study of high frequency components in electrocardiogram...

Study of High Frequency Components inElectrocardiogram by PowerSpectrum Analysis

By Ernst K. Franke, Dr. Ing., John R. Braunstein, M.D., Ph.D.,and David C. Zellner, M.D.

• In a paper read before the Chelsea ClinicalSociety on March 19, 1912, Binthoven statedthat a string galvanometer, with a deflectingtime of 0.01 sec./mm. when standardized inthe usual manner, was adequate to recordthe form and dimensions of the electrocardio-gram.* He further stated that if the stringcould be made 10 or 100 times faster, the sen-sitivity remaining constant, no perceptiblechange would be noted in the record.1 Thisconclusion remained unchallenged until 1933,when Reid and Caldwell2 reopened the issue.Using a vacuum tube amplifier and a bifilaroscillograph, they showed in a tracing takenon a normal subject that the one hundredthharmonic was twice as large as the funda-mental. Since then, although a number ofworkers have examined the problem," theissue remains open. It is not, however, thepurpose of this investigation to shed newlight on the proper techniques for recordingthe form and dimensions of the conventionalelectrocardiogram, but rather by a newmethod to explore the possibility that thehigher frequency components may containuseful information. This approach was sug-gested to us particularly by the work ofLangner, in which he showed beading andnotching of the QTCS complex in patients withischemic heart disease."

A study of the higher frequencies may beundertaken in a number of different ways:first, the usual amplitude-time record can be

From tlie Laboratories of Cardiology and Bio-physics, University of Cincinnati, and tlie Depart-ment of Medicine, Cincinnati General Hospital, Cin-cinnati, Ohio.

Supported in part by IT. S. Public Health ServiceGrant H-37(i9.

'Received for publication November 10, 196.1.•For a critically damped string this is about 6

decibels down at 20 c.p.s.

870

written with sufficient width and speed toobtain the necessary resolution at high fre-quencies ; second, an autocorrelation-time dis-placement record can be obtained by analogor digital computation techniques; and, third,a method can be used to study the powerspectrum.

In the present investigation, we haveadopted the latter. In doing so, we havechosen to consider some of the average sta-tistical properties of the electrocardiogramrather than its details. Since it was not knownin the beginning for which details one wouldhave to look, nor in which frequency rangethe significant features would be found, thisapproach seemed to have the advantage thatit might permit a general survey of the higherfrequencies without the accumulation of un-manageable amounts of records. Althoughsome information is evidently lost in this sta-tistical condensation process, it has greatlyfacilitated the solution of such problems asthe signal-to-noise level and the optimal choiceof frequency bands which can ultimately bechosen for a detailed analysis.

MethodsThe subject was put in a low Fowler's position

on a standard hospital bed, which was enclosed ina double-copper screen shielding cage to eliminateexternal electrical disturbances. Standard eleetro-enrdiographic leads I, II, and III were used. Thesewere connected one by one to a battery-powered,low noise, transistor amplifier (Medistor A-32 B),set to a gain of 1,000. This gain was in all easessufficiently high so that the amplifier could beconnected to the rest of the instrumentation, whichis outside the shielding' cage, without unduly rais-ing1 the noise level. The signal was monitored byan oscilloscope and further amplified and filteredby a variable band-pass (Krohn Hitc, ultra lowfrequency band-pass 333-A). The output of theband-pass was then fed into an amplifier withcontinuously adjustable, calibrated gain (Philbriek

Circulation Research. Volume X. June 19G2

by guest on June 10, 2018http://circres.ahajournals.org/

Dow

nloaded from

http://circres.ahajournals.org/

POWER SPECTRUM ANALYSIS 871

co-efficient unit). This was done for the purposeof raising the signal to the proper recording level.The signnl was subsequently electronically squaredin a quadratic rectifier (Philbrick squaring unit),the output voltage of which was proportional tothe signnl power. Since the power spectral densityis defined as an average (ideally to be taken over arecord of infinite length), the output of the quad-ratic rectifier must be integrated and the integraldivided by the length of the integration interval.A low puss filter with a properly set band limitwas the electronic analog for this part of the datn-proeessing sequence (Krohn Hite, ultra low fre-quency band-pass 330-A, with the lower bandlimit set as low as possible. This substitution ispermissible because the system need not pnssD.C.). The output of the integrator was then pro-portional to the power spectral density of the signalgenerated by the electromotive force of the heartin the respective band-pass and was recorded on astandard two-channel direct-writing- electrocardio-graph, together with a. standard low frequencyelectrocardiogram.

