AU599407B2 - System and method for predicting ventricular tachycardia - Google Patents

Description

F-

is i 59940 7 S F Ref: 34481 FORM COMMONWEALTH OF AUSTRALIA PATENTS ACT 1952 COMPLETE SPECIFICATION

(ORIGINAL)

FOR OFFICE USE: Complete Specification Lodged: Accepted: Published: Class Int Class This (o incnt cdvru c, 1[ S:tii;i 49 and is c.sr; ct i'(r printing. j 0 p 6 Priority: Related Art: t p Name and Address of Applicant: Address for Service: Arrhythmia Research Technology, Inc.

1300 East Anderson Lane Austin Texas 78752 UNITED STATES OF AMERICA Spruson Ferguson, Patent Attorneys Level 33 St Martins Tower, 31 Market Street Sydney, New South Wales, 2000, Australia t 9 p9 P1 Complete Specification for the invention entitled: System And Method For Predicting Vent-icular Tachycardia The following statement is a best method of performing it full description o' this known to me/us invention, including the 5845/3 N b-- 00 0 a 0 0 0 0 0 0 SYSTEM AND METHOD FOR PREDICTING VENTRICULAR TACHYCARDIA This invention relates to electrocardiography and more particularly to an improved system and method for predicting potential ventricular tachycardia in a patient.

This application is a divisional application from Australian Patent No, 561744. AU 614-7-4 -spe if al J a t-hmod0 adaptive filtering to obtain the objects of the inve The present divisional applicatio attain the objects of the invention by use of highi rectlonal filtering preferably in conjunctioni ining the time derivative of the unidirectional high Sudden death from acute arrhythmia is a major risk in the first few hours after a myo::ardial infarction. During the first days, the incidence of ventricular arrhythmia is approximately 90%. The percentage of arrhythmias decreases considerably after the first several days but still presents a substantial risk to the myocardlal infarct patient.

Statistically, without treatment, approximately 50% of all inrarct patients will eventually die of ventricular arrhythmia.

o A reproducible and consistent ability to predict a patient's 2, popensty for lapsing into an arrhythmia is needed, Several investigators, employing signal averaging techniques, have detected, on the body surface, small amplitude high frequency potentials in the late QRS and ST-segments of electrocardiograms in patients and animals prone to S ventricular tachycardia., (Uther, et al.: "The Detection of Delayed Activation Signals of Low Amplitude in the Vector Cardiogram of Patients with Recurrent Ventricular Tachyi:ardla by Signal Averaging", in Management of Ventricular Tachycardia Rolo of Mexiletine, edited by E. Sandoe, at al., Excerpta Medica, Amsterdam, 1978, pp. 80-82). D's, Uther, et al.

found that these potentials did not occur in healthy, young people and suggested that they represented areas of delayed myocardial depolarization.

Obviously, If it car be shown that the high frequency signal In the late QRS of a myocardial Infarct patlentt is common to most, If no, all, Sinfarct patients who are subject to ventricular tachycardia, an important new diagnostic tool would be available. Technically, however, It Is extremely difficult to isolate, accurately, high frequency signals late in the QRS complex. A filter must be used to eliminate the lower frequency portions and 1o analyze the late QRS for high frequency content.

Unfortunately, substantially all filters employed In the prior art "ring" 0 aa 00 aY6 -00 a~ 0 Xrc T"'T G-i~6 Rii/ fi 6 14- 13369/83 I.rl$ 2 -2for a period of time after application of the relatively high energy, initial portion of the QRS waveform. This ringing effectively hides any low amplitude, high frequency portions in the QRS.

Some prior art systems dealt with ringing problems by plotting or displaying the waveforms at various resolutions and left prediction to the doctor. Another approach which has been utilized in the prior art is to reverse time filter the waveform to analyze the tail of the QRS, and then to forward time filter the waveform to determine the start of the QRS portion and therefrom to deduce the QRS portion's width. The filters employed in prior systems have predominantly been recursive, sharp cutoff filters.

In a large clinical trial supervised by Dr. Michael Simson, University of Pennsylvania, using an electrocardiographic analysis system, .o It was found that 92% of postmyocardial infarct patients who were subject "1'5 to ventricular tachycardia, did, indeed, exhibit a distinctive high o 0 frequency signal tall In their late QRS signal. This signal was present in 09 only 7% of post infarct patients who were free of ventricular tachycardias. In addition, it was found that a patient suibject to ventricular tachycardia will exhibit a QRS signal of substantially longer duration than patients without ventricular tarhycardla.

It is an object of the present invention to predict potential ventricular tachycardia with unidirectional time filtering while overcoming filter ringing problems.

In accordance with the present inventiona there is disclosed a system 25 for predicting ventricular tachycardia from an electrocardiograph input waveform detected from a heart by detecting a high frequency signal in the late portion of the QRS portion of the electrocardiograph input waveform, the high frequency signal In the late portion of the QRS portion being predictive of the propensity of the heart to ventricular tachycardla, said system comprising: means for converting the electrocardiograph input waveform to a digital signal corresponding to said electrocardiograph input wavefori; means for reducing a signal-to-noise ratio of the digital signal and producing a reduced noise digital signal; 14, IAX 4o /VTr 0 2A finite impulse response high pass digital filter means receiving said reduced noisc digital signal for non-recursively processing said digital signal in a single time order to produce a finite impulse response high pass filter output signal; and means for generating an output indication responsive to the finite impulse response high pass filter output signal, said output indication indicating a propensity for the heart to experience episodes of ventricular tachycardla.

@tt tt oO et 0 et ea a IAD/884o providing a preselected reference signal, comparing the filter outpu ence signal, and indicati iction of potential ventricular tachycardia In accordance with some of the illustrated preferred embodiments of the present invention, each of u patient's X, Y, and Z electrocardiographic signals are converted from analog to digital values, and stored, and are then processed to select only normal or typical QRS waveforms, The selected waveforms are signal averaged over several hundred beats to obtain a relatively noise-free composite QRS. The latter portions of the X, Y, and Z digital QRS signals are then applied in either forward or reverse time order to an adaptive finite impulse response high pass filter. The finite impulse response filter processing enables the ringing artifact to be eliminated from the filter's output, The resulting filtered outputs are combined to create a composite filtered QRS waveform, The last 40 (or so) milliseconds of the filtered composite is isolated and measured to obtain an indication of the level of high frequency energy content indicative of a 00o propensity for episodes of ventricular tachycardia. The overall QRS .20 waveform is also processed in the same time order to determine Its total 0 00 o duration which provides a second indication of a propensity for Ventricular 0 0 STachycardia.

