A Cross Wavelet Transform Based Approach For ECG Feature Extraction and Classification Without Denoising

Automatic classification of cardiac patterns has become a challenging problem as the morphological and temporal characteristics of the ECG signal shows significant variations for different subjects. Most of the classification methods use explicit time-plane features information like presence of abnormal Q wave, QS complexes, ST segment, R height, QT segment measurement etc. Also ECG signals gets corrupted by various forms of noises. Before any feature extraction technique ECG requires to be preprocessed for removal of artifacts and other high frequency noises. This paper presents an ECG based feature extraction and classification technique which does not require conventionally used time plane features also the features in use are extracted from noisy data. The developed method also extracts parameters which have sufficient distinguishing characteristic to classify normal and abnormal cardiac patterns. The proposed algorithm analyses ECG data through the scope of cross-wavelet transform (XWT) and explores the resulting spectral differences. R peaks are detected for beat segmentation and extraction of any other explicit time plane features is not required. The cross-correlation between two time domain signals gives the measure of similarity between two waveforms. The application of the Continuous Wavelet Transform to two time series and the cross examination of the two decomposition reveals localized similarities in time and frequency. Application of Cross Wavelet Transform to a pair of signals yields wavelet cross spectrum (WCS) and wavelet coherence (WCOH). A heuristically determined mathematical formula extracts parameter(s) from the wavelet cross spectrum and coherence. Empirical tests establish that the two parameter(s) are relevant for classification of normal and abnormal Cardiac patterns. Advantage of this method is: i) It efficiently works in noisy environment ii). Explicit time plane feature extraction is not required and eliminates the use of rule mining procedure thus reducing the computational complexity of the classifier.