A pattern-recognition program for continuous ECG processing in accelerated time.

Abstract A minicomputer-based system has been developed for beat-by-beat shape classification of long-term ECG recordings being replayed at high speed. Hardware preprocessing of the signal permits the use of a small computer with only 8K bytes of memory. The computer program can establish prototypes of up to four different QRS morphologies present in the signal, and compares every incoming complex with all of these in turn. The prototypes are allowed to adapt to gradual changes in the morphology. Coupling interval histograms are generated for each type of beat encountered. Some elementary statistical processing of these histograms is also performed. The program runs interactively, a continuous graphic display of the results of the sorting procedure being provided to allow rapid intervention in case of errors. Classification failures such as artifacts are also displayed for possible operator action. The results of a preliminary validation of the program will be presented.