Professional Documents
Culture Documents
Short Communications: Limiiits, (V, M, Ax
Short Communications: Limiiits, (V, M, Ax
Short Communications
AZTEC, Preprocessing Program for
a description of the processing algorithm) sets initial con-
Real-Time ECG Rhythm Analysis ditions on tWo Iimits: Vmax=Vmin=VO. Samples obtained
at subsequent interrupts are compared to these limits.
Abstract-A preprocessing program developed for real-time
monitoring of the electrocardiogram by digital computer has proved If exceeded, a limit is replaced by the voltage just
useful for rhythm analysis. The program suppresses low amplitude samlpled. As long as the difference betweein the limiiits,
signals, reduces the data rate by a factor of about 10, and codes the (V,m,ax - Viin), does not exceed an experimentally deter-
result in a form convenient for analysis. mined threshold, the fluctuating voltage is considered
to be adequately represented by a constant voltage, or
Digital computers are able to provide the biological "line,'" midway between the limnits. When finally a sam-
scientist witlh extensive comlputational capacity. How- ple would necessitate separating the limits by more than
ever, a common difficulty that biologists encounter is the threshold, the preceding average of the two limits is
the transformation of the observed analog signals into stored in the miiemory of the computer and called the
a form that can be easily processed by a digital com- value of the line. The time since the limits were initial-
puter. This poses a significant problem for those who ized is stored as the duration of the line. In order to
wish to use a small computer for on-line or real-time avoid long delays in the presentation of data to the
interaction with their experiments. Our problem is the ECG analysis program, a long line is terminated after
real-time analysis of electrocardiographic rhythm and 126 nis. In our program, the duration of a line occupies
we have developed a scheme to assist this analysis called six bits of the first of a pair of twelve-bit wvords and its
AZTEC (amplitude-zone-timie-epoch-coding). value occupies the second word.
The electrocardiogram is composed of low frequency When the difference between the voltage limits ex-
components (P and T waves, ST segments) and medium ceeds the threshold and a pair of data words is recorded,
frequency components (QRS complexes). We have the process is restarted by setting Vrnax and Vmin equal to
chosen to ignore higher frequency components in the the latest sample voltage. When a signal of higher fre-
QRS.1 The signal amplitude is 0.1 to 1 mV, and fre- quency and amplitude such as the QRS begins, the
quently contains undesired signals of similar amplitude voltage samples will change rapidly, and lines of short
ranging from very lowA frequencies (respiration at about duration will be formed. A series of lines, each containing
0.2 Hz) to high frequencies (muscle noise up to about four samples or less, is considered to be adequately rep-
200 Hz). Power line interference at 60 Hz is also corn- resented by a constant rate of voltage change, or slope,
mon. For accurate quantification of the duration and as long as the voltage difference between adjacent lines
amlplitude of the rapidly changing QRS we have chosen does not change sign. The slope is terminated by a line
to sample the electrocardiogram at a rate of 500 samples longer than four samples or a change in signs. The slope
per second.2 For anything but a trivial continuous real- duration and the voltage betweein the lines bounding
time analysis, this data rate would overwxhelm most the slope are then stored as the next pair of data words.
computers. The memory size of the LINC3 employed in In order that this pair may be recognized subsequently
this work, 2048 twelve-bit words, made the need for as a slope, the duration wvord is marked by setting its
data reduction even greater. In addition, some form of sign bit.
preprocessing was desirable to filter unavoidable noise Fig. 1 illustrates an ECG signal and its resulting
recorded with the ECG signal. AZTEC representation. The data reduction is from a rate
Our AZTEC preprocessing program is activated by a of 500 samples per second to an average of 25 word pairs
real-time interrupt that converts a sample of the input (50 words) per second, or about 10 to 1. The high fre-
voltage to digital form every 2 milliseconds. The first quency but low amplitude noise is interpreted simply as
sample v0 (see Appendix for notation and a compact a line as long as the peak-to-peak amplitude does not
exceed the threshold. The QRS consists of seven AZTEC
word pairs, four of which are slopes.
Manuscript received September 25, 1967; revised November 5,
1967. This work was supported by the Division of Research Facilities The AZTEC algorithm can be easily followed by addi-
and Resources, National Institutes of Health, Bethesda, Md., under tional digital processing, since the calculation time per
Grant FR-00161.
I P. H. Langer, Jr., D. B. Geselowitz, and F. T. Mansure, "High- sample is considerably less than the time between sanm-
frequency components in the electrocardiograms of normal subjects ples for most digital nmachines. For example, we are cLr-
and of patients with coronary heart disease," Am. Heart J., vol. 62,
pp. 746-755, July-December 1961. rently using it as input to a program that evaluates the
2 C. A. Steinberg and L. WV. Paine, "Methods and techniques of signal noise level and, where possible, detects QRS comi1-
data conversion," Ann. N. Y. A cad. Sci., vol. 115, pp. 614-626, July
31, 1967. plexes, measures their width, and codes their shape in
3 V. A. Clark and C. E. Molnar, "A description of the LINC," in terms of their sloping segments. Our implementation of
Computers in Biomedical Research, vol. 2, Stacy and Waxman, Eds.
New York: Academic Press, 1965, pp. 35-66. this detection algorithm is quite fast and can identify
SHORT COMMUNICATIONS 129