A detailed mathematical justification for thisanalog- computation procedure is beyond the scopeof this paper, particularly since it can be foundin a number of standard communication engineer-ing texts. A succinct treatment of the topic maybe found, for instance, in reference 32.

The complete recording of all frequency bands(0 to 1,000 c.p.s.) used took 30 to 40 minutes.The records were then analyzed. The main powerwas concentrated in a. small time interval corre-sponding to the QRS complex. The height of thepulses in this record was proportional to thepower in the respective beats. Since breathing hasan influence on the height of the pulses in thehigh frequency tracings in both the normal sub-jects and those with heart disease, the averagepower was taken over 10 cycles. The breathingeffects were averaged out in this way. Four differ-ent groups of subjects were investigated : normal,young adults; patients with myocardial infarcts;patients with ischemic heart disease without demon-strable infarction; and middle-aged, normal sub-jects.

ResultsIn figure 1, a typical single-lead record is

shown ; the upper trace is the standard elec-trocardiogram, the lower trace is the filtered,squared, and integrated cardiac potential(lead II). The height of the wave of the lowertrace is proportional to the power generatedby the electromotive force of the heart.

In tables 1 through 4, the original dataobtained by the power spectrum analysis have

1

•

int

; • •

: : : ;

: : : :

- ~

i

lii

i

A

Lead II Range 90 - r 180FIGURE 1

This figure illustrates a typical record. A Sanbornt'win-viso was used for this purpose. Standard leadIT is illustrated in the upper trace. In the lowertrace the filtered, squared, and integrated poten-tial is recorded simultaneously for standard leadIT hi the range from 90 to ISO c.p.s.

been compiled. In the top row, the bandlimits of the filters are listed in cycles persecond. Overlapping octave bands were vised,the lower frequency limit of any band beingthe center frequency of the preceding one. Inthe second row, a weighting factor is shownfor each band. This factor was introducedfor convenience, in order to bring the numer-ical values of the relative power in the variousbands to the same order of magnitude. Inthe columns corresponding to the variousbands, the ratio of the spectral power in therespective band to the power in the 0 to (i0c.p.s. band (standard eardiographie range,not shown), multiplied by the weighting fac-tor, is listed. To calculate the actual powerratio between any of. the bands, one has todivide the data of any one column by its re-spective weighting factor.

The weighting factor was calculated in thefollowing way: A group of 12 young, healthysubjects (male college freshmen) was initiallychosen, and their spectra determined. Thepower ratio was averaged for each band anda numerical factor determined which wouldreduce these averages to unity. The adoptionof this way of representing the data has alsogreatly facilitated the graphical representa-tion.

Figure 2(A) shows the typical power spec-trum of a normal, young adult. On the ab-

Circulation Research, Volume X, June 19G2


Dow

nloaded from


872 FRANKE, BEAUNSTEIN, ZELLNER

Power Spectra: Healthy Young AdultsTABLE 1

Range(c.p.s.)

Weightingfactor

# 1# 2# 3# 4# 5# «# 7# 8# 9#10

#11#12#13#14#15#10#17#18#19#20

#21#22#23#24#25#20#27

30-60

1.00

1.49

1.16

0.94

0.9G

1.03

0.75

O.iSG

1.450.S71.941.121.121.011.401.101.080.961.111.430.901.600.991.101.070.881.321.04

40-80

1.93

2.281.911.001.121.160.930.571.241.132.822.001.191.372.731.41

o.so0.900.981.871.672.150.970.601.380.731.341.17

60-120

2.54

2.262.310.091.301.72

'1.080.732.500.942.602.161.051.702.771.490.750.830.052.083.023.081.070.091.330.041.061.41

90-180

3.54

1.622.020.091.271.411.140.603.811.462.180.780.591.301.350.900.811.050.581.572.243.801.321.000.770.600.921.37

130-260

6.11

2.171.651.971.271.151.140.362.420.841.400.010.521.210.810.810.801.870.551.072.403.921.331.830.500.690.411.05

