relationship between heart rate and electromechanic systole in orthostatic test during autonomic blockade. Acta Cardiologica, 41: 271-281. Mantysaari, M., Antila. K. and Peltonen, T. (1984) Relationship between systolic time intervals and heart rete during four circulatory stress tests. Ew. 1. Appf. Phgiol.. 52: 282-286.
Mohapatra, S.N. (1981) Non-invasive cardiovascular monitoring by electrical impedance technique, Pitman Medical. Newlin, D. and Levenson, R.W. (1979) Pre-ejection period: measuring betaadrenergic influences upon the heart. Psychophysiology, 16: 546.
Rasmussen, J.P., Sorensen, G. and Kahn, T. (1975) Evaluation of impedance cardiography as a non-invasive means of measuring systolic time intervals and cardiac output. Acta Anaesth. Stand., f 9: 21 O-2 18. Sheps, D-S., Petrovick, M.L., Kizakevich, P.N., Wolfe, C. and Craige, E. (1982) Continuous non-invasive monitoring of left ventricular function during exercise by thoracic impedance cardiography-automated derivation of systolic time intervals. Am. Heart J., 103: 519-524. Sherwood, A., Allen, M.T., Fahrenberg, J., Kelsey, R.M., Lovallo, W.R. and Van Doornen, L.J.P. (1990) Methodological guidelines for impedance cardiography. Psychophysiology, 27: 1-23. Sziligyi, N. and Ling, E. (1989) Computer methods for online automatic acquisition of systolic time intervals measured by impedance cardiography. Int. J. Psychophysiof., 7 (Abstracts of the 4th Conference of the International Organization of Psychophysiology held in Prague, 19881:406-407. Van der Hoeven, G.M.A., Clerens, P.J.A., Donders, J.J.H., Beneken, J.E.W. and Vonk, J.T.C. (1977) A study of systolic time intervals during uninterrupted exercise. Br. Heart J., 39: 242-254.
Weissler, A.M., Harris, W.S. and Schoenfeld, C.D. (19691 Bedside techniques for the evaluation of ventricular function in man. Am. J. Cardiol., 23: 5’77.
45
INTPSY 00392
ase N&a SziGgyi, Eszter Lhg Department of Comparatiw
and L&z16
Physiology, and Psychophysiology Research Group of Hungarian Academy of Science, Etitriis Lorcind Unirersily, Budapest (Hungary)
(Accepted 12 March 1992)
Kq words: Systolic time interval: Impedance cardiography; Computerized measurement; Waveform recognition
A computerized system has been developed based on impedance cardiography for processing systolic time intervals (STI) in psychophysiological experiments. The conventional method of ST1 determination is based on transducing three signals: electrocardiogram, phonocardiogram and pulse tracing (carotid pulse). Because of its negligible pulse transmission time the first derivative impedance cardiogram (dZ/dt) is more reliable than c:her pulse signals. Since the sharply demarcated points of the dZ/dt waveform occur simultaneously with the cardiac events the dZ/dt makes phonocardiogram processing unnecessary. Computer algorithms for ST1 assessment are based on processing two signals: the ECG and dZ/dt. The pre-ejection period (PEP), the left ventricular ejection time (LVET) and electromechanical systole (QS2) are derived by recognizing the B and X points on the dZ/dt signal as the endpoints of PEP and LVET respectively. X point identification consists of two steps: (i) the estimation of the QS2 based on the regression relationship between QS2 and heart rate (HR) and (ii) a filtering procedure for exact localization of the X point. The B point is determined by calculating the curvature function of the dZ/dt and employing a clustering procedure. The accuracy and reliability of the software were tested by processing data from 40 subjects under stress corldition (cold pressor and mental arithmetic).
INTRODUCTION Cardiologists recognized long ago the importance of systolic time intervals (ST11 as contractility-based measures in the assessment of cardiac performance (see Lewis et al., 1977). In the field of psychophysiology STIs are useful non-invasive indicators of sympathetic influences reflecting inotropic changes of myocardial activity (Newlin and Levenson, 1979; Light, 1985; L5ng and Sziligyi, 1988). The three basic STIs are the electromechanical systole (QS2), the left ventricular ejection
Correspondence to: N. Szilrigyi, Eiitvijs Lorind University, Department of Comparallve Physiology, 1088 Budapest Muzeum krt. 4/A, Hungary.
