A knowledge-based system for arrhythmia detection

Current arrhythmia detectors perform well on patients who have stable normal QRS complex morphologies and distinct ventricular QRS complex morphologies in clean ECG signals. However when normal QRS complex morphologies are not stable and/or noise is present, these arrhythmia detectors perform poorly. The result is an intolerable number of false positive arrhythmia alarms and some missed arrhythmias. Thus, there is a tremendous need for a highly sensitive and accurate arrhythmia detector. In this work, we have developed such a highly sensitive and accurate arrhythmia detector. We employed a set of features used by clinicians to classify ECG beats. These features are obtained from single or multiple ECG channels as available. The posterior probability of a class given a feature was used to represent the classification knowledge. This knowledge is derived from a subset of the MIT-BIH database. The new arrhythmia detector was tested on both the MIT-BIH and AHA databases. It exhibited a significant improvement over the existing state-of-the-art arrhythmia detector.