180-360

13.0

2.961.321.850.750.740.930.292.160.580.840.740.420.760.550.620.841.440.941.742.402.901.411.970.660.990.611.09

250-500

34.0

3.591.911.740.840.760.820.442.010.671.260.890.600.811.230.080.991.330.552.072.202.092.412.420.821.460.570.93

350-700

81.0

3.112.001.450.700.700.780.542.640.821.070.870.680.591.810.700.891.570.581.581.451.00—

2.700.541.450.670.01

scissa, the bands are indicated; the ordinatesshow the weighted, relative power, taken fromtable 1. From table 1, the medians were alsocalculated for every band. As a measure forthe scatter of the data, the semi-interquartilerange is used. These quantities are plottedin figure 2(B) ; the. semi-interquartile rangesare represented by gray shading.

It is not possible to summarize tables 2 and3 in the same way, because the spectra ofthese groups of patients have wide variations.The spectra may be divided, however, intothree relatively uniform groups. There aresome spectra which resemble the normal onesrather closely. Figure 3 (A) shows mediansand semi-interquartile ranges of nine subjects,of whom six had acute and three old myo-cardial infarcts. Tracings on the six acutecases were taken prior to discharge, and inall instances the S-T segments of the electro-

cardiogram had returned to the baseline. Theelectrocardiographic age of the infarcts variedfrom 25 to 32 days. The age of the remainingthree infarctions varied from one to fiveyears. In two, there was no S-T elevation ordepression, and in one the persistent elevationof the S-T segment in V4 for over a year sug-gested the presence of an aneurysm. This grouporiginally consisted of 10 patients, but onehad to be discarded because the patient diedand no infarct could be demonstrated at au-topsy. In a small number of patients, ex-tremely high power was found in some, if notall, of the higher frequency ranges. Medianand semi-interquartile range of this group(six subjects) is plotted in figure 3(B). Allof these were recovering from acute infarc-tions ranging in age from 6 to 37 days.Four showed definite residual displacementof the S-T segment. Of the remaining two,

Circulation Research, Volume X, June iiiC,2


Dow

nloaded from



both with inferior infarction, one showed 0.5-mm. displacement of the S-T segment in leadIII and marked coving. The other showed nosegment displacement, marked coving, andchanges in the precordial leads suggestingposterior ischemia. There is, finally, an inter-mediate group, having moderately elevatedpower between 40 and 180 c.p.s. A graph forthis group is shown in figure 3(C), compiledfrom data of seven patients Avith old infarcts,and five patients recovering from acute in-farcts. In this group, four showed displace-ment of the S-T segments. Three of thesetracings were on patients recovering fromacute infarction and about to be discharged.The fourth was on a patient with an infarc-tion approximately five years old. Shortlyafter the tracing was taken, the patient diedon the commode, and the infarct was demon-strated at autopsy.

The data of table 3 may be subdivided inthe same way a,s those of table 2. Again,there are a number of spectra which are in-distinguishable from normal ones, as shownin figure 4(A). Of these data, seven are frompatients with chest pain as a prominent symp-tom, seven from patients with the same diag-nosis but without conspicuous chest pain.Spectra with unusually high power at thehigh frequency end are also found amongfour patients of this group, as shown in figure4(B). In all these cases, this finding was asso-ciated with chest pain. In figure 4(C), finally,spectra with moderately high power in thehighest ranges are shown. The spectra onwhich this figure is based came from six pa-tients, of whom four had heart failure andpain ; in two, pain was not a symptom. Inthe standard 12-Jead electrocardiograms, onlytwo from the group shown in figure 4(A) ex-hibited an ischeinic pattern; in those fromfigure 4(B), all exhibited nonspecific T-wavechanges only; and in those from figure 4(C),one showed an ischeinic pattern.

No figure is shown for the data of table 4,because it is practically the same as that ofhealthy, young, normal subjects.

It is also important to know the range ofvariation of a spectrum of a particular in-dividual after the lapse of some time. Meas-

2 - i

I -

2 - ,

LJ>

< I -UJ

30 4060 180

60120

90180

130260

180360

250500

350700

FIGURE 2Histogram showing: (A) the typical weightedpower spectrum of a normal, young adult (thefrequency bands are shown on the abscissa); (B)the medians and semi-interquartile ranges ofweighted power spectra from table 1—healthy,young adults (the former are shown as a hori-zontal base, the latter by the shaded areas).

urements were, therefore, made in intervalsof several days on a number of normal andabnormal subjects. The semi-interquartilerange for these repeat spectra of a singlesubject was one-half as large as the semi-inter-quartile range for normal subjects as shownin figure 2(B).