time (LVET) and the pre-ejection period (PEP). The PEP is the time interval between the Q wave of the ECG and the aortic valve opening. The LVET is the period of time between the opening and closing of the aortic valve. The QS2 consists 1 \IFT of the PEP anu-I +ha Lllcl-. - =. Trnditinnsl a I__________rnpacIIre ____-____ment of STIs requires processing of three signals: the electrocardiogram (ECG), the phonocardiogram (PCG) and the carotid pulse tracing (CPT), to time the opening and closing of the zortic valve. Measurements of aortic root pressure were compared with simultaneously recorded external carotid pulse waves. Phase analysis of the pulses showed a definite change in phase angle between the two pulses - the phase shift being most prominent at higher frequencies. This phase shift could affect the validity of the carotid pulse since landmarks of STIs could be influenced (Lewis et al.,
1977). A more useful noninvasive method which provides direct measures of STIs is impedance cardiography, originally developed by Kubicek ( 1966) for NASA. In this technique a low-intensity, high-frequency alternating current (4 mA at 100 MHz) is caused to flow longitudinally through the thorax and to oscillate between two ribbon electrodes applied around the neck and abdomen. The amount of electrical impedance which varies with the heartbeat can be detected by another two electrodes attached around the thorax above and below heart level. The first derivative of the impedance signal (dZ/dt 1 has several advantages over other pulse tracings. The two most important advantages are that dZjdt is a central pulse with negligible pulse transmission time and that it is less prone ttr noise interfercncc. dZ/dt can be used for direct timing of STls, replacing the CPT and eliminating the need for the PCC. (It is a very difficult task to guarantee high quality of PCGs for processing. 1
dZ /dt
----3
QS* I
0
1
1
02
Fig. I. Ikfini~ion tion pxioc1: LVET.
04
of rhc sysUlic
I
06
t
00
lirnc intewuls.
Icfr vcnlricular
I
1
PEP.
time(s) pre-cjec-
cjcclion tinic: OS?. Ax-
~ronl~chi~nici~l >~\~oIc.
COMPUTERIZED MEASUREMENT SYSTEM PROVIDING STls BASED ON ECG AN dZ/d t All the information concorning the above mcntioncd physioioglcai inrcrvals (US?. LVET. PEP) is inciuded in the dZ/dt signal together with the ‘iIVC.
The morphology of the dZ/dt is shown in Fig. 1: the nomenclature is adopted from Lamberts et al. (1984). The dZ/dt signal has characteristic points occurring simultaneously with the cardiac events in question (Lahabidi et al., 1970). The B point of dZ/dt corresponds to the opening of the aortic valve, while the X point indicates its closure. (The Y point indicates pulmonary valve closure.) The Q wave of the ECG curve is often absent or not differentiated. The d&y bctwcen the R and Q waves does not affect the observed changes in STI unless the conduction characteristics of the heart are drastically altered, a situation rare in psychophysiological measurements. Therefore.
many investigators USC‘the R wavt’ instead of the Q wave as the onset of QS2. An abbreviated PEP. calculated as the interval between the ECG point. may also be R wave’ and the dZ/dt sufficient ;IS a mcasurc of myocardial contractility. In order to accomplish continuous SWanalysis we have developed a computerized system based on an IBM/PC AT computer. Data acquisition is ital interface board (UAM 510 performed by a ungary) which contains A/D Unielectrotcrv. and D/A converters, 16 I/O channels and a special event-input, e.g., for capturing scores of the ECG QRS complex. (These are only its main functions. 1 The impedance cardiogram was obtained using 3M electrode tape No. M6001 (3M Company. Minnesota) and I PG-24 1 Dual Channcl Tctrapolar Impcdancc Plethysmograph (Hungary). The dZ/dt and mean body impedance (Z(l) arc sampled at 500 Hz and 2 Hz, respectively. The R wave is scored on the EGG by a Beckman R411 dynograph. R to R intervals were measured
47
dZ’/dt
(,X1
WAVESHAPE
Ohm/s) SUbJeCt
1400
(‘4) Algorithm for X ware iderrtificatim
600.
; 100.
300
i
i
5cO
;
;
700
t
900 ttme (ms)
( ,001
T
Ohm/s
1 subject
ts30
cold
pressor
1400 t
78.
600
HEART PERIOD
RR PEP LVE T
=656 = 914 -88 = 306
Cl52 PEPILVET
‘394 z
HR
.2876 time
Fig. 2. Upper
half: the plot of the first derivative
cardiogram (dZ/dt beginning
(rns.1
impedance
). One cardiac cycle of one subject. Bottom
half: the smoothed dZ/dt The
SOFTWARE
cold pressor
ff30
i
dZ/dt
RECOGNITION
curve superimposed
on the original.
and the end of the left ventricular
ejection
time as indicated by the computer.
by the real time clock, controlled by the computer, with 2 ms accuracy. The measurement was managed by the control program, which included sample processing routines, file maintenance, peripheral device handling, preliminary data analysis and conversion of the processed data to ASCII for later statistical analysis. The experimenter controlled the program using a menu and could edit his/her own experimental schedule. Waveform recognition was performed by application modules.