0°:0 Further features and advantages of the invention will become more °o0. readily apparent from the following detailed description, when taken in conjunction with the accompanying drawings, wherein; 0. FIG. 1 is a block diagram of an electronics based embodiment of the invention; o o o ,0 FIG. 2 Is a block diagram of a microprocessor based embodiment of the cri invention;

S

t 30 FIG. 3 is a graph showing incoming waveform registration in i ccordace with one embodiment of tha present invention; FIGS. 4A-C are graphs Illustrating general finite impulse response °0 (FIR) filters; 0 FIGS, 5A-B are graphs Illustrating the filter output response to the illustrated inputs for an FIR filter corresponding to a high pass filter constructed usin the low pass filter of FIG. 4A; FIG. 6 is a graph illustrating an FIR high pass filter response corresponding to a high pass filter constructed using the low pass filter 3 13369/83 b i. ~i -il: -r-u~ted-: -usin t he ow. s P 1=ter of FIG. 4C; FIG. 7 illustrates one realization of an FIR filter; FIG. 8 illustrates a realization of an adaptive FIR filter; FIGS. 9-11 are graphs illustrating finite impulse response filters as utilized in accordance with one embodiment of the present invention; FIGS. 12-14 are graphs Illustrating finite impulse response filters as utilized in accordance with another embodiment of the presert invention; and FIGS. 15A-D are simplified flowchart of the overall orocessing flow utilized to predict potential Ventricular Tachycardla from an ECG input according to an embodiment of the present invention.

FIG. 16 shows certain waveforms derived from the output of a high pass filter for the interior portion of the QRS complex In a control group without Ventricular Tachycardia; FIG. 17 shows certain waveforms representing output derived from a high pass filter for the interior portion of the QRS complex in a control group with Ventricular Tachycardia; FIG. 18 shows digital filter outputs derived from published emperical data relating to a control group of patients without ventricular %29 tachycardia; FIG. 19 represents digital filter outputs derived from published emperical data representing signal processing In patients with ventricular tachycardia; FIG. 20 shows filter outputs of electrocardiogram signal, output from recursive sharp cutoff IIR filters derived from clinical tests on patients 0o having ventricular tachycardia; and u a FIG. 21 shows filter outputs of electrocardiogram signals output from 0 6 0 Srecursive sharp cutoff IIR filters derived from patients not suffering from (f&A ventricular tachycardla.

DESCRIPTION OF THE ILLUSTRATED EMBODIMENTS Referring now to FIG. 1, a block diagram of an apparatus constructed In accordance with the present Invention Is shown. Each of leads 10, 12 o and 14 Is a bipolar electrocardiographic electrode lead. The X electrodes 0 0 are applied to the patient's midaxillery line at the fourth intercostal space (under the left arm between the fourth and fifth ribs), The Y electrodes are placed at the superior aspect of the sternum and 'rie proximal left leg, The Z electrode Is a the "V2" position (left of sternum at the nipple ltne), and the other electrode is directly .RJE 26z 4 13369/83 1 1 posterior. Each of the respective X, Y, and Z leads (10, 12 and 14) is coupled to a respective one of three ECG amplifiers 90 (such as Analog Devices Model 283J isolation amplifier). The output of each amplifier is passed to a switch contact, through switch 103, and to a low pass filter 110. Filter 110 characteristically attenuates all signals above 250 Hz.

The output from filter 110 is fed to an analog to digital converter 120 which samples the incoming voltage every millisecond and converts it to a 12-bit binary signal (such as an Analog Device Ad572 used at a sample rate of 1,000 samples per second). The time segment outputs from A to C converter 120 are stored in the order sampled In storage means 130, such as on tape, disk, semiconductor memory or other electronic or magnetic storage means.

The X, Y, and Z ECG signals (100, 101, 102) are sequentially coneacted to the filter 110 and to A to D converter 120 by the operation of the switch 103. The filter 110 can be replaced by any of a number of preprocessing functions, The output from each is sampled for 133 seconds at the 1,000 sample/sec rate to obtain the necessary continuum of recorded signals.

The switch 103 can alternatively be a multiplexer for simultaneously 20 measuring the X, Y and Z leads signals during each ECG cycle. Furthermore, o« the switch 103 can be eliminated and three separate processing paths each having an analog filter 110, and an A/D converter 120 can be coupled to with each of the X, Y and Z amplifier outputs coupled to a respective "analog filter 110, and with each respective A/D converter 120 coupled to the secondary storage means 130.

In either of these alternatives, only a single 133 seconds period is required to record a continuum of recorded X, Y and Z leads signals.

Sometimes the leads of the ECG are multiphasic, noisy, or contain extra beats. It is therefore desirable to select the ECG lead for a reference which has the best unambiguous trigger and least abnormal output. Nhile any of the X, Y and Z leads can be chosen, experimentation has shown the Z lead to usually S 326z I nIo a IJaU3!o0 -Ll'l~"f;rrWer~~- i -6be the best.

The output from the Z ECG amplifier (90Z) is also coupled to an input of reference comparator 140. A bandpass filter (such as 8-40 Hz) can be inserted between the ECG lead output signal from the amplifier 102) and the reference comparator 140 to clean the signal. A reference voltage is coupled to a second input of the reference comparator 140, which sets the comparison level. When the QRS portion of the Z lead ECG signal appears on line 102 and passes through the reference voltage, the reference comparator generates a reference bit which is recorded along with the corresponding time segment output of A to D converter 26. This reference bit enables all QRS waves to be overlaid, one on another, for selection and averaging purposes. Alternatively, the reference comparator can use parameters in addition to or instead or voltage level. For example, the maximum or minimum slope can be used to establish the reference bit position. The .20 reference time is a common time from lead to lead Y, Z) and from ECG cycle to cycle.

The samples of the waveform are taken from the secondary storage 130 and put in a fast primary storage from which the editing function 150 is performed. The editing function works in conjunction with the feature o selection to discard waveforms which are nonstandard.

Alternatively, where a large primary storage is availabe, no secondary storage need be employed.

Particular features of the waveform are preselected as standard, such as by experimentation, and any waveform not meeting the standard is rejected by the feature selection means 160. All waveforms meeting the standard are averaged to reduce noise. Such edited, averaged waveform is then passed on via coupling node 155 to an adaptive digital high pass filter 170, preferably of the finite impulse response type, and to a peak finder 180 which locates the peak of the QRS complex of the -7- ECG input. The digital filter response is a function of the 'location of this peak in the illustrated embodiments, and does not require both forward and reverse filtering; either forward only or reverse only filtering is sufficient. The filter output is passed to the two subsystems 200 and 210. One subsystem, 200, provides means for determining the amount of high frequency energy in the tail of the QRS portion of the ECG input. The tail of the QRS section is first accurately determined, as will be described later, and then the energy in that tail is measured.