Discussion

When the first evidence was obtained'--7-s

that the electrical output of the heart con-tained a measurable amount of energy at fre-quencies which were cut off by conventionalelectrocardiographs equipment, it was natu-ral to speculate whether valuable informationmight be lost by the band-width limitationimposed by the usual procedure. It was alsoevident from Langner's work7-8- " that therelative amplitudes of high frequency contri-butions were very small. The "notching" or"beading" of the oscillographs record re-ported by Langner clearly indicates that thehigh frequency components that caused it hadamplitudes which were of the same order of

Circulation Research, Volume X, June 1962


Dow

nloaded from


874 FRANKE, BRATJNSTEIN, ZELLNER

TABLE 2

Power Spectra: Patients with Well-documented Myocardial Infarction

Range(c.p.s.)

Weightingfactor

# I# 2# 3# 4# 5# c# 7# 8# 9#10#11#12#13#14#15#10#17#18#19#20#21#22#23#24#25#26#27

30-60

1.60

1.520.621.440.470.641.810.860.881.030.630.910.750.600.741.311.030.910.300.531.340.830.360.610.841.261.270.83

40-80

1.93

2.250.432.340.721.142.961.981.191.400.610.970.710.641.272.551.731.620.390.061.991.220.610.771.062.181.431.02

60-120

2.54

2.680.581.772.060.884.406.110.912.060.520.780.930.741.353.013.002.300.440.572.171.731.060.741.512.801.002.30

90-1S0

3.54

1.221.000.741.850.653.547.610.632.800.720.430.880.590.690.882.122.440.520.491.621.750.650.863.841.841.262.13

130-260

6.11

0.601.700.801.581.494.74

10.30.662.990.950.660.911.051.100.692.721.710.560.521.462.060.631.062.931.451.562.34

180-360

13.0

0.952.110.930.942.078.836.900.615.251.070.790.821.251.530.852.501.010.790.401.382.120.730.963.852.281.412.90

250-500

34.0

0.544.940.920.781.86

11.44.880.754.151.110.840.881.122.210.832.641.050.660.331.272.011.091.511.553.081.275.46

360-700

81.0

4.901.081.341.51—

4.350.593.120.89—

0.941.10

—0.7.12.781.100.710.541.372.120.92—

1.412.860.943.24

magnitude as the line width of the record.This small, relative power of the high fre-quency part of the spectrum of the electro-motive force of the heart has led someinvestigators13 to conclude that it does notcontribute significant information. It must bepointed out, however, that the absolute mag-nitude of a signal is in no way a measure forits information content. The only relevantquantity in this respect is the signal-to-noiseratio. "With the power spectral method, wehave found components as high as 3,000 c.p.s.i.n some subjects, which were well above thenoise level.

Although it is not a priori admissible todiscard the high frequency part of the spec-trum, it is equally evident that somethingmust be done to prevent it from being maskedby the energetic low frequency components.

There are, basically, three avenues open toapproach this goal, mentioned briefly above,corresponding to the basic representations ofa stationary time series. (The electrocardio-gram may be considered as a stationary timeseries or a stochastic process in a mathemat-ical sense.) The first is the amplitude versustime representation, which is the same that isbeing used in standard electrocardiography.Langner has used it preferentially and hasovercome some of the limitations mentionedabove by means of expanded amplitude andtime scales. He estimates that visual inspec-tion of such an expanded record should revealsufficient detail up to 500 c.p.s. The ampli-fication is, nevertheless, basically limited inthis case by the Fournier component with thehighest amplitude and the dynamic range ofthe system.

Circulation Research, Volume A*. June 19GS


Dow

nloaded from



The autocorrelation function, which is theFournier transform of the power spectraldensity function, should at least be mentionedas a second possibility. Except for some spe-cial cases,14-]r> this approach would necessitatethe use of high-speed computers with largememories and, although attractive in theory,must be considered as not economical, at leastfor the present.

