WO2010018608A1

WO2010018608A1 - System for detecting pacing pulses in electrocardiogram signals

Info

Publication number: WO2010018608A1
Application number: PCT/IT2009/000374
Authority: WO
Inventors: Paolo Banelli; Alessandro Polpetta
Original assignee: Universita' Degli Studi Di Perugia
Priority date: 2008-08-14
Filing date: 2009-08-07
Publication date: 2010-02-18
Also published as: ITPG20080041A1

Abstract

A method and a system for detecting pacing pulses in electrocardiogram (ECG) signals are presented. The method, and system are able to detect pacing pulses even when the ECG waveforms have low signal-to-electromyographic-noise (EMG) ratios. The system can be, in an exemplary embodiment, a part of a generic electrocardiograph. Shortly, in a first step, in order to prevent pacing pulse widening and thus their superimposition to physiological ECG components, the anolog ECG signal of a patient having a pacemaker is digitalised employing a high sampling rate,- thus, in a second step, the method detects the pacing pulses by means of a non-linear filtering (potentially after a linear filtering step) whose output is compared with a threshold.

Description

SYSTEM FOR DETECTING PACING PULSES IN ELECTROCARDIOGRAM SIGNALS

TECHNICAL FIELD The present invention relates to a method, and its implementation on a digital system, for detecting pace pulses in electrocardiogram (ECG) signal of a patient with a pacemaker.

BACKGROUND ART

Generally, a pace pulse that is acquired by the electrodes of an electrocardiograph, can be approximated by a square wave whose amplitude is bigger than 0.5 mV and whose duration lies within 0.1 to 2 ms [2]. The same square wave can overwhelm the physiological ECG signal in the diagnostic bandwidth (up to 150 Hz) [1]. Moreover, the overall signal that is acquired by the electrodes is typically characterized by different noise components that sometimes can disturb the pace pulse detection. A pacing pulse that is subject to a low-pass (LP) filtering in the diagnostic bandwidth, can be significantly widened depending on the filter impulse response; similarly when it is subject to a high-pass (HP) filtering, a tail at the end of the pulse can be created. This effects can cause both the pace pulse to overlap the physiological ECG signal and a failure of the pace pulse detection [2] [3]. For these reasons, it is necessary to detect and remove pacing pulses before applying any analog filter, which are typically used in ECG systems and typically work in the diagnostic bandwidth. The object of the present invention is a method to detect pace pulses in ECG signals, and its implementation in a fully digital equipment, with a flexible architecture, easily upgradable, which does not require any analog filtering in the diagnostic bandwidth before the ECG analog-to-digital conversion, and that is able to profusely exploit the electrical and statistical characterization of the pace pulse. Due to the high frequency nature of the pacemaker pulses, many methods and devices, like those proposed in [4] [5] [9], use dedicated analog circuitry to detect pacing pulses before the signal is passed to an analog-to-digital converter (A/D). Other methods and devices, such those in [6] [7], combine analog circuits with digital algorithms to aid the detection performance. Unlike the methods and devices that, because of their analog circuits, cannot be characterized by a flexible architecture, the fully digital methods and devices have a upgradable and flexible architecture that can be very useful to face with the evolution of the developed pacemakers. However, among all the known digital methods and algorithms, those that have a low complexity by employing simple linear filtering approaches [1] [2], cannot guarantee very good performance even in the presence of wide band electromyographic noise; on the other side, those algorithms [8] [11] [12] that employ a nonlinear filtering approach, do not fully exploit the pacemaker electrical and statistical characterization and thus are not totally protected against impulse and strong electromyographic noise.