The filtered output 175 is analyzed by means 210 which determines the duration of the QRS complex from the signal at node :155. The QRS duration and high frequency tail content are correlated by decision means 220 with emperically derived standards to predict Ventricular Tachycardia. For example, the diagnosis can be made by thz decision and indication means 220 based on either or both the duration of the QRS complex and the high S 20 frequency energy at the tail of the QRS complex. If both these indicators are positive, indicating Ventricular Tachycardia, then the diagnosis is positive. If both are negative, then the diagnosis is negative. If one of these indicators is positive and 5 the other is not, then the decision subsystem 220 can Qu .l provide an indication of the conflicting data and new data can be taken for confirmation. The plotter or CRT can also be used so that a physician can look at the edited/averaged waveform being processed and make an independent judgment. If the peak finder subsystem 180 is not implemented in a particular embodiment,, then Shuman inspection of the edited/averaged waveform can be used to determine the peak of the QRS complex of the ECG input. It is not necessary to Wnow the precise location of the peak in order LO practice the present invention, anJ an approximate peak location is adequate.

In an alternative embodiment of FIG. 1, the U-LL-LL----L C~ 1 -8filter 170 can be comprised of first'and second filter means 170A and 170B, respectively. The signal output of editing means 150 is coupled to the input of a finite impulse response filter 170A, where the time points tl and t 2 are identified as the starting and ending point of when the input signal exceeds a preselected level (such as 50uV). A gain control signal is generated during the t 1 to t 2 time interval. The signal output of editing means 150 is also applied to filter means 170B, which can be a finite or infinite impulse response filter having a gain control input. The gain of filter 170B is attenuated by the gain control output i of filter 170A, thereby suppressing the gain of filter 4. 170B during the high amplitude high frequency portion of S 15 the QRS waveform. This reduces ringing artifacts and permits unidirectional filtering.

"Referring to FIG. 2, an alternate embodiment of the system of FIG. 1 is shown. The subsystem functions of editing, 150, feature selection, 160, digital 20 filtering, 170, peak finding, 180, processing for the amount of high frequency energy in the QRS tail, 200, duration of the QRS complex, 210, and the decision and indication means, 220, of FIG. 1 are implemented, in a computer system utilizing a computer program. The computer system may be comprised of a microprocessor based system, minicomputer, or other computer. There are several reasonably inexpensive microprocessors which will perform the necessary functions at adequate speeds.

A general description of the operation of the systems of FIG. 1 and 2 shall now be described. First, the z lead waveform of the ECG input is analyzed to establish a reference waveform. (Alternatively, either 'the X or Y waveform of the ECG input may be first analyzed to determine the reference.) It is analog filtered, sampled, digitized and stored (where digital filtering is used), compared to the reference voltage -9- (which has been previously predetermined), and a reference bit is generated at the time segment corresponding to the Z waveform value equal to the reference voltage. Each new waveform that is sampled is then processed using the refe:ence bit as thus described. Where 512 samples are utilized for each ECG cycle, each sample is stored in a slow secondary storage means 1,30 (or where available in a portion of a large primary storage) for each of the X, Y, and Z waveforms of the ECG input$ for later use by the system. The reference bit technique is one of several which may be used to eliminate erroneous waveforms from being utilized in analysis by the system.

Referring to FIG. 3, a graph illustrating one S 15 waveform registration technique is shown. Many i techniques are available for proper registration ot the *waveforms to one another from cycle to cycle. The following alternatives are given as examples.

The incoming waveform is compared with ore i' 20 voltage such as the peak voltage value VP and the time instant at which the incoming waveform equals the peak voltage is recorded. The other incoming waveforms are S shifted such that each is properly registered with respect to this one point.

The incoming waveform is compared with several referenne voltage$ and the time instance at which these voltages occurred on the incoming waveform are recorded, as illustrated in FIG. 3. Each incoming waveform is then shifted right or left until the match to these reference voltages is the best, As illustrated in FIG. 3, the solid line 250 is the new waveform and Vl, V2, V3, and V4 are reference voltages from the reference waveform 260 (dotted) which have been previously determined. By shifting the solid waveform to the left in the illustrated example, better match and proper registration is obtained for the illustrated example of FIG. 3.

C*

Y C i r_ i i 1 function 150, of FIG. 1, may be performed in hardware or software. Many techniques can be used. For example, template selection or signal averaging can be used.

Subsequent waveforms can be selected by adjusting DC levels and time shifts in both plus and minus directions relative to the template. Initially, a single beat, including a QRS, is accessed from the secondary storage and placed in a buffer register. The reference bit is here employed to grossly acquire the reference location points on the QRS. Starting with the reference bit and ending with a reference location 128 milliseconds thereafter, eight location samples are selected and stored. This process continues for four QRS counts, and S 15 enables the establishment of the initial template against which succeeding QRS signals will be tested.

After the fourth QRS signal is stored, the maximum and minimum voltage values for each of the eight voltage points on the four recorded QRS waveforms are S 20 tabulated and become the initial template.

9 Statistical analysis can be used to reject noisy signals from use in the template. Then, the next IQRS signal is selected, its eight voltage points are determined and stored, and each point is selectively tested against the stored maxima and minima to determine whether it falls within or without the respective values. If it is found that there is a mismatch in any one of the eight points, the signal is rejected as not J being a QRS or being some other artifact which is not of interest. If all eight points fall within the maxima and minima, the waveform is accepted as a QRS, and its 512 voltage points, spanning the -accepted QRS, are then averaged with the corresponding b12 poiits of the previously stored QRS signals, and the resulting averaged value stored in a buffer memory. This routine is repeated for 150 QRS's which are subsequently passed through the template, averaged, and thsn stored to i i i 1 accomplish a compos ite-aver aged QRS *wave for the X lead. The template voltage minimum and maximum test points can be updated during the processing to assure accurate QRS selection. The same procedure is then repeated for the Y and Z leads, and the averaged valoea for each of the composite Y and Z QRS signals also are respectively stored in the buffer meiory. The buffer memory can form a part of the oditing means 150, or can be provided as a separate means, and can utilize semiconductor, bubble, disk and/or other storage.

The above processing greatly reduce$ the noise inherent in the QRS signal by the square root of the number of averaged beats and provides three averaged D0 QRS waveforms which are relatively noisefree and suitable for subsequent processing. Approximately 150 beats per lead are signal-averaged and reccQrded. At this point, the recorded QRS waveforms oan be coupled to the remaining processing means and/or can be plotted out on plotter 190 (or 240) 'for examination by the physician, The plot also enables the physician to pick out the midpoint of the QRS for the sub'seqent filtefring step, zs a substitute or voc-ificotion 6o.k finder giczns 180.

