Arrhythmia Classification using DWT-Coefficient Energy Ratios

Certain features present in electrocardiogram (ECG) signals are used to detect different heart conditions. Hence, by developing a system to extract these features, useful information related to the heart conditions could be obtained automatically. Specifically, degree of arrhythmia, i.e., how much the heartbeat is deviating from the normal pattern. In this paper, we present a classification algorithm based on features extracted from the discrete wavelet transform (DWT) of ECG signals. Each ECG signal was analyzed using multi-level wavelet decomposition. Then, the energy of the obtained detail and approximation coefficients for each level was computed. The distribution of the signal's energy along the DWT levels was used to compare between different ECG signals and to classify them into the correct classes. Four features were used in the classification algorithm; in which four abnormal conditions are classified. After testing the developed classification algorithm on different beats, the algorithm had an accuracy of more than 95% for detecting any of the five classes. The technique was found to be scalable to any other number of arrhythmia classes.