DISCLOSURE OF INVENTION

In order to amplify the pace pulses with respect to the physiological ECG signal and also to all the noise components that affect an electrocardiogram measure, the present invention exploits the structure and the statistical characterization of the pace pulses in an ECG signal. The said goal is reached, by means of what has been discovered, by a procedure that includes a non-linear filtering that can enhance the pace pulses with respect to wide band noise and potentially, but not limiting, also a linear filtering whose aim is to remove great part of the ECG physiological signal and the noise in the diagnostic bandwidth. The said goal is reached, by means of what has been discovered, by a procedure that includes a non-linear filtering that can enhance the pace pulses with respect to wide band noise and potentially, but not limiting, also a linear filtering whose aim is to remove great part of the ECG physiological signal and the noise in the diagnostic bandwidth. The following description aids efforts to understand how the two filtering stages helps the pace pulses detection. The signal that is acquired by the electrocardiogram electrodes is composed by the addition of the physiological ECG signal, of the pace pulses and of the additive noise which in turn can be expressed by the addition of three components:

• baseline wander, due to the patient respiration and movement, with very low frequency content ([0.05-1] Hz) [10];

• power-line interference, which consist of a 50/60 Hz pickup (depending on whether we are in Europe or in U.S.A.), and whose amplitude can reach a maximum of 50% of the peak-to-peak ECG amplitude;

• electromyographic noise, which is introduced by the electrical activity, and is characterized by a wide band frequency content.

Thanks to the fact that the ECG frequency content is almost on the diagnostic bandwidth [0- 150] Hz, and that the pace pulses (whose minimum duration is 0.1 ms) are conversely characterized by significant high frequency energy due to the sharp transition at the rising and falling edges of the pulse, it is easy to understand that:

• the significantly different frequency contents allow us to isolate pace pulses from the ECG signal, the baseline wander and the power-line interference simply by means of a linear filtering, without modify the electrical characterization of the pace pulses,

• the statistical characterization of the pace pulses allow us to design a non-linear filter able to enhance them with respect to the wide band electromyographic noise, even if the two components have very similar frequency content,

• finally, the pace pulses detection is obtained simply by comparing the non-linear filter output with a given threshold. Summarizing, with reference to the present invention, in order to effectively sample also the pace pulses with minimum duration (0.1 ms), the analog signal, which is acquired by the electrodes of an electrocardiograph, is analog-to-digital converted with a sampling rate which is higher than those typically used in the diagnostic system. The digital signal is successively high-pass filtered in order to remove the low frequency content components due to the physiological ECG signal and to the baseline wander and power-line interference noise components. The following non-linear filter is designed first by sorting the digital samples in an opportunely defined temporal window, then by differentiating the samples with adjacent ranks (see (8)) which belong to the vector which has been previously sorted. This approach significantly improves the pace pulse detection performance with respect with the linear filtering methods conventionally used in ECG systems [1] [2]. Indeed, simply by using one or more linear filters, due to the very similar frequency content of the pace pulses and the electromyographic noise, noise spikes can be frequently and erroneously interpreted as pace pulses. Moreover, the proposed non-linear filtering, that is mainly characterized by a sorting procedure, provides very higher robustness against wide-band noise, also with respect to the typical non-linear filtering approaches described in [8] [11] [12] that, on the contrary, do not exploit the different statistical characterization of the pace pulses with respect to that one of the wide-band noise. Indeed, the non-linear filtering approach of the present invention, which represents the signal derivative in the sorted domain, allows the detector to avoid false detections, which are typically caused by wide-band noise spikes, because it would compare, with a high probability, a noise spike with another noise spike of quite the same amplitude. To this end, it is very important to correctly choose the set of samples that have to be sorted in order to avoid the possibility to have a single noise spike in the considered window, thus causing a false detection. BRIEF DESCRIPTION OF DRAWING

Fig. 1 represents the architecture of an electrocardiograph that exploits the pace pulse detection system described by this invention. Fig. 2 and 3 show the block diagram and the flowchart of a possible, but not limiting, embodiment of the pace pulse detection method which exploits the invention, respectively.