Other alternative tecthniq s oim employed. For Eixamp3.ef ion Qrdier to t wavetorm for the averaging Vvocossl certt~-t! ot therefrene wvefrmcan be evalvated. Tbhec fQeU~res 0can be called the reference features. it the new waveform has features which are rot "stnlficontly" different from the ttfe-rence featkires thet-,j~ 18t s "'good" wave-Form and it is accepted fo aVwtqq; otherwise it is rejected. Some Of the reference geatures are: Value of the voll prespecified times; (ii) Values of the j th t$tne at which it occursl Y _ll _1-_II1. E-ii~iii -12- (iii) Reference times when'the waveform is at, exceeds, or is below a certain voltage level.

(iv) Sum of the samples of the X, Y, or Z lead inputs during each ECG cycle having an amplitude greater than a predefined threshold; Sum of the positive and negative values around a certain voltage level. For example, if Vn are samples, then 2:(Vn-V) C X(Vn-V)g (Vn-V) where f (Vn-V)=l if (V otherwise, f(V and g(Vn-V)=l if (V -V)1O o0 otherwise, g(V o (vi) A combination of features can als be used. These features can be referred to as F F 2 Fk k separate features). The reference features can be measured and denoted by values FRl,

FR

2 ,r F*R. A new waveform is then accepted for averaging if its features F i Fk are not vufficiently different from the reference features, i.e.

S* Accept the waveform if SIF Ri-Fil threshold, therwise reject it.

The threshold is preselected, based' on experiments. Different features can be given different importance by considering a weighted sume, i.e.

Accept the waveform if Wi *Ri-Fi threshold, otherwise reject it, where weights IWil are all positive. A feature that is considered most important, or which should be very close t, the reference features, should have a high weighc. The other features should have a smaller weight. All the accepted waveforms are averaged as described in the above section regarding signal averaging. Thus by averaging 150 waveforms, a -4.

to bohx 960 wt -V tcatloR of potential ventt ular LOchlcardi Is qlvent -13composite averaged QRS waveform X-lead is created.

Similarly composite-averaged Y and Z leads are created.

The next s\ep is to find the peak of the QRS complex and its sample number, or the time at ohich this peak occurs. This is the function of the peak finder means 180 of FIG. 1. Several algorithms for determining the peak exist. One of these is described hereinafter, as follows: Let samples occur at time t i having a corresponding value V.

Step 1: Assume the pea, is at t call if P.

Step 2: If V 2 V, is unchanged If V 2

V

1 then P=t 2

S

6 Step n: If Vn Vn-1 P is unchanged.

If V >Vn-l' t n Thus by making sequential comparison, the peak of the QRS wave (from node 155) for each of the X, Y, and Z leads are found. This peak can be used in controlling the operation of the digital filter; as will be desccibed hereafter in greater detail. As mentioned above, it is not necessary to precisely locate the peak, o- and therefore, the peak may be found by simple human observation of the averaged waveforms, either on a CRT or a plotter.

It is not required that both forward and backward filtering be utilized. Prior systems have used forward filtering to determine the beginning time point S* of the QRS complex, and used reverse time (backward) filtering in analyzing the QRS tail so that the high energy portion of the main QRS waveform does not spill over into the tail of the QRS (the high frequency energy content of which indicates a propensity for ventricular tachycardia). Bidirectional filtering was employed by prior systems because recursive, sharp cut off filters were used which exhibit significant ringing. The use of adaptive time varying, and/or Finite Impulse Response (FIR) filters can overcome this difficulty. Adaptive '4 -14- FIR filters also have much more flexibility and provide features which are difficult or impossible to obtain using the recursive filters.

A simple high pass FIR filter can be modelled as having an input sequence X I and an output sequence "Y where Y X {low pass filter output?. The low pass filter cao' be modeled as YnlP Xn 0.25 (X nl X nl), as shown in FIG. 4a; Ynl1/ 5 (Xn Xn 1 Xn Xn+1 X+2), for five samples, as shown in FIG. 4b; or a low pass filter whose impulse response is an approximation to the Gaussian function, which would also have the Gaussian function as its frequency response, as shown in FIG.

4c. The step response of the filter of FIG. 4c would be the Error Function, which is quite smooth with no ringing. (For great r detail on this see, for example, .r "Fourier Integral and iLs Applications", by Papoulis.) Many other FIR configurations exist, which can be chosen to approximate a given frequency response 20 characteristic. (For examples see "Digital Signal Processing" by Oppenheim and Schafer, Prentice Hall.) Referring to FIGS. 5A-B, a graph illustrating the filter- output (Y corresponding to utilization of the model from FIG. 4A, is shown for two input waveforms (X where the filter output Yn X n n 0.25(Xn Xn+) 0.5 X 0.25(Xn +Xn) It is clearly seen from FIGS. 5A-B that this filter exhibits almost no ringing. (For the problems encountered in using recursive sharp cutoff high pass filters see "Use of Signals in the Terminal QRS Complex to Identify Patients with Venrtricular Tachycardia After Myocardial Infarction," by Dr. Michael Simpson, in Circulation, Vol, 64, No. 2, and especially see FIG. 1 on page 237.) Referring to FIG. 6, the frequency response of Gaussian high pass filter is shown, corgesponding to the 0 -R Gaussian low pass filter according to the relationship: Gaussian High Pass Output 1 (Gaussian Low Pass Output). The Gausia~ high pass frequency response curve can be shaped by controlling the width of the Gaussian low pass curve. In practice, the filter response must be truncated to finite terms. An FIR approximation for the Gaussian high pass filter can be readily constructed in a straightforward manner, The realization of the FIR filters is generally represented as Yn A *Xn A *Xn- n n 2 n1n Graphically, this is illustrated as shown in FIG. 7, for n A X A 2

A

3

X

n n 2 n-l 3 n+1' This filter can be constructed in a straightforward manner in hardware or software. FIR filters can also be used adaptively for time-varying filtering, where the coefficients can be changed as a function of the input v;t and/or output signal, or as a function of time. Thus, the filter response'car change to attain optimum response for a given input function. For example, where the filter is output is of high amplitude and is not changing in o wavy manner, the filter response can be t adapted to have a sharper high pass cutoff, thus making S detectioh pore reliable. An illustration of an adaptive FIR high pass filter is shown in FIG. 8.