Another approach to the problem of dem-onstrating the small high frequency compo-nents is the use of electrical band filters. Withfiltering, the amplification of the high fre-quencies is limited only by the signal-to-noiseratio of the lead electromotive force in thatparticular band. On the other hand, filteringwill not only introduce the desired linear dis-tortion into the system but will alter at thesame time the phase relationship of the indi-vidual waves of which the time series is com-posed. No further loss of information will,therefore, result when the phase informationis discarded altogether by squaring and inte-gration. This is the approach we have usedin this work.

The resolving power of the filter method ofpower spectrum analysis is limited by thelength T of the available data. If the filterpass-band were too narrow, the stability ofthe recorded power spectrum would be toolow, i.e., large statistical fluctuations in therecord would result.12 In fact, the filter pass-band should be many times as wide as 1/T,or conversely, T should be large compared tothe smallest band width. But what is T inthe present case? We have found that prac-tically all the electrical energy in the higherfrequency bands is concentrated in the QRScomplex (see fig. 1) and that the interval iselectrically silent in all but the standard 0to 60 c.p.s. range. Since the interval is muchlonger than the QRS complex, no energy fromthe previous event is stored in the filters andintegrators when the next QRS pulse arrives.Each of the individual QRS complexes should,therefore, be considered as an individual, in-dependent record; the time T is the length ofthe average QRS complex.

The narrowest of the filter bands used is30 c.p.s. wide. The width corresponds to a


FIGURE 3Median and semi-interquartile ranges of weightedpower spectra. (A) Patients with old mijocardialinfarcts. (B) Patients recovering from acute vn/o-curdial infarctions. (C) Mixed group of patients(of the patients whose tracings fall in this pat-tern, some had old infarcts; some were recoveringfrom acute infarctions).

minimum record length T = 1/30 second;the band width is, therefore, sufficiently large.The relation between band width and recordlength becomes more and more favorable atthe higher frequencies, which are more im-portant from the point of view of the presentinvestigation. For the highest band, 350 to


Dow

nloaded from


876 FRANKE, BRAUNSTEIN, ZELLNER

TABLE 3

Tower Spectra: Patients with Ischemia Heart Disease but No Known Infarction

Ttnnpre(c.p.s.)

AVoiglitingfnetor

# 1# 2# 3# 4# 5# 6# 7# 8# 9#10

#11#12#13#14#15#16#17#18#19#20#21#22

#23#24

30-60

1.60

1.471.160.620.911.0!)0.820.691.150.640.820.650.800.920.831.411.240.791.511.210.890.421.171.040.77

40-80

1.93

2.661.340.501.141.470.821.621.980.701.130.530.551.551.192.14n zo

0.972.801.061.190.491.331.551.01

60-120

2.54

3.161.340.871.232.231.594.912.020.730.780.660.312.171.012.035.371.173.840.651.080.770.953.421.32

90-180

3.54

2.351.520.730.812.562.147.461.122.040.610.660.270.980.980.906.780.882.310.581.081.391.302.520.67

130-260

6.11

1.611.290.420.583.401.916.450.792.520.710.570.350.811.700.916.501.013.180.791.501.361.382.960.65

180-360

13.0

1.460.860.520.564.611.343.200.980.400.830.370.471.552.451.735.3S1.923.221.001.960.761.114.8.10.92

250-500

34.0

1.330.810.240.956.751.642.680.880.540.800.770.512.022.473.354.122.303.121.072.330.721.06

12.81.43

350-700

81.0

3.280.74

—

1.246.871.242.950.910.61—

0.61—

1.922.431.742.352 7°3.320.772.14—

1.1315.2

1.30

700 c.p.s., for instance, the minimum neces-sary record length is .1/350 second only.

As Langnev.10 has already pointed out, thepower spectrum analysis will not permit usto distinguish (at least if a single lead onlyis used) between several possible types ofQES complexes which all may produce highfrequency components in the spectrum, de-pending on the position of the heart vector.From a theoretical viewpoint, it would beideal to obtain the total energy output of theheart by averaging the potentials over thewhole body surface. This is practically im-possible though ; one should at least try totake account of the variation of position ofthe vector in the frontal plane. To accom-plish this, the average power in the standardleads was algebraically added (the powermust be added algebraically, in contrast tofield quantities, which are added vectorially).This procedure is admittedly a crude approx-imation, but jt is sufficient for our immediate

problem, i.e., to determine the significance ofhigh frequency components of the "heartgenerator." The use of the standard leadsappeared advantageous, also, for the reasonthat it eliminates uncertainty concerning theposition of the electrodes, is sufficiently sim-ple and universally accepted. It seems cer-tain that future research will come up witha better lead system. At present, there is nosufficient reason to prefer any of the manylead systems or lead fields that have beenproposed.

