Envelopment filter and K-means for the detection of QRS waveforms in electrocardiogram

• Published in: Medical engineering & physics Vol. 37; no. 6; pp. 605 - 609
• Main Authors: Merino, Manuel, Gómez, Isabel María, Molina, Alberto J.
• Format: Journal Article
• Language: English
• Published: England Elsevier Ltd 01.06.2015
• Subjects: Algorithms; Arrhythmias, Cardiac - diagnosis; Arrhythmias, Cardiac - physiopathology; Biosignal processing; Databases, Factual; ECG signal; Electrocardiography - methods; Envelopment; False Positive Reactions; Heart Rate - physiology; Humans; K-means clustering; QRS detection; Radiology; Sensitivity and Specificity; Signal Processing, Computer-Assisted; Time Factors; QRS detection; Biosignal processing; ECG signal; K-means clustering; Envelopment
• ISSN: 1350-4533; 1873-4030; 1873-4030
• DOI: 10.1016/j.medengphy.2015.03.019

•We develop a free-parameter processing to detect. QRS-waveform.•We have defined an offline and real-time versions with similar results.•A high level of correct classification is achieved (>99.75%).•The computational load is linear.•The ECG data were obtained from Physionet Database, everyone can repeat tests. The electrocardiogram (ECG) is a well-established technique for determining the electrical activity of the heart and studying its diseases. One of the most common pieces of information that can be read from the ECG is the heart rate (HR) through the detection of its most prominent feature: the QRS complex. This paper describes an offline version and a real-time implementation of a new algorithm to determine QRS localization in the ECG signal based on its envelopment and K-means clustering algorithm. The envelopment is used to obtain a signal with only QRS complexes, deleting P, T, and U waves and baseline wander. Two moving average filters are applied to smooth data. The K-means algorithm classifies data into QRS and non-QRS. The technique is validated using 22 h of ECG data from five Physionet databases. These databases were arbitrarily selected to analyze different morphologies of QRS complexes: three stored data with cardiac pathologies, and two had data with normal heartbeats. The algorithm has a low computational load, with no decision thresholds. Furthermore, it does not require any additional parameter. Sensitivity, positive prediction and accuracy from results are over 99.7%.