FIR filters can be implemented simply as a weighted average,. Thus if £In is input sequency and Q\ is the output sequence, then a *I n aLa m n-m where a are preselected coefficients, Thus, the m filter uses samples onr both sides of the siample "n" to obtain the filtered output. The filter spread is said to be (2,L since (21 input samples are used to derive an output. Choise of coefficients determines the characteristics of the filtar. Seceral methods of design of these coefficients exist. Digital filters are well-known in the art, and will not be discussed in

L

-16general terms in any substantial detail herein.

Reference is made, however, to two recognized works Digital Signal Analysis by S.D. Stearns, Hayden Book Company, Inc., (1975) pp. 182-222; and Digital Signal Processing by Oppenheim and Schafer, Prentice-Hall, Inc., (1975) pp. 195-282], the contents of both of which are incorporated herein by reference.

The aforementioned excerpts teach, in detail, the methods for designing various digital implementations of analog filters. Another reference book teaching similar methods is Theory and Application of Digital Processing by Rabner and Gold, Prentice-Hall, Inc.

It is straightforward to design a certain set of coefficients which yields the desired proper 15 characteristics, i.e. adaptive high pass filter characteristics. In order to avoid "spill-over" (e.g.

Sringing effects) in the tail of the QRS portion of the ECG input from affecting adjacent samples (which in turn affect the filter output), adaptive filtering in accordance with the teachings of the present invention can be used, as illustrated in FIGS. 9-11. Either S. forward only or reverse only (backwards in time) filtering direction can be used. If so desired, a combination of forward and reverse direction filtering 25 can be used.

Assuming for sake of illustration that the reverse only filtering direction is selected, waveform analysis would commence at the right side of the waveform, sample number n 512 (end of waveform), where n a running index of the current sample number.

(forward filtering would commence with n=l).

The peak of each ECG input waveform is determined for each ECG cycle as described above herein. The filtering can then proceed as described below.

Referring to FIG. 9, a sample waveform to be filtered is shown. The filter output can be defined as I r" -17- +4 Q a

I

I where the filter input sample spread is defined as 2-t 1.

(For either direction of filtering, In+m can be replaced by In-m.) For filtering of n>p i.e. n p +4 1, 512, then 0 for n m=512. For filtering of p a portion of the input samples used by the filter in deriving an output occur on the n(p side of the input waveform, which should not be allowed to change the filter output for p +4>n>p.

To prevent the samples n p from affecting the filter output for p the waveform is effectively extended by a constant for n p such that In+m= p for all n m p, as shown by the dotted line in FIG. 10. The resultant waveform produces a 15 filter output whenever oh n p but n p This has the effect of eliminating "spill-over" from the left-hand side (n p) to the right-hand side (n p) of the input waveform. Since the filter is high pass, the extension of the waveform by a constant has no effect on thb filter output. In a similar manner, for filtering o of n p, the waveform is extended by a constant on the t rht-hand side of the peak p, such that I.+ m

I

n+m p S> p, as shown in FIG. 11. This has the effect of e-iminating spill-over from the right-hand side (n p) to the left-hand side (n p) of the input waveform.

The net effect is that ringing artifacts are reduced to almost zero, while complete filtering action is preserved. The reference characteristic need not be limited to the peak p, and can be chosen according to desired filterinq characteristics and known waveform criteria. Additionally, other adaptive and finite impluse response high pass filters can be constructed in accordance with the teachings of the present invention.

Referring to FIGS. 12-14, a preferred alternative high pass filtering embodiment of the present invention are illustrated where two reference electrodes are placed at the superior aspect of the sternum and ne proximal left leg. The Z electrode is a the "V 2 position (left of sternum at the nipple line), and the other electrode is directly r ra A jj RJE' EP26z

.I

-4- 13369/83 I ~Cr I I~C~r I

C

4r 4*, r* e 4 *4 4 4*t 4( I 4 411L 4 sample position numbers, Al and A2, are chosen as waveform extension breakpoints, in addition to determination of the peak position reference number p.

The filter output is designated Qn, where Qn m I n+m Referring to FIG. 12, filtering for n>p A 2 is shown, where the sampled input waveform is extended by a constant such that In+m= i A2 for n+m<p+A2.

Referring to FIG. 13, filtering for p Z <n< p+AL2 'is shown, where the sampled input waveform is extended by a constant, twice, such that In+m= I 4 for n+m>p+A2 and In+m=I -A for n+m(p Al.

Referring to FIG. 14, filtering for n p 6 is shown, where the sampled input waveform is extended by a constant such that I I for ii+m p-~Al n+m,>p 61 Since the filtered values inside the QRS complex portion of tts ECG input are not of much 20 interest in the prediction of the ventricular tachycardia, the above described filtering technique will work extremely well. The values for A 1 and 6 should be experimentally determined to insure best results.

Many other alternatives exist other than those illustrated in FIGS. 12-14. For example, if filtering is desired for p A 2 then the sampled input waveform is extended by a constant, twice, such that In+m I P+A2 for n+m>p 2 and In+m I P for n+m<p.

Again, experimentation can determine which technique will work best for the particular test equipment utilized and the particular data characteristics which are to be analyzed.

Qa Returning again to FIG. 1, the filter 170 provides a filter output 175 in accordance with the teachings of the present invention, The filter output I I jLi i 1; I _I -19- 175 is coupled to means 200 for processing the filter output to determine the amount of high frequency energy in the anterior portion of QRS complex (QRS tail), The portion of the filtered QRS which corresponds to the late section containing the potential high frequency energy of interest must be located. This can be achieved by first selecting a 40 millisecond sample substantially after the termination of the major portion of the QRS t 300ms to t 260ms) and averaging the Vn values to achieve an average noise voltage for that sample. That average noise is stored, and a standard noise level deviation is calculated employing the following equation: n=300 n=300 2

F.

0 15 Standard deviation =V 2 V z n T n 0o n=260 n=260 0 040 0 0I 0 0 V.4t0.40 0o 39 This standard noise deviation is stored, and a millisecond sample of the QRS is selected from t 250ms to t 255ms). The average value of the time o segment voltages from t 250ms to t 255ms is calculated and compared to the average noise level plus three standard deviations (as previously determined) If the calculated value for the 5 millisecond sample does not exceed the total, the time segment is decremented by one time slot one millisecond), 25 and the process repeated until the selected average voltage of the smaple does exceed the level of the calculated noise plus three standard deviations. This occurrence indicates that the selection process has arrived at the termination of the QRS signal the middle time segment of the 5 millisecond sample is defined as the end of the QRS.) In order to determine whether the QRS signal has or does not have the high frequency tail referred to

L

above, the voltage sample in the middle time segment (t of the 5 millisecond sample is then selected as well as the next lower 39 voltage time segments from t 255 to t 186). The root mean square value of all of these voltages is then calculated as: V, 2 s t)+V n 2ts Vn 2 (t- 40 The RMS voltage of the 40 ms sample is then compared to 25 microvolts, and if it exceeds microvolts it is indicative that the patient is not susceptible to ventricular tachycardia whereas, if it is less than 25 microvolts it is indicative that the patient is subject to ventricular tachycardia. Medical o, researchers have found that the high frequency component found in patients with ventricular tachycardia extends S 15 thetail of the QRS by several tens of milliseconds, but at a relatively low level. Thus, a low level measurement indicates that there is a low level, high frequency tail of energy appended to the QRS. If the voltage exceeds the 25 microvolt level, it is indicative 0 t 41 l 20 that, in lieu of there being the aforementioned tail of Shigh frequency energy, the measurement is actually being Smade on the major portion of the QRS signal which has high levels of high frequency energy. The results of these tests can be displayed or printed out as shown in 2 5 FIG. 1 for the physician's use. Parameters such as milliseconds, 5 milliseconds, and 25 microvolts should be optimized for the selected filter characteristics.