When the present work was begun, the au-thors were primarily interested in a deter-mination of the general significance of theinformation obtained when the frequencyrange of the electrocardiogram would be ex-tended. Mainly on account of the work ofLangner, one could anticipate an increase inthe power of the high frequency componentsamong patients having myocardial infarctsor coronary heart disease without infarction.

Circulation Rcscarchi «¥i J"1\C 1961


Dow

nloaded from



For tliis reason, these two groups were com-pared witli young, healthy normals. Later, agroup of. elderly, normal subjects with no evi-dence ol! heart disease was included in theinvestigation. It should be pointed out hereIhat it was by no means easy to find a strictstandard of normality which one could con-fidently apply to the choice of subjects. Ab-sence of evidence of heart disease, electro-cardiograph ic or otherwise, appeared not tobe enough. An increase in high frequencypower of the spectrum was found in a fewinstances in young adults and was associatedwith overweight, high blood pressure, and, inone instance, earcinoid. Because the numberol: such cases so far found was small, and theassociated pathology too diverse, no estimateof their significance is possible at present.However, the occasional appearance of suchabnormalities has led us to adopt rather strin-gent limitations in the choice of normal sub-jects. The age limits were set between 15 and25 years. A large number of our normal sub-jects were college students; hospital patientswere accepted as normal only if there was nosystemic disease (most of them were hospital-ized because of fractures or minor surgery).Applying the above-mentioned precautions,normal limits for the algebraic sum of spec-tral power density in leads I, II, and 111 wereestablished and are shown in figure 2(B). Inview of the great care which was exercised inthe choice of subjects, we do noi expect thatfuture work will greatly alter this standard.

Comparing the spectra obtained from pa-lients with myocardial infarcts with the nor-mal ones, one will sec at: once that there arethree classes of spectra. The first, shown infigure 3(A), is almost identical with that ofnormal subjects. The second, shown in figureH(C), is somewhat elevated between 40 and180 c.p.s. The difference, although easily dis-cernible by visual inspection of the individualgraphs (not shown) is not significant statis-tically, according to the Mann-AVhitney li-tes t . ' -* Moreover, the ratio of old to acuteinfarcts is the same in both groups, so thatthe spectra of figure 3(C) must be classifiedat present as a statistical variant of the nor-

Circulation Research. Volume X, June 196$

3 - i

8 06 0

1209 0

1801302 6 0

1803 6 0

2 5 05 0 0

3 5 07 0 0

3060

FIGURE 411 ist ogram, showing median and semi-interquartileranges. (A) First group of patients diagnosed ashaving ischemia heart disease without infarction(these spectra cannot be distinguished from thenormal). (B) Second group diagnosed as havingischemia heart disease without infarction (the highrelative pouter noted in this group was associatedin all instances with eotispicnous chest pain). (0)Third group diagnosed as having ischcinic heartdisease without infarction (this group is inter-mediate between those shown in A mid B, but dif-fers significantly from the group illustrated in A.

*Tliis test appears to be particularly suitable forthis type of data. A nonparametric test was chosenhcfiiuse it appears doubtful at present that thepower is normally distributed. The U-test is aboutequally as powerful as the t-test in rejecting thenull hypothesis.


Dow

nloaded from


878 FRANKE, BRAUNSTEIN, ZELLNER

TABLE 4

Power Spectra: Healthy Middle-aged Subjects

Range(cp.s.)

