Biomedical Signal Processing and Control 10 (2014) 41–49 Contents lists available at ScienceDirect Biomedical Signal Processing and Control journal homepage: www.elsevier.com/locate/bspc Technical Note Automated analysis of ECG waveforms with atypical QRS complex morphologies Reza Tafreshi a,∗ , Abdul Jaleel a , Jongil Lim a , Leyla Tafreshi b a Mechanical Engineering Program, Texas A&M University at Qatar, Doha, Qatar b Lehigh Valley Health Network, Allentown, PA, USA article info Article history: Received 2 September 2013 Received in revised form 16 December 2013 Accepted 21 December 2013 Available online 20 January 2014 Keywords: ECG analysis QRS detection abstract Automated detection of the various features of an electrocardiogram (ECG) waveform has wide appli- cations in clinical diagnosis. Although detection of typical QRS waveforms has been widely studied, detection of atypical waveforms with complex morphologies remains challenging. The importance of detecting these complex waveforms and their patterns has grown recently due to their clinical implica- tions. In this paper, we propose a novel algorithm for detecting the various peaks of such complex ECG waveforms. It is identified that most of the well-formed ECG waveforms – both typical and complex – fall into nine broad categories according to the standard nomenclature. Motivated by this ECG wave- form classification, our algorithm uses signal analysis techniques such as first and second derivatives and adaptive thresholds to classify these waveforms accordingly by detecting the various features present in them. Temporal coherence along a single lead as well as spatial coherence across the 12 leads are used to improve performance. For waveform and pattern analysis, data from 50 healthy subjects and 50 patients with myocardial infarction were randomly selected. Results with an overall sensitivity of 99.06% and overall positive predictive value of 98.89% validate the effectiveness of the approach. Further, the algorithm gives true detections even on waveforms with fluctuations in baseline and wave amplitudes, proving its robustness against such variations. © 2013 Elsevier Ltd. All rights reserved. 1. Introduction The electrocardiogram (ECG) has been widely known as one of the most reliable and noninvasive tools for monitoring the heart rhythm. One heart rhythm or beat on an ECG signal consists of the P-QRS-T wave as seen in Fig. 1. The P wave, the QRS complex, and the T wave respectively represent atrial depolarization (contraction of the upper heart chambers), ventricular depolarization (contrac- tion of the lower heart chambers), and ventricular repolarization (relaxation of the lower heart chambers) [1]. Apart from the main features of a single beat – P, QRS, and T waveforms – there are other points of interest especially for the purposes of clinical diagnosis. ST segment is defined as the seg- ment between the end of the S wave and the start of the T wave. J point is defined as the junction between the QRS complex and the ST segment. [1,2]. Further, the isoelectric line is defined as the baseline of each ECG beat and is generally identified as the segment (1) between the end of the P wave and the start of the Q wave (PR), and (2) between the end of the T wave and the start of the P wave ∗ Corresponding author. Tel.: +974 44230237. E-mail address: reza.tafreshi@qatar.tamu.edu (R. Tafreshi). (TP). All these features have proven to be vital for the detection of many clinical conditions such as a myocardial infarction (MI) [3]. The development of accurate and quick methods for automatic QRS detection is of major importance in automated ECG waveform analysis. For more than 30 years, many QRS detection algorithms have been proposed. Most commonly used algorithms in QRS detec- tion are first derivative and/or filter based algorithms [4,5], neural network based algorithms [6,7], wavelet based algorithms [8–10], and independent component analysis (ICA) [11]. Among them it is reported that neural network, wavelet, and ICA based algorithms are computationally intensive. Alternatively, the first derivative based algorithms are widely utilized because of their low computa- tional cost [4,5,10]. However, the current QRS detection algorithms are unable to identify various types of noise or disturbances and sudden changes in the complexes [11–14]. Our previous study presented a QRS detection algorithm based on the use of simple pattern matching techniques [15]. The algo- rithm first classified the waveforms into five fundamental types, the most commonly observed ECG waveforms [16], as shown in Fig. 2. It then used properties such as the temporal correlation between successive ECG beats for the QRS detection in each category. The algorithm resulted in a true detection rate of 98.9% for typical ECG waveforms. 1746-8094/$ – see front matter © 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.bspc.2013.12.007