Also, squares used in the V RMand standard deviation equations can be replaced by absolute values.

The filter output 175 is also coupled to means 210 for determining the duration of the QRS portion of the ECG waveform. The width of the QRS waveform has been found to have a relationship to a patient with ventricular tachycardia. In order to measure the width of the QRS in the. above system, it is sufficient to 'if- -21obtain an indication of the beginning of the QRS waveform, as the end of the QRS has already been determined. The initiation of the QRS is calculated in much the same m-n.er. In specific, from t 40 -o t 1, a 4C(-millisecond sample of noise measurements is averaged, and the standard deviation calculated. Five millisecond values are then selected and tested to determine whether the average value of each sample exceeds the average noise plus three standard deviations. For the 5-millisecond sample which does exceed ti.at level, the beginning of the QRS is then defined as the middle time segment of that segment. The duration of the QRS then stretches from the middle of that segment to the end of S" 15 the QR'S as defined above. Again squares can be replaced by absolute values.

SReferring to FIG. 15, a flow chart of the overall processing flow utilized to predict ventricular tachycardia from an ECG input'in accordance with the illustrated embodiments of the present invention is shown. 'he starting point of the flow chart is based on the assumption that an electrocardiograph has been a attached to a subject patient. The X, Y, and Z ECG leads waveforms are periodically sampled during each of t 25 a pluraltiy of ECG waveform input cycl's, each ECG cycle corresponding to a heart beat, as shown in box 500.

Separate samr,'ling is done for of the X, Y, and Z input leads. The samples are filtered to reduce noise artifacts, as shown in box 502, and the filtered samples are digitized as shown in'box 504. The digitized samples are stored in a secondary storage, as shown in box 506. A beat count is initialized at 0, as shown in box 508, which tracks the number of beats utilized in subsequent steps in averaging and dotermining waveform features. The digitized X lead QRS ECG waveforms are accessed from the secondary storage, and the beat count is incremented, bpx 512. Selected features of the -22accessed waveform are computed, box 520, and the values of the reference sample points for the selected features are stored, box 530. The beat count value is tested, as illustrated for a count of 4, at box 540, to track the number of input waveforms utilized in obtaining the averaged values. If the beat count is not yet 4, or whatever other number may be determined as desirable, the process returns to step 510 and continues to access waveform samples. If the waveform beat count is equal to 4, then the process continues at box 550, where the values of the features for the 4 beats are averaged to determine reference features.

Next, another QRS X lead input is accessed from secondary storage, and the selected features are computed for it, box 560. If the computed features as r t determined at box 560 are sufficiently close to the reference features as determined at box 550, then the waveform accegFed at box 560 is utilized as an acceptable input. 'If the 'features of the sampled waveform as determin-d at box 560 differ significantly from the reference featues as determined at box 550, then the signal waveform accessed at box 560 is rejected as an invalid signal, and processing returns to box 560 for accessing of another QRS X lead input. Where the sampled waveform of box 560 is found acceptable by decision logic as shown at box 570, processing proceeds at box 580. The QRS voltage values of all accepted QRS waveforms for the X lead are averaged, box 580, and the .a averaged values are stored, box 590. The steps from 560 to 590 are repeated for one hundred and fifty (150), or whatever other number is determined desirable, QRS X lead waveform inputs as shown at box 600. Next, the steps from 508 to 600 are repeated, for the Y, and then Z, QRS lead waveform inputs, Thus, as the system completes the process to the step of box 610, averaged X, Y, and Z waveform samples are provided for a single averaged ECG cycle. For reference sake, the averaged .I Li I i. -I -23samples are designated X Yn, and Zn, corresponding to the averaged X, Y, and Z, waveforms, respectively, where n equals the reference sample point within the given ECG cycle, such as n 1 to 512, as shown at box 620.

Next, the peak value of the averaged waveform samples is determined, first for the X averaged waveform, as shown at box 630, and then for the Y and 2 averaged samples, as shown at box 640. Then, the average value of n, designated np at which the peak of the X Yn and Z averaged waveforms occurs is n. n n determined, as shown at box 650. As described herein with reference to FIGS. 12-14, L1 and 6 2, are selected, as shown at box 660. Next, the filter output, 15 Qn' is obtained from each of the averaged input samples X Yn and Z respectively, and separately, as shown at boxes 670 to 750.

The filter output is obtained by adaptively filtering the averiged input waveform samples in a 20 sectional manner. tihe sectioning is done based on the reference sample position, n, where the input tXn is modified to eliminate ringing effects outside the section being analyzed. This technique is described herein with reference to FIGS. 9-14. For example, as 25 illustrated at box 670, to obtain a filter output for the section of the input sample waveform of n >p+A2 a ba 'new input X is created, where X' X X, the original input waveform average sample values, for n n +A2, thereby maintainig the integrity of the p filtering action for those sample values, while creating in an adaptive manner a new input X' 1

X

n p for all values of n<n p+2.

The next step is to initialize the value of n for the filtering action. Where forward filtering is used naO is the initialized value. Where backwards filtering is used (reverse time filtering) n is initialized at n 4 5U.2as shown at box 680. A filter 1 -24output Q is obtained for the section n n '2.

as shown by box 690. The filtering action is implemented by decrementing n for backward filtering or by incrementing n for forward filtering. The filter output Q O is the summation of the filter ottput for the combination of input samples utilized in obtaining the output. The next step is to obtain a filter output for a different section of the averaged input waveform. For obtaining a filter output for n -AI 1 n<n +a2, as shown at box 700, a new input is created Xn" X for nnp 6 2, and X Xp+ to" for n n A 2. Finally, a filter output is S' obtained for the section of the averaged sample input waveform for n<n1 l, as shown in box 720. This is done by creating a new input which is adaptive to the section of the averaged input waveform being sampled by .the filter, such that X n for nnp A, and Xn for n>n A The net o n- 0, tt filter output for the three 'sections of the averaged X 20 lead sample inputs is equal to the sum of the filter outputs as derived at boxes 690, 710, and 730. The filtering steps 670 to 730 are repeated for the Y lead averaged sample inputs and then for the z lead averaged sample inputs to derive outputs Qn and Q 25 as shown ab box 740.