Our algorithms for the automatic recognition of impedance waveforms described below come from the software library recently developed in our laboratory for psychophysiological measurements and signal processing. The general idea was that if more specific feature characteristics of the waveforms are used, a more effective pattern recognition can be achieved. The X point is the endpoint of the QS2 and LVET. The assumption that the X point is the nadir (the most positive peak) of the dZ/dt curve serves as a base for many computational strategies. The fact that this is not always the case gave us the idea of developing an identification scheme which discards the above restriction. Our algorithms consisted of two consecutive steps: (8 predicting the probable occurrence of the X point, and (ii) exact localization of the point by a smoothing procedure (Szilagyi et al., 1989). (i) The time of the X wave was estimated for each cardiac cycle by calculating the regression of the QS2 against the heart rate of previous beats. In general, Weissler’s regression (Weissler, 1969) obtained with subjects lying supine was used for estimation. However, in stress situations van der Hoeven’s equation ( 1977), based on exercise experiments, proved to be more adequate. Despite these well established general equations, individual regression coefficients seemed to be more suitable in physiological or pharmacological interventions where the intra-individual variation is great (Mantisaari et al., 1984, 1986; Donders, 1980). Therefore, following the first ten cycles (in which QS2 is calculated using the van der Hoeven or Ueissler equation) the computation of QS2 was based on the subject’s own actual regression line. However, whenever the correlation did not differ significantly from zero, the program reverted to the van der Hoeven or Weissler equation. In order to eliminate artefacts only HPs within 1 S.D. were used. A 50 ms wide search window was placed symmetrically around the predicted endpoint of the QS2 (since the time differ-
48
ence between the X and Y waves of the dZ/dt is about 25 ms (Lababidi et al., 1970). (ii) The exact localization of the X point inside the window is achieved by a frequency discrimination procedure. The frequency components in the original dZ/dt that belong to the X wave (and ail the higher frequencies) were removed by an appropriate low-pass digital filter. The iocation of the maximal difference between the original and the filtered version of the dZ/dt signal is considered as the X point (Fig. 2). For some of the subjects no sharp, definite X wave is apparent (Kubicek et al., 1970): in such cases automatic waveform detection is unlikely to be reliable. Thus, it is recommended, not only for detection of the X wave, but also for detection of point I3 (and for ail the waveforms in general Sherwood et al., 1990) for the program to allow the operator to check the shape of the dZ/dt signal cycle by cycle, and to supervise decisions made by the computer. The interactive graphics editing capability of our semi-automated program provided an opportunity for human intervention each time signal quality was inadequate. (If the impedance signal proves to be quite inappropriate for computer processing, it is advisable to use a phonocardiogram.)
(B) Algorithm for B point idcrttificatim
QS2 is automatically divided into LVET by identifying the B point of t cath-holding, end-exnt lies on the zero reference line of the dZ/dt. This is why Rasmussen et al. (1975) used the zero crossing point prior to the most negative peak of the dZ/dt as the onset of LVET. Due to small oscillations in the waveform, Kubicek et al. (1970) suggested using the point on the curve equal to 0.15 dZ/dt/max. This is obtained by going back in time down the dZ/dt wave from the most negative peak. In stress situations, however, certain alterations may occur in tho impedance waveform (Hypoxic stress ( aiasubramanian et al., 1978); and in the mental arithmetic test (Lang et al., 1989a) even if the respiration-related baseline movements are eliminated.
These facts confirm the importance of defining the I3 point as the onset of left ve tion, i.e., as the onset of the rapid upstroke of dZ/dt (Sheps et al., 1982; Sherwood et al.. 1990). Our algorithm identifies the onset of the rapid upstroke in a special way. The shape of the dZjdt is characterized by the curvature function. This measure has its highest values at that part of dZ/dt which corresponds to the beginning of ventricular ejection. These characteristic points of the dZ/dt curve constitute a loosely contiguous cluster which is determined by an agglomerative hierarchical clustering atgorithm. ‘I’he first element of the cluster corresponds to the B point.
EVALUATION OF IMPEDANCE ANALYSIS SOFTWARE
SIGNAL
The advantpges of our dZ/dt processing program were shown in experiments when true STIs of 40 healthy male subjects were studied under stress conditions (cold pressor test, mental arithmetic). The evaluation of the physiological findings of the experiment will be discussed in a separate paper. A service program has been developed which continuously monitors the system’s r each cardiac cycle t curve with the corn mined events (endpoints of PEP and QS2) and allows the operator to reject physiologically unreasonable data (artifacts). To evaluate the accuracy and reliability of the ST1 analysis program the computer processed events were compared to those marked by the experimenter. Of 90 randomly selected periods, 8 were marked erroneously by the computer in the case of the point (correct identification: 91%) and 15 in the case of the I3 point (correct identification: 83%). The missing X point seems to be due to the discrepancy between the estimated and real occurrchce time of the X wave for the actual period. The algorithm might be more effective by adding some monitoring of the sudden changes (e.g., in arrhythmia) of the heart period.