Fig. 4 (a) shows the pace pulse response of a possible, but not limiting, embodiment of the linear filter, while Fig. 4 (b) shows the absolute value of the corresponding frequency response of the same linear filter. Fig. 5 shows the flowchart of an efficient implementation, but not limiting, of the sorting algorithm which can be used by the pace pulse detection system and apparatus

DETAILED DESCRIPTION

In order to clarify the adopted notation that is used in the following description and in the attached figures, the digital signal s[n] , which is acquired by the electrodes of an electrocardiograph, is expressed as s[n] = x_ECG [n] + p[n] + w[n] , ( 1 ) where x_ECG[n] represents the physiological ECG signal in the absence of noise, p[ri] is the signal caused by the presence of the pacemaker and

is the additive noise, which in turn is expressed as

w[«] = w_m [»] + ^WPLI M + ^WEMG W» (²) where w_mv[n] , w_PLI [n] and w_mG[ή] are the baseline wander, the power-line interference and the electromyographic noise contribution, respectively. Even if a pacemaker pulse in an ECG signal may have both positive and negative amplitude, without loss of generality, the rest of the description and the attached figures refer simply to a sole positive impulse pace pulse. The following description relates to a possible, but not limiting, realization of an apparatus, which can be used into an electrocardiograph, for pace pulses detection that exploits the invention.

Fig. 1 shows how the electric signal of a patient with a pacemaker (101 ) is processed by a wide-band analog circuit (102) whose aim is to amplify the signal and adjust its dynamic range to the analog-to-digital converter (104) input. After the signal is low-pass filtered (103) in order to avoid aliasing, it is digitally converted (104) employing a sampling frequency f_s higher than those used by the typical monitoring systems, in order to guarantee the pace pulse detection and also to prevent the pacing pulses widening; the digital signal which is obtained this way is processed by the pace pulse detection method (105) in accordance with the present invention. Once the pacing pulses are detected, they can be easily removed (106) at the high sampling frequency f_s by simply replacing a portion of the signal before and after the time instant of the detected pulse with a signal average over the recent past. The high sampling frequency which is adopted, also permits to implement some or all the algorithm for the detection of the typical parameter of the high-resolution electrocardiography (109). Successively, it is possible to down-sample (107) the digital signal for a standard ECG monitoring system (108).

A possible, but not limiting, embodiment of the pace pulse detection method, in accordance with the invention, is described by the block diagram and the flowchart which are shown in Fig. 2 and Fig. 3 respectively. In this embodiment of the invention it is possible, but in a not limiting way, to consider an analog-to-digital conversion of the signal with a sampling frequency f_s = 10kHz . Optionally, it is possible to introduce a linear high-pass filter (201) and (301), which can be implemented by the average of two consecutive two-step (2 + / ) finite differences (derivatives), as summarized by the following equation s_HP[n] = s[n] + s[n -l] -(s[n - 2 - l] + s[n - 3 -l]). (3) The goal of this high-pass filtering is to enhance the rising and falling edges of the pacing pulse with respect to the low frequency contents of the physiological electrocardiograph signal. x_£CG[n] and of the noise components w_m[n] and

.

Fig. 4 (a) shows the output of the linear filter when its input (401 ) is the electrocardiograph signal s[ή] where the pace pulse signal p[n] is positive and has a width of d samples, distinguishing the case (402) when I < d from the case (403) when / > d . In this exemplary embodiment, the signal s_HP\n\ is characterized by two trapezoidal waves with opposite amplitude, whose discrete duration l_F and distance d_F (total duration d_τoτ = 21 _F +d_F ) are Ip = min( d + 1, 1 + 3); d_F = max(d, l + 2) -l_F . (4)

Clearly, the parameter / in (4) controls the filter time-span and, consequently, frequency response H_HP(ώ) that is associated to eq. (3), and expressed by

TT , x , . -i<^- ■ ( 2 +lλ (ω\ ,^,,

H_HP(ω) = 4j e - sin ω cos — , w) and whose absolute value \ H_HP(ω) \ is shown in Fig. 4 (b) for different values of the parameter / . It is clear that the low frequency content of the electrocardiograph signal and of the noise components w_BW [n] and w_PLI [n] are widely eliminated in all the conditions. For example, a possible, but not limiting, choice for the parameter / in the exemplary embodiment is to maximize the low frequency reduction and thus to use / = 0.