A composite filter output is then computed as V n n (X)I +|on and the composite filter is output designated V n as shown in box 750, An average noise nA300 voltage value V is calculated as V 1 0 and n-260 resultant value V is stored, as shown at box 760.

The use of samples 260 to 300 is based upon experimental r i i 1 h data published in medical periodicals indicating that samples 260 'to 300 represent the tail of: the QRS waveform. The use of these samples numbers, and of the sample total, i 1 n computing the average noise voltage can be changed as determined by exper~mental results.

Next, a composite 3randard deviation i~i computed, equal to CSD /40 1/2/4 The resulting composito standard deviation is then stored, as shown at box 770. Where samples :260 to 300 are utilized f'or boxes 760 and 770, a 5 millisecond segni,-nt of Vis averaged and called V f with n starting at nr- 300, with n decreasing for reverse filtering. The vajue as determined at box 780 is n compared to a reference level Vref 3XCSD, I 4S n shown at box 790, and it the average value Vnis less than V ref, then the value of n is decremented, and a' new 5 millisecond segment is selected. Where the value 0 of the averaged 5 millisecond Segment is greacer than V Aa&2 0 refe processing proceods at box 810, and voltagc sample is selected in the middle Of the 5 Millisecond 000 segment, plus in the middle Of the next 39 time segment sap.The RM voltage V is computed efrc level rmss as determined by experimentation, such as 25 microVoltS, as shown at box 830, said reference level being a threshold indicative of ven~tricular tachycardia. where the RMIS voltage is greater than the reference level, 'indicating the abse*nca of potential ventricular tachycardil, at! appropriate indication Is given as shown at box 870. Where the RMS vo).!agje lis less, than the reference level, then procoitsinq proceeds from box 830 C _l _I is i) c ou

D

o oil o o o o j J~ a a

I

i pc s

P

O

ii

C

to box 860 wh, 'ication of potential ventricular tachycardia is given, In parallel to steps 750 to 830, the duration of the QRS complex of the filtered ECG input waveform is derived from the filter output, as shown at box 840. The QRS duration is then compared to a thKi'shold value T, as shown in box 850. which reference value is a threshold for detecting potential ventricular tachycardia. Nhen the QRS duration is determined to be less than the threshold, ventricular tachycardia is indicateu, and proper indication is provided to the user of the system as shown at box 860. Nhere the QRS duration Is greater than the threshold T, an indication 10 of no potential ventricular tachycardia is provided as shown at box 870.

Nhere there is a positive indication of poteitial ventricular tachycardia provided by the RMS voltage decision box 830, and a negative indication of potential ventricular tachycardia from tne QRS duration decision box 850, or, vice versa, an indication is provided to the user of the conflicting 15 measurements, and the measurement steps are repeated again, starting at step 500, as shown at box 880.

The flow chart and description of FIGS. 15 A-D are meant to be illustrative only, and should not be construed In a limiting sense. Many other processing flows could be utilized, and, as described elsewhere within this specification, alternative techniques of determining high frequency low amplitude energy from the noise content of the sampled waveform exist (such as utilizing a method different from the 5 millisecond sample stepping window technique). Other alternative means of indicating potential ventricular tachycardia and conflicting QRS duration and Vrms data can be provided. Additionally, other reference parameters can be utilted in addition to or in place of QRS duration and Vrms voltage of the high frequency contents in the tail of the QRS complex portion.

RJE/Z326z 26 13369/83 r L -27- Heretofore, the use of recursive sharp cutoff filters has been ineffective in unidirectional filtering of the late QRS portion of the ECG to predict VT, due to ringing and masking problems.

In accordance with one aspect of the present invention, the filter output of er a high pass, recursive, sharp cutoff IIR filte. an FIR high pass filter or an adaptive high pass filter is analyzed by taking the time derivative of the late QRS portion of the ECG, in either a forward only or reverse only direction, or bidirectionally. A resultant derivative waveform "reference" pat'--rn is then established for two co;.trol groups non-VT (see for example FIGS. 16B and o n B 18B and 20B) and VT (see for example FIGS. 17B and u~o 15 19B and 21B). In one embodiment the first time derivative is used. Alternative embodiments utilize second and higher order time derivatives. Additionally, other linear and differential techniques can be used to o ~transform the signal and retain the information for 0 00% 20 analysis. Furthermore, other pattern recognition techniques can be utilized in analyzing the signal.

j Additionally, where two or more linear processes are utilized, the result ci processing is the same regardless of the order. Thus, filtering and 25 differentiating can occur in either order. Each control group will exhibit a unique reference pattern easily distinguishable from each other.

Even the ringing artifacts of the sharp cutoff recursive IIR filters result in derivative waveforms which have distinguishable VT and non-VT reference patterns. Derivative analysis is best implemented where the signal waveform being differentiated exhibits a high signal to noise (S/N ratio).

It is well known in the art that noise can add extraneous components to a signal waveform, which can cause problems. It is therefore desirable to maximize the S/N ratio. All of'the waveforms illustrated herein LI t -28are clean with good S/N ratios.

Analysis of the derivative waveforms can be done by numerous techniques, such as evaluating the number of positive and negative transitions, evaluating the area of positive and negative pulses about a reference line, tc.' The present invention can be practiced in analog or digital form. However, digital filtering provides cleaner filter output and allows the use of signal averaging, templating, and other techniques.

Finite and/or infinite pulse response filters and/or adaptive filters can be utilized with the present invention.

The repeatability, reliability, and statistical accuracy of VT prediction using this invention will require clinical verification.

Referring -now to FIGS. 16A and 17A, certain waveforms are shown representing the output of the high V't pass filter for the interior portion of the QRS complex in a control group without ventricular tacnycardia (FIG.

idA) and patients with ventricular tachycardia S representing the second control group (FIG. 17A). These outputs are obtained by constructing a digital filter having a Butterworth high pass filter response, and 25 utilizing signal averaging prior to filtering in a reverse direction.) The waveforms of FIGS. 16A and 17A represent clinically derived data which was published in a medical periodical, Circulation, Volume 64, No. 2, SAugust 1981, pages 235-242, entitled "Use Of Signals In The TermirLal QRS Complex To Identify Patients With Ventricular Tachycardia After Myocardial Infarction", by Dr. Michael Simson, and specifically page 238, Fig. 3. These curves were obtained by signal processing in patients with anterior myocardia infarctions. FIGSo 18A and 19A similarly represent digital filter outputs derived fromf empirical daL. published in the 0 Circulation publi.cation citea above, at page 238, Fig.

