Evaluation of a computer-based system for detecting ventricular arrhythmias.

A digital system for real-time arrhythmia monitoring in the coronary care unit has been designed. The system is based on an algorithm for discrimination between normal complexes and ventricular ectopic beats (VBs). A beat is classified as normal if the absolute difference from a running average of the patient's normal QRS is below an adaptive threshold. To prevent artifacts and beats of non-ventricular origin from being falsely interpreted as VBs, each abnormal beat is correlated with a typical VB waveform, incorporated into the program. A VB is recognized only when the correlation coefficient exceeds 0.8. In a performance study, ECGs from 15 patients were recorded on magnetic tape and replayed to the computer. Independent evaluation by two physicians showed a total of 1 306 VBs, 94% correctly classified by the computer. In the group labelled "suspected VBs" the detection rate was lower (average 69%). Out of the whole number of complexes (53 260), 0.45% were falsely interpreted as VBs by the computer. Artifacts giving rise to false VBs are included in this figure. The causes of false positive and false negative VBs were thoroughly investigated and on the basis of these results, possible improvements in the system are discussed.