The linear filter output s_HP[n] is processed by the non-linear filtering stage expressed by (302) and (304). Specifically, we define the (N + T) -dimensional vector s_N corresponding to an observation window extended towards the (N + V) samples, as expressed by s_N = [s_HP[n], s_HP[n -k], s_HP[n - k -ϊ], ..., s_HP[n - k-N + l]]. (6)

It is clear that the observation window is composed by the current sample, which has to be detected as a pace pulse or not, and by the JV contiguous past samples (stored into a FIFO memory (203) and obtained by a delay block (202)) at a discrete temporal distance k .

A possible, but not limiting, choice of the parameter k , in an exemplary embodiment in 65 accordance with the present invention, could be to consider a 23 samples discrete temporal distance, which is not else but the discrete temporal distance d_τoτ = 21 _F + d_F between the beginning of the first and the end of the second of the two trapezoidal waveforms in Fig. 4 (a), when 1 = 0 , the sampling frequency f_s = 10 kHz , and d = 20 , which corresponds to the wider possible pace pulse duration [2]. This way, in the exemplary embodiment considered, it 0 is possible to avoid the comparison between rising and falling edges of the same pace pulse. Indeed, by using k > 23 , the observation window is composed only by the sample that potentially belongs to a pace pulse and by those samples that potentially do not belong to it. The non-linear filtering which is obtained by sorting (204) (302) the observation window s_N , produces a variational series \_N[n] (303) expressed by S_v r_w-i _ L(D „(2) „(#+1) 1 _{with v}(i) < „(2) < < „ (N₊I) _m

/0 ^N lⁿi - [^SHP ' ^SHP > — ' ^SHP J> ^{WITπ S}HP - ^S HP ^ — ^ S_Hp . { / )

Thus we can define the rank R(γ_N[n],i) as the position of the sample s_HP[n — i] in the variational series Y_N[n] and

as the value of the sample whose position is m in the variational series \_N[n] . It is thus possible to define the differential rank dr[ri\ as

which is defined either as the difference between the current sample with the next (or previous) one in the variational series if the position of the current sample is smaller (or higher) of the median in the sorted version of the observation window, or as zero if the said position exactly corresponds to the median. The median of a variational series \_N[n] of

85 N + l samples is defined as

V(y_N[n],N/2 + l) if N + l is odd m^{edian(Υ[n]) =} \ ~(v(y_N[nUN-l)/2)₊V(y_N[nUN + l)/2)) if JV^" + 1 is even O) We identify a pace pulse by comparing the differential rank

with a threshold T (206) (305), which for example can be set to 0.35 mV. In order to avoid a false pulse detection induced by a sudden voltage rise caused by an electrode-skin contact loss, it is necessary

90 that the sample s[ή] with the whose corresponding dr[n] is bigger than the threshold T , is

. followed, within a discrete temporal distance L (306), by a sample s[k], obtained by the delay block (206), whose dr[k] has both its absolute value bigger than the threshold (308), and the opposite sign of dr[n] (309).

With reference to the parameters that have been chosen in this exemplary embodiments in

95 accordance with the present invention, a possible but not limiting choice of the parameter L is L = 23 samples, which represents the minimum distance between the two trapezoidal waveforms in s_HP[n] that are obtained by a pace pulse in p[ή] . In a wide band noise environment, typically caused by the electromyographic noise x_mG[ri] , the differential rank signal dr[ή] can increase the detector performance with respect to the simple use of s_HP [n] .

100 Indeed, if we would simply use the filtered signal s_HP [ή] , some noise spikes could produce some false detections; on the contrary, the differential rank signal

, which represents the signal derivative in the sorted domain, allows the detector to avoid false detections because it would compare, with a high probability, a noise spike with another noise spike of quite the same amplitude. It is clear that if there is only a single noise spike into the

105 observation window, the detector may occurs in a false detection also with the differential rank signal dr[n] .