Weightingfactor

# 1# 2# 3# 4# 5# 6# 7# 8# 9#10

#11#12#13#14#15#16

#17#18

30-60

1.60

0.611.120.650.741.161.141.37\ oo

0.930.901.270.700.S90.901.120.S80.971.01

40-80

1.93

0.481.711.961.271.811.092.112.801.551.021.690.751.002.280.901.121.771.73

60-120

2.54

1.411.482.531.440.801.011.774.671.700.972.040.980.533.610.771.102.851.56

90-180

3.54

1.850.661.261.351.321.230.951.611.100.501.470.450.271.290.820.562.700.76

130-260

6.11

2.050.621.411.201.020.961.071.411.020.391.640.440.440.930.860.391.760.52

180-360

13.0

2.550.75.1.511.421.430.780.981.371.230.481.700.700.870.730.940.311.420.59

250-500

34.0

2.051.101.501.762.040.761.001.061.330.661.661.000.8S0.641.070.401.490.74

350-700

81.0

1.731.081.662.131.800.881.001.331.190.871.760.S1

0.720.920.501.620.78

in a I ones until more data will have accumu-lated.

The spectra shown in figure 3(B), however,are in a completely different class; there seemsto be no continuous transition between thisand the other two subgroups. All of the spec-tra of which figure 3(13) is composed areassociated with acute infarction. It would beinteresting1 to find out whether the spectraof this group were produced by the high fre-quency waves which Langner has found byhis methods. A direct comparison is, unfortu-nately, not possible because suitable recordingequipment was not at our disposal; however,there is some indirect evidence. For instance,in the paper by Langner and Geselowitz10

(table 1), so-called high-pass filter residualsare listed; they are the per cent output-to-input ratios of high-pass filters and are ameasure for the signal energy which was leftafter the low frequencies were cut off. In thistable, a few abnormals are listed, residual ofwhich in leads III and VL> is 3.5 to 6 per centof the original signal. These cases have incommon with our own findings the fact of afairly wide separation, in amplitude from therest of the subjects. Their number is rela-

tively .small, as is that of ours. It is possibleto compare the "residuals" of Langner withour own data of figure 3(B) by convertingour own data into amplitude ratios. Takingdata from the ranges between 250 and 700cp.s., an average of 6 per cent results. Thisis in good agreement with Langner's data. Itseems, therefore, that the data of figure 3(B)correspond to the more extreme cases of"beading" and "notching" which Langnerreports.

The findings in the group of subjects with-out infarction are similar to those for themyoeardial infarcts. They may also be di-vided into three subgroups. The first of these(fig. 4A) has again a normal spectrum ; thesecond (fig. 4B), an extremely high high-fre-quency content.

As before, there is also an intermediategroup, which is shown in figure 4(C). Whilethe intermediate group among the myoeardialinfarcts could be classified as a statisticalvariant of the normal, the population of figure4(C) differs significantly from the normal one(in fig. 4A) on the 0.0.1 level in the 350 to700 cp.s. band, according to the Mann-Whit-ney U-test.

Circulation Research, Volume A", June, ions


Dow

nloaded from


POWEE SPECTRUM ANALYSIS 879

The great number of acute cases amongthe patients with especially large high fre-quency power seems to indicate that it ismostly the acute phase during which thisphenomenon occurs, and that the high fre-quency content may again diminish duringthe course of healing. Up to now, it was notpossible to follow the course of the myocar-dial infarct from the beginning through theacute phase, mainly because a screened cagehas to be used for the recording, so that trac-ings cannot be conveniently done at the bed-side.

It was already pointed out above that nospectral abnormalities were found amongelderly individuals who were definitely freefrom cardiac disease. In view of this, it wouldbe interesting to know the time when spectralabnormalities first occur: before, during, orafter the onset of an attack. It Avould also beinteresting to know the effect of variants suchas exercise, hypoxia, and certain drugs. Toattack all these problems, modification of ourpresent equipment to allow records at thebedside will be necessary.

SummaryA method has been developed for recording

the power spectrum of the heart's electromo-tive force for the time interval correspondingto the QRS complex of the conventional elec-trocardiogram. Standards for healthy, youngnormals have been established. No significantchange was noted in a group of healthy, mid-dle-aged normals without heart disease. Threedifferent patterns, however, were found forpatients with ischeinic heart disease, with andwithout evidence of infarction.

References1. EINTHOVKN, W.: Different forms of the human

electrocardiogram :uid their signification.Lancet 1: 853, 1912.

2. REID, W. D., AND CALDWELL, S. H.: Research

in clcctrociircliogriiphy. Ann. Int. Mod. 7:3G9, 1933.

3. GROEDEL, M. F.: Das Elektrokardiogram. Dres-den and Leipzig, Theodore Steinkopf, 1934.

4. GILFORD, S. R.: Engineering aspects of biologi-cal research design. Paper given at A.I.E.E.-I.R.E. Conference on Electronic Instrumenta-

tion in Mechanics and Medicine, New York,November 29, 1948.