-29- 4, representing signal processing in patients with inferior myocardia infarctions. FIG. 18A represents a control group of patients without ventricular tachycardia, while FIG. 19A represents the second control group with ventricular tachycardia.

Referring now to FIGS. 16B, 17B, 18B, and 198, the first time derivative of the respective curve of FIGS. 16A, 17A, 18A, and 19A, are shown. Comparing the curves of FIGS. 16B and 17B, it is seen that the first time derivLtive waveforms of patients without, and with, ventricular tachycardia are substantially different in terms of the number of transitions, the duration of S. pulse widths, and the relative position of transitions.

Additionally, the area under the pulses differs for the two curves. Thus, the prediction of ventricular tachycardia can be made in a straightforward manner from the derivative waveform curve either by human analysis or by electronic hardware analysis or by digital Sanalysis via computer.' Referring to FIGS. 18B and 19B, it is again seen that the first time derivative of the filtered ECG ORS complex waveforms provide readily C distinguishable waveform patterns making detection of people with ventricular tachycardia (FIG. 19B) from those without ventricular tachycardia (FIG. 18B) a relatively straightforward task. Again the number of transitions, the time position of the transitions, the pulse widths between transitions, and the areas displaced by the pulses are readily distinguishable between the resultant derivative waveforms of FIGS. 18B and 19B.

Referring to FIGS. 16C, 17C, 18C, and 19C, the second time derivative waveforms of the filtered outputs of FIGS. 16A, 17A, 18A, and 19A, respectively, are shown. Referring to FIGS. 16C and 17C, it is seen that the non-VT (FIG. 16C) versus.the VT (FIG. 17C) second time derivative waveforms differ significantly, and provide an easily distinguishable mechanism by which -i i i i i -C ventricular tachycardia can be predicted. The amplitude, time position, and number of transitions differs between the two waveforms of FIGS. 16C and 17C, and can be easily analyzed by a human, electronic hardware, or a digital computer system. Referring to FIGS. 18C and 19C, it is again clearly seen that the second time derivative of the filtered waveforms of FIGS. 18A and 19A, respectively, yields readily distinguishable characteristics for analyzing non-VT (FIG. 18C) versus VT (FIG. 19C) patients. The waveforms of FIGS. 18C and 19C are readily distinguishable based upon the number of transitions, the amplitude of the transitions, and the time position of the transitions.

oIt can now be understood that utilizing the first or second (or higher) time derivative of the filtered waveform yields an output by which patients with and without a propensity for ventricular tachycardia can be distinguished and analyzed.

r Referring to FIGS. '20A-B and 21A-B, the filter 20 outputs of electrocardiogram signals as output from 4' 0 recursive sharp cutoff IIR filters is shown. FIGS. J and 21A show a high resolution filter output while FIGS.

and 21B show the low resolution filter outputs.

FIGS. 20A and B represent waveforms obtained from clinical tests on patients having known episodes of chronic ventricular tachycardia, while FIGS. 21A and B represent waveforms obtained from clinical data of patients with no known episodes of chronic ventricular tachycardia but who had chronic ventricular arithramias. The portion of the waveform marked in FIG. 20A represents the delayed waveform present in patients having known episodes of ventricular tachycardia. This data and these curves were published in the medical publication Circulation, Volume 63, No.

5, May 1981, pages 1172-1178, at page 1175, entitled "Body Surface Detection Of Delayed Depolarizations In Patients With Recurrent Ventricular Tachycardia And Left Ventricular Aneurysm", by John J. Rozanski, et al. FIGS. 20C and 21C represent the first time derivative of the respective curves of FIGS. 20A and 21A for the relevant portion of the waveform in the time vicinity of the delayed waveform D. As seen by the arrow in FIG. 21A, there is no delayed waveform D occurring at FIG. 21A.

Referring now to FIGS. 20C and 21C, the first time derivative DX/DT of the waveforms of FIGS. 20A and 21A, respectively, are shown. It is clearly seen that the first time derivative waveforms of FIGS. 20C and 21C are distinguishable from one another, each having a unique reference pattern. The number of transitions, time position of transitions, and area displaced by the o eQ pulse width between transitions differs for each of the waveforms of FIGS. 20C and 21C. Thus, it is straightforward, in the manner as discussed above, to differentiate between patients exhibiting a propensity for ventricular tachycardia'as in FIG. 20C versus %120 patients not indicating a propensity for ventricular tachycardia, as shown in FIG. 21C. This can be done by Rhuman analysis, or it may be done by electronic analog or digital techniques. Additionally, second and higher order derivatives of the filtered waveform may be utilized and analyzed to distinguish the waveforms of VT and non-VT patients.

Although this invention has been described with reference to the illustrated embodiments, this description is not meant to be construed in a limiting sense. Various modifications of the disclosed embodiments, as well as other embodiments of the invention, will become apparent to those skilled in the art upon reference to the drawings and description of the invention, It is therefore contemplated that the appended claims will cover any such modifications as fall within the true scope of the invention.

i

Claims (4)

1. A system for predicting ventricular tachycardia from an electrocardiograph input waveform detected from a heart by detecting a high frequency signal in the late portion of the QRS portion of the electrocardiograph input waveform, the high frequency signal in the late portion of the QRS portion being predictive of the propensity of the heart to ventricular tachycardia, said system comprising: means for converting the electrocardiograph input waveform to a digital signal corresponding to said electrocardiograph input waveform; means for reducing a signal-to-noise ratio of the digital signal and producing a reduced noise digital signal; finite impulse response high pass digital filter means receiving said reduced noise digital signal for non-recursively processing said "o digital signal In a single time order to produce a finite impulse response oo high pass filter output signal; and means for generating an output indication responsive to the 0 00 finite impulse response high pass filter output signal, said output 0°o. Indication indicating a propensity for the heart to experience episodes of ventricular tachycardla.

2. The system as claimed in claim 1 further comprising: means for signal averaging the digital signal prior to processing by said filter means,

3. The system as claimed in claim 1 further characterized in that said filter means is a Butterworth filter.

4. The system as claimed in claim 1 further characterized in that said filter means is an adaptive high pass filter. SO4: DATED this TWENTY-THIRD day of APRIL 1990 0o Arrhythmia Research Technology, Inc. Patent Attorneys for the Applicant SPRUSON FERGUSON IAD/884o