Professional Documents
Culture Documents
Abstract-The five step digital filter has been developed which re- the drawbacks of the conventional methods listed above. The
moves components other than those of QRS complex from the re- algorithm which extracts QRS complexes from the recorded
corded electrocardiogram (ECG). The final step of the filter produces ECG will be described in detail. In this study, a single lead is
a square wave whose on-intervals correspond to the segments with QRS
complexes in the original wave. used to record ECG's.
ALGORITHM
INTRODUCTION As first step, high-frequency components are separated
a
UP to this time, special purpose analog equipments have from the sampled ECG. ECG is sampled at the rate of 1000
been used to conduct continuous care on arrhythmia. It points/s.
is, however, not a easy task to incorporate fairly complicated Step 1:
logical circuit into analog equipment, which is necessary in The averages for every four sampled points are computed
order to monitor and diagnose every pattern of arrhythmia in real time and are stored. Let {Yo0, Yo2, * , Yon } be the
precisely. Most of the existing equipments employ simple sequence of data points thus stored. {Jo,} (i = 1,.* , n)
judging criteria, and there have been unsatisfactory perfor- may be regarded as the data sampled at the rate of 250
mances with them such as raising an alarm too often and so points/s. (From here on, the notation {Yji} will be used to
forth [1]. denote the waveform consisting of the points yjijs.) Smooth
The awkwardness of analog equipment has stimulated the {y0j} (i = 1, 2, *, n) by three-point moving average
development of digital computer techniques for the processing weighted with coefficients { 1, 2, 1 }. The resultant waveform
and analysis of electrocardiograms (ECG's) in the patient care is represented as {y1j} (i = 1, 2, * , n).
system. Although there have been many softwares developed Step 2:
for the computer analysis of ECG, most of them will not be Let {YJiT denote the output from the low-pass filter. Com-
applicable to the system where monitoring of ECG is per- pute the squared differences between {Yi j} and {y)j} (i =
formed for a long period of time. To obtain the satisfactory m + 1, ,n- m)as
functioning from monitoring system for arrhythmia, further
reliability must be added to the existing techniques for the Y2i=(Yli-
Y-i)2 i=rm+ 1,* n-rm (1)
computer analysis of ECG. where
This paper reports the developed method for recognizing
QRS complex which is the most important part in the com- 1 i+m
puter ECG monitoring system. QRS complex contains relatively 2m + 1 k=i-m
high-frequency components compared with other waves. Sev-
eral algorithms have been devised based solely on the point that The data pointsY2mi1,Y2m+2,** ,Y2n-m are the output
[2]-[4]. Most of them, however, have some drawbacks such from the bandpass filter which forms peaks at locations where
as the following: they may fail to detect ventricular ectopic the spikes exist in {Yiy}. Fig. 1 shows frequency character-
beats with small amplitude; may mistake abrupt disturbance istics of the filter (1).
of base line for QRS complex. In case there exist several QRS In the obtained waveform {Y2i}, there are fault peaks, e.g.,
complexes with different amplitudes and T waves with high the peaks formed by the existence of sharp T waves from
amplitude at the same time, it is difficult to set up the bound- which QRS complexes must be differentiated.
ary to distinguish between these two different waves. More- Step 3:
over, the existing methods often record three leads simul- The high-frequency components of QRS complexes appear
taneously, and compute spatial velocity for detection of QRS for a long time period from the onset to the end compared
complexes [5] -[7], but it is desirable to use as small a num- with those of other waves. Noting this point, compute the
ber of leads as possible in a patient monitoring system. modified waveform {Y3i} as
The method developed here has achieved improved reliabil-
ity in the automatic detection of QRS complexes, eliminating y3-Y2i (i+M Y2k 12
64 klk-m 64
Manuscript received November 29, 1976; revised April 15, 1979. i=MT+bLmb+t2, ne (2)
The author is with the Department of Neurophysiology, Brain Re-
search Institute, Niigata University, Niigata, Japan. The division by 64 is to prevent overflow during computation
Authorized licensed use limited to: Achmad Arifin. Downloaded on November 13, 2008 at 23:05 from IEEE Xplore. Restrictions apply.
OKADA: QRS COMPLEX DETECTION 701
Authorized licensed use limited to: Achmad Arifin. Downloaded on November 13, 2008 at 23:05 from IEEE Xplore. Restrictions apply.
702 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. BME-26, NO. 12, DECEMBER 1979
iA
0 .1 150
A-fA WV_
.1
..1.
A- 92 A :A..
100
Of
0
(a)
: 0
50
Fig. 3. Tolerance versus noise. Ordinate and abscissa represent the out-
put from the filter and the amplitude of 50 Hz noise.
TABLE I
COMPARISON OF VK.
Intervals with Intervals without
(b) QRS complex QRS complex
I
I
1. original ECG 1.03487 1.01379 - 1.51009
. -
2. the first derivative* 1.58155 0.05629 - 0.24858
.0~ ~~~
y
3. a.m.a.l.f.** 0.97807 0.07443 - 1.32840
..
z o' S .~;.. .0 *
The first derivative of 1.; ** waveform processed by-autoregressive-
moving average low-pass filter.
