Adaptive wavelet representation and classification of ECG signals

ECG signals are adaptively approximated as a weighted linear combination of translated and dilated mother wavelets. An ECG frame is thus represented by a limited number of adaptively estimated parameters indicating translation, scaling, and weights. Also, a neural network classifier that utilizes adaptive wavelet based features is used to discriminate between normal and abnormal beats. The ECG signals used in the experimental phase of this study are extracted from the "MIT/BIH" arrhythmia database.