A computer measures the ECG unaided

In February 1962, in Circulation, Cesar Caceres and colleagues at the heart disease programme of the US Department of Health, Education and Welfare showed that a program can measure the waves and intervals of an electrocardiogram by itself, without human intervention. They processed 36 leads from 20 people, and the measurements agreed with careful visual measurement to two decimal places. The program made no diagnosis.

The authors are from the Instrumentation Unit of the Heart Disease Control Program, Division of Chronic Diseases, US Department of Health, Education and Welfare, in Washington; part of the work was done by Airborne Instruments Laboratory (Deer Park, Long Island) under a Public Health Service contract. The programming was by Norman F. Ogilvie. Data: 36 electrocardiographic leads from 20 clinically normal individuals (not necessarily electrocardiographically normal), each recorded on magnetic tape with a frequency modulation system; the authors chose two leads to give a range of wave shapes. The leads were digitised at 625 samples per second, so time was measured to 0.0016 second and values to one part in a thousand, and stored on punch cards for analysis by a "general-purpose computer". The initial work was done on a "small, low-speed computer": measuring one lead took 64 minutes, and the authors estimate 15 seconds to 2 minutes on faster computers. Method (the appendix of the paper): the signal was smoothed by fitting a nine-point least-square parabola, its derivative was taken, and the point of greatest negative derivative, which falls in the QRS complex, served as the reference; an R wave was taken as present if the maximum positive derivative exceeded 9.375 millivolts per second within 0.128 second before it. The measured parameters were the P, Q, R, S, T and U waves and the PQ, ST, QT and RR intervals; the extra P', R' and S' waves were deliberately excluded at first. Results: the values were exactly reproduced on every run; the measurements were comparable in the first two decimal places to careful direct visual measurement of the standard electrocardiogram; the program did not depend on the lead and coped with 60-cycle noise. Arrhythmias were not classified. What the record does not state. The paper names no make or model of computer, so "general-purpose" and "small, low-speed" are all that is known of it. It does not count how many measurements were compared with manual ones and gives no error rate: the comparison is described in words. The discussion says each lead was recorded for 5 seconds (at least three beats), while the methods say "1 minute each"; the paper does not reconcile the two. It contains no diagnosis: the authors only name systems of logic for later use, with Bayes and Fisher among the examples.