Thus, for this reason, it is very important to correctly choose the observation window width. In the exemplary embodiment, in accordance with the invention, a possible but not limiting choice of the parameter N of the observation window is JV = 100 as the result of a tradeoff no between reliability (the false detection probability is reduced by a wide support window) and accuracy (a smaller value of N reduces the probability to have a missed detection). Once a pacing pulse has been detected thus obtaining p[n] ≠ 0 in s[n] , the detection algorithm considers a refractory period of R samples (310) in which it does not search (309) for any other pulses in order to avoid multiple detections of the same pulse. A possible, but

115 not limiting choice of the refractory period R is i? = 1000 samples.

The flowchart of an efficient implementation of the sorting algorithm is shown in Fig. 5. The finite difference s_HP[z] of an ECG signal, which is obtained in a generic sampling interval z , is with high probability very similar to the value s_HP[z-ϊ\ that is calculated in the previous sampling interval. For this reason, for each step of the sorting algorithm, the variational

120 series \[n] is obtained by putting the samples s_HP[n] and s_HP[n -k] (that are represented by s_HP[z] in Fig. 5) into v[» -l] , which is obtained in the previous step, and discarding the samples s_HP[n -N -k] and $_HP[n -l] from it. In order to decrease the computational complexity, instead of sorting the vector by comparing the samples s_HP[n] and s_HP[n — k] ($_HP[∑] ) with those either at the outermost or in the middle of the previous variational series

125 \[n -1] , we exploit the low variability of the ECG signal by starting the sorting by comparing the new samples s_HP[n] and s_HP[n -k] with those samples in the previous variational series v[« -l] that correspond to the next samples (501) s_HP[n -l] and s_HP[n -k -l] (s_HP[z-ϊ\ ) and by moving towards higher or smaller values depending on the result of the comparison (502) (503). To this end, in order to both correctly insert the new samples s_HP[tt] and

130 s_HP[n -k-Y\ (504) (505) into and correctly remove the older ones s_HP[n -k -N] and s_HP[n -Y\ from the previous variational series \[n -ϊ] , it is important to control the structures by two doubly linked lists, the first to control the temporal sorting and the second to control the amplitude sorting of the samples. Bibliography

[I] E. D. Helfenbein.et. al., "A Software-based Pacemaker Pulse Detection and Paced Rhythm

Classification Algorithm," Journal of Electrocardiology, vol. 35, pp. 95-103, 2002.

[2] C. E. Herleikson, "ECG Pace Pulse Detection and Processing," US Patent 5,682,902, HP Company, 1997.

[3] R. F. Donehoo, D. W. Browne, "Pacemaker Pulse Detection and Artifact Rejection," US Patent 5,660,184, J&Johnson Medical, 1997.

[4] M. N. Shaya, B. L. Wyshogrod, "Pace Pulse Identification Apparatus," US Patent 4,664,116, HP Company, 1987.

[5] R. J. Regan, "Pace Pulse Signal Conditioning Circuit," US Patent 4,574,813, HP Company, 1986. [6] J. Wang, M. N. Shaya, "Pace Pulse Eliminator," US Patent 4,832,041, HP Company, 1989.

[7] B. H. Taha, S. B. Reddy, "Method and Apparatus for Automatically Detecting and Interpreting Paced Electrocardiograms," US Patent 6,304,772, GE Medical System Inf. Technologies, 2001.

[8] JR. Zinser et al., "Method and System for Enhancing Pace Pulses", US Patent 0069321 A1 , 2006.

[9] J. Yinbin, Y. Youli, J. Jie, H. Yecho, "Development on Pacing ECG Monitoring System," Proc. of IEEE EMBC 1998, vol. 20, n. 1 , 1998.

[10] J. M. Leski, N. Henzel, "ECG baseline wander and powerline interference reduction using nonlinear filter bank," Signal Proc, Elsevier, vol. 85, n. 4, pp. 781-793, April 2005.

[I I] Alexander Holland, "Nonlinear Method and Apparatus for Electrocardiogram Pacemaker Signal Filtering", US Patent 7,383,079, Welch Aliyn, Inc. 2008.

[12] JR. Zinser et al., "Method and System for Detecting Pace Pulses", US Patent 0020219 A1 , 2006.