5. GILFORD, S. R.: High fidelity electrocardiograph}'.In Proceedings of Second Joint A.T.E.E.-I.R.E.Conference on Electronics in Mechanics andMedicine, New York, October 31, 1949.

G. DUNN, F. L., AND ROHM, W. D., JR . : Electro-cardiograph}-: Modern trends in instrumen-tation and visual and direct recording electro-cardiograph}'. Ann. Int. Med. 32: 611, 1950.

7. LANGNER, P. H., JR. : Value of high fidelityelectrocardiograph}' using the cathode rayoscillograph and an expanded time scale. Cir-culation 5: 249, 1952.

8. KERVVIN, A. J.: Effect of the frequency responseof electrocardiographs on the form of elec-trocardiograms and vectorcardiograms. Cir-culation 8: 98, 1953.

9. LANGNER, P. H., JR. : High fidelity eleetro-cardiography: Further studies including thecomparative performance of four differentelectrocardiographs. Am. Heart J. 45: 683,1953.

10. EVANS, W., AND MCRAY, C.: Lesser elcct.ro-cardiographic signs of cardiac pain. Brit.Heart J. 14: 429, 1952.

11. LANGNER, P. H., JR . : Further studies in highfidelity electrocardiography: Myocardial in-farction. Circulation 8: 905, 1953.

12. BLACKMAN, R. B., AND LUKEY, J. \V.: Measure-ment of Power Spectra. New York, Dover,1958.

13. SCHER, A. M., AND YOUNG, A. C.: Frequency

analysis of the electrocardiograph. Circula-tion Research 8: 344, 1960.

14. FRANKE, E. K., AND BRAUNSTEIN, J. R.: Powerspectrum analysis of the high frequency elec-trocardiogram. Paper read at the BiophysicalSociety Meeting, Boston, February 5, 1958.

15. FRANKE, E. TC, AND BRAUNSTEIN, J. R.: Auto-correlation analysis of the high frequencyelectrocardiogram. Paper presented at theEleventh Annual Conference on ElectricalTechniques in Medicine and Biology, Minne-apolis, November 19-21, 1958.

16. LANGNER, P. H., JR., AND GESELOWITZ, D. B.:

Characteristics of the frequency spectrum inthe normal electrocardiogram and in subjectsfollowing myocardial infarction. CirculationResearch 8: 577, 1960.

17. MANN, H. B., AND WHITNEY, D. R.: On a test

of whether one of two random variables isstochastically larger than the other. Ann.Math. Stat. 18: 50, 1947.

18. SIEGEL, S.: Non-parametric Statistics for theBehavioural Sciences. New York, McGraw-HillBook Co., 1956.



Dow

nloaded from


Ernst K. Franke, Dr. Ing., John R. Braunstein and David C. ZellnerStudy of High Frequency Components in Electrocardiogram by Power Spectrum Analysis

Print ISSN: 0009-7330. Online ISSN: 1524-4571 Copyright © 1962 American Heart Association, Inc. All rights reserved.is published by the American Heart Association, 7272 Greenville Avenue, Dallas, TX 75231Circulation Research

doi: 10.1161/01.RES.10.6.8701962;10:870-879Circ Res.

http://circres.ahajournals.org/content/10/6/870World Wide Web at:

The online version of this article, along with updated information and services, is located on the

http://circres.ahajournals.org//subscriptions/

is online at: Circulation Research Information about subscribing to Subscriptions:

http://www.lww.com/reprints Information about reprints can be found online at: Reprints:

document. Permissions and Rights Question and Answer about this process is available in the

located, click Request Permissions in the middle column of the Web page under Services. Further informationEditorial Office. Once the online version of the published article for which permission is being requested is

can be obtained via RightsLink, a service of the Copyright Clearance Center, not theCirculation Research Requests for permissions to reproduce figures, tables, or portions of articles originally published inPermissions:


Dow

nloaded from

http://circres.ahajournals.org/content/10/6/870

http://www.ahajournals.org/site/rights/

http://www.lww.com/reprints

http://circres.ahajournals.org//subscriptions/


study of high frequency components in electrocardiogram...

Documents