Authorized licensed use limited to: Achmad Arifin. Downloaded on November 13, 2008 at 23:05 from IEEE Xplore. Restrictions apply.
OKADA: QRS COMPLEX DETECTION 703
TABLE II ACKNOWLEDGMENT
THE DATA USED AND THE RESULTING PERFORMANCE
The author wishes to thank Dr. N. Maruyama and Miss M.
A*2 B.2 C*2 D*3 Okada for their helpful suggestions and Miss R. Sato for
typing the manuscript.
total number of 1818 857 1230 1085
beats analyzed REFERENCES
number of normal*I 1737 857 1105 605 [1] D. A. Frost, F. G. Yanowitz, and T. A. Pryor, "Evaluation of a
beats computerized arrhythmia alarm system," Aner. J. Cardiol., voL
39, pp. 583-587, 1977.
number of abnormal*' 81 0 125 480 (2] C. A. Caceres, "Electrocardiographic analysis by a computer sys-
beats tem," Arch. Intern. Med., voL 111, pp. 196-202, 1963.
[3] T. A. Pryor, R. Russell, A. Budkin, and W. G. Price, "Electrocar-
number of false 3 0 1 4 diographic interpretation by computer," Comput. Biomed. Rea.,
positives vol. 2, pp. 537-548, 1969.
number of false 0 0 2 0 [41 C. C. Wilton-Davies, "Computer-assisted monitoring of ECG's and
negatives heart sounds," Med. Biol. Eng., vol. 10, pp. 153-162, 1972.
[5] H. K. Wolf and P. M. Rautaharju, "An on4ine program for acquis-
tion and analysis of resting and exercise electrocardiograms," in
*1 The figures are obtained by careful doctor interpretation. Proc. XIth Int. Vectorcardlography Symp. Amsterdam, The
Netherlands: North-Holland, 1971, pp. 231-236.
*2 The cases with sporadic ventricular ectopic beats. [6] J. Cornfield, R. A. Dunn, C. D. Batchlor, and H. V. Pipberger,
"Multigroup diagnosis of electrocardiograms," Comput. Biomed.
*3 The case with atrial flutter with repeated multifocal ectopic beats. Res., vol. 6, pp. 97-120, 1973.
[7] M. L. Simoons, H. B. K. Boom, and E. Smallenburg, "On4ine pro-
cessing of orthogonal exercise electrocardiograms," Comput.
for performance evaluation, there have been some methods Biomed. Res., vol. 8, pp. 105-117, 1975.
reported [9]. In this study, we evaluated the performance by [8] C. H. Klingenmaier, P. R. Moyer, J. I. Aunon, M. L. Shaffer, F. A.
investigating the false positive and false negative which oc- Siegel, and J. C. Rios, "A method of computer-assisted pacemaker
surveillance from a patient's home via telephone," Comput. Bkob
curred in QRS complex detection. The procedure was applied med. Res., vol. 6, pp. 327-335, 1973.
to the data obtained from four patients in surgery. Table II [9] C. A. Swenne, J. S. Duisterhout, and N. M. V. Hemel, "Interactive
summarizes the performance. As for the data obtained from computerized CCU-monitoring," in MTends in Computer-Processed
Electrocardiograms, J. H. V. Bemmel and J. L Willems, Edi
the patients A, B, and C, the cases with sporadic ventricular Amsterdam, The Netherlands: North-Holland, 1977, pp. 59-63.
ectopic beats, a total of 3905 beats (3699 normal and 206 ab-
normal beats as judged by doctor interpretation) was analyzed. Masahiko Okada (M77) was born in Niigata,
As a result, detection of QRS complexes was falsely reported Japan, on July 28, 1946. He received the M.D.
four times (false positive) and was missed twice (false nega- degree from Niigata University, Nilgata, Japan,
tive). Every false positive was due to the abrupt fluctuation of hin 1972.
From 1972 to 1973 he received his tining
the base line. Every false negative occurred at widened QRS in internal medicine at the Department of Med-
complexes caused by bundle branch block. As for ECG of the icine, Niigata University School of Medicine,
patient D, there was atrial flutter with repeated multifocal Niigata, Japan. In 1973 he joined the Brain
Research Institute, Niigata University, for the
ectopic beats which made it difficult to visually identify nor- study of medical information processing. He
mal configuration of QRS complexes. The method was ap- has been working on computer diagnosis, data-
plied to 1085 beats of ECG of the patient (605 normal and base, patient monitoring system, microcomputer application, and so on.
Dr. Okada is a member of the Japanese Society of Medical Electronics
480 abnormal), and obtained was such satisfactory results as and Biological Engineering, the Institute of Electronics and Communi-
only four false positives and no false negatives. cation Engineers of Japan, the Japanese Circulation Society, ACM, etc.