Claims

1. A method to identify pacemaker pulses in an electrocardiogram signal, which exploits the non-linear processing of an ensemble of samples obtained by sampling, and possible processing, of said electrocardiogram signal. Said non-linear processing is characterized by the fact that it comprises the succession of the following processing steps:

.A. ordering by amplitudes of part of the said samples that belongs in time to an opportune observation window;

B. a linear filtering that works both on that sample, which we want to know if it contains, or not, a pacemaker pulse, and on the samples in its vicinity in said ordered ensemble;

C. comparison either of the output of the previous filtering stage, or of its possible further processing, with an opportune threshold, in order to detect the presence of a pacemaker pulse.

2. The method of claim 1, wherein the said observation window in time of the samples used in the said ordering is formed by the current sample and by an ensemble of consecutive samples which precede and/or follow said current sample and that are at an opportune temporal distance from it (402).

3. The method of claim 1 , wherein said linear filtering implements (404) either the difference of the current sample with the preceding one in said ordered ensemble, if said current sample is greater than the median of said ordered ensemble, or the difference of the current sample with the successive one in said ordered ensemble, if said current sample is lower than the median of said ordered ensemble, or the difference of the current sample with itself, if said current sample is equal to the median of said ordered ensemble.

4. Apparatus which contains means to detect pacemaker pulses in an electrocardiogram signal, which exploits the non-linear processing of an ensemble of samples obtained by sampling, and possible processing, of said electrocardiogram signal. Said apparatus includes the following functional blocks:

A. a first non-linear filtering stage that associates to each signal sample at the input an ordered ensembles of samples obtained by amplitude ordering the samples that belong to an opportune observation window in time

B. a second linear filtering stage that compares the current sample with samples that are close to said current sample in the said ordered ensemble, which was previously obtained;

C. a third stage that establishes the presence, or not, of a pacemaker pulse into the sample obtained at the output of the chain of said preceding filtering stages, by the comparison of this sample with an opportune threshold.

5. The apparatus at claim 4, which contains a further filtering stage, before said non-linear processing, which implements the average of the finite differences of two, or more, couples of samples, where each couple is at an integer distance greater than one (401).

6. The apparatus at claim 4, wherein

A. said observation window in time (402) contains the current sample and an ensemble of consecutive samples, which precede and/or follow said current sample, and are placed at a distance equal to 2 milliseconds from it, and that belong to a temporal interval with almost 10 milliseconds of duration;

B. said comparison is implemented (404) either by the difference of the current sample with the preceding sample in the said ordered ensemble, when said current sample is greater than the median of said ordered ensemble, or by the difference of the current sample with the successive sample in the said ordered ensemble, when said current sample is lower than the median of said ordered ensemble, or by the difference of the current sample with itself, when the said current sample is equal to the median of said ordered ensemble.

7. The apparatus at claim 4, wherein it is employed an efficient ordering method that, for every new input sample, updates the ordered ensemble of samples that were associated to the previous input and that, in order to reduce the computational complexity, exploits the low variability of the electrocardiogram signal and orders each new sample by a comparison with the samples of the ordered vector of the previous input, starting this comparison for each new sample from the position, into the ordered ensemble, of the sample immediately before it in time, and by moving towards either greater values, or lower values, when the new sample results to be greater, or lower, than the sample was compared with, respectively (502-503).

8. The apparatus of claim 4, which contains means that detect a pace pulse at a discrete temporal instant when said threshold (about 0.35 millivolt) is exceeded by the absolute value of two samples at the output of said filtering stages, which have opposite signs, and where the first represents said output obtained for to said discrete temporal instant of interest, and the second corresponds to said output obtained for a successive discrete instant whose temporal gap is less than, or equal to, about 2 milliseconds.

9. The apparatus of claim 4, wherein immediately after a pace pulse detection, it is inhibited the detection of further pacemaker pulses by using a refractory period with time duration of about 100 milliseconds (410).

10. Monitoring system for electrocardiogram signals, which detect pacemaker pulses and it works either by means of the method in claim 1 or by means of the apparatus in claim 4.