49
REFERENCES Balasubramanian. V.. Mathew, O.P., Behl. A., Tewari, S.C. and Hoon. R.S. (1978) Electrical impedance cardiogram in derivation of systolic time intervals. Br. Heart 1, 40: 26% 275.
Divers,
.T.. Katona. P.G., Dauchot, P.J. and Hung, J.C. tinuous real-time computation and display of systolic time intervals from surgical patients. Camp. Biomed. Res.. 10: 45-59. Donders. J.J.H. (1980) Computer algorithms for automatic determination of systolic time intervals. In W.F. List, J.S. Gravenstein and D.H. Spodic (Eds.1, Systolic rime inrerrals. Springer-Verlag, Berlin, pp. 100. Kubicek. W.G.. Karnegis, J.N., Patterson, R.P.. Witsoe, D.A. and Mattson, R.H. (1966) Development and evaluation of an impedance cardiac output system. Aerospace Med., 37: 1208-1212. Kubicek. W-G.. Patterson, R.P. and Witsoe, D.A. (1970) Impedance cardiography as a noninvasive method of monitoring cardiac function and other parameters of the cardiovascular system. Ann. NY. Acad. Sci., 170: 724-732. Lababidi, 2.. Ehmke. D.A., Durnin, R.E., Leaverton, P.E. and Lauer. R.M. (1970) The first derivative tboracic impedance cardiogram. Circulation, 41: 65 l-658. Lamberts, R., Visser, K.R. and Zijlstra, W.G. (1984) Impedance cardiography, V,an Gorcum, Assen. Lang. E.. Sziligyi. N. and Adam, G. (1989) Simultaneous non-invasive measurements of parasympathetic and sympathetic influences on the heart in stress related reactions in humans. In T. Radil and Z. Bohdanecky (Eds.), froceedijlgs of the Fourtl7 Conference of the fnrernational Organization of Psychophysiology, Prague, September 12- 17,
1988, pp. 158. Lewis, R.P.. Rittgers. S.E.. Forester, W.F. and Boudoulas, H. (1977) A critical review of the systolic time intervals. Circulation. 56: 146-158.
Light. K.C. (1985) Cardiovascular and renal responses to competitive mental challenges. In J.F. Orlebeke, G. Mul-
der and L.J.P. van Doornen (Eds.1, Psychophysiofogy of cardiorascular control, Plenum Press, New York, pp. 683. Mantysaari. M., Antila, X., Halkola. L., Kero, P. rind Lansimies, E. (1986) Individual relationship between heart rate and electromechanic systole in orthostatic test during autonomic blockade. Acta Cardiologica, 41: 271-281. Mantysaari, M., Antila. K. and Peltonen, T. (1984) Relationship between systolic time intervals and heart rete during four circulatory stress tests. Ew. 1. Appf. Phgiol.. 52: 282-286.
Mohapatra, S.N. (1981) Non-invasive cardiovascular monitoring by electrical impedance technique, Pitman Medical. Newlin, D. and Levenson, R.W. (1979) Pre-ejection period: measuring betaadrenergic influences upon the heart. Psychophysiology, 16: 546.
Rasmussen, J.P., Sorensen, G. and Kahn, T. (1975) Evaluation of impedance cardiography as a non-invasive means of measuring systolic time intervals and cardiac output. Acta Anaesth. Stand., f 9: 21 O-2 18. Sheps, D-S., Petrovick, M.L., Kizakevich, P.N., Wolfe, C. and Craige, E. (1982) Continuous non-invasive monitoring of left ventricular function during exercise by thoracic impedance cardiography-automated derivation of systolic time intervals. Am. Heart J., 103: 519-524. Sherwood, A., Allen, M.T., Fahrenberg, J., Kelsey, R.M., Lovallo, W.R. and Van Doornen, L.J.P. (1990) Methodological guidelines for impedance cardiography. Psychophysiology, 27: 1-23. Sziligyi, N. and Ling, E. (1989) Computer methods for online automatic acquisition of systolic time intervals measured by impedance cardiography. Int. J. Psychophysiof., 7 (Abstracts of the 4th Conference of the International Organization of Psychophysiology held in Prague, 19881:406-407. Van der Hoeven, G.M.A., Clerens, P.J.A., Donders, J.J.H., Beneken, J.E.W. and Vonk, J.T.C. (1977) A study of systolic time intervals during uninterrupted exercise. Br. Heart J., 39: 242-254.
Weissler, A.M., Harris, W.S. and Schoenfeld, C.D. (19691 Bedside techniques for the evaluation of ventricular function in man. Am. J. Cardiol., 23: 5’77.
