QRS Detection by Rules Based Multiple Channel Combinatorial Optimization

Abstract

A multiple channel QRS detector is described. Separately for each channel, the detector generates sequences of peaks and statistically scores them according to: 1) peak prominence; 2) temporal regularity; 3) peak shape similarity; and 4) number of skipped beats. In the case of unstructured rhythms, the temporal regularity score is null and does not contribute to sequence quality. If at least one winning score from any channel exceeds a quality threshold, multi-channel sequences are generated from the winning sequences peaks and scored according to the above measures and peak time coherence across channels. The winning multi-channel sequence is then selected. The algorithm was applied to both channels of the MIT-BIH Arrhythmia Database. Over the entire database, the sensitivity (SE) and positive predictive value (PPV) were 99.93% and 99.96% respectively. For record 203, generally considered the most difficult one in the database, the SE and PPV were 99.80% and 99.76% respectively. The present algorithm fits within the framework of a previously described algorithm that can detect sinus rhythm in high noise conditions (e.g. waist based textile electrode recordings of a jogging subject).