Validation of automated arrhythmia detection for Holter ECG

This paper describes a fast and very effective feature extraction technique for detection and discrimination of QRS on a microprocessor-based Holter ECG analysis system. The technique converts long term (up to 24 hours) recorded ECG into a positive waveform by signal preprocessing. Three characteristic factors, the duration, the areas, and the original slope of the positive waveform are detected when the onset and end points of each pulse have been detected by dynamic threshold detection. The prominent feature is extracted from a product of these three factors. It is used to identify normal beats and arrhythmias. This method has been examined using 47 different patients' ECG signals on a MIT/BIH database. The accuracy of QRS detection was 99.3% in validation. The identification sensitivity of PVC beats was 95.2% with 14 different arrhythmia patients. The method has also been implemented on a microprocessor based Holter ECG analysis system. A record of the MIT database recorded ECG can be completely analyzed within 30 second for reporting the heart rate variations, heart beat classifications and arrhythmia analysis.