Cardiac Arrhythmia Classification Using Wavelet Analysis of Electrocardiograms

Cardiac Arrhythmias are heart rhythm abnormalities that seriously impact the quality of life and can be lethal. Four major types of cardiac arrhythmia that originate from atria and ventricles are atrial flutter (AFL), atrial fibrillation (AF), ventricular tachycardia (VT), and ventricular fibrillation (VF). Annually about 50,000 embolic strokes mostly due to AF and 35,000 sudden cardiac deaths mostly due to VF are reported in Canada. Accurate detection and segregation of these arrhythmia swiftly is an essential requirement for appropriate treatment. Automating this process is especially critical and valuable for implantable devices and long-term monitoring scenarios. 

In this thesis, with the above motivation, we analyzed the electrocardiograms (ECGs) recorded during these 4 types of arrhythmia using wavelet transform. A total of 100 ECG segments con- taining 25 ECG segments each for AF, AFL, VT and VF, obtained from well-known open source databases were used in this study. Discriminative features were extracted from the wavelet coefficients and fed to a linear discriminant analysis based classifier. Based on the proposed scheme, best classification accuracies using library wavelets and adaptive continuous wavelets (ACW) are as follows: (i) for four group classification, Paul CW and A-pattern ACW attained 77% and 81% respectively, (ii) for the two group classification of AA, Paul CW and M-pattern ACW attained 76% and 86% respectively, (iii) for the two group classification of VA, Bump CW and M-pattern ACW attained 92% and 94% respectively.