Sparse Representation Based Anomalies Detection in Electrocardiography Signals

In this article, we present the use of sparse representation of signal and dictionary learning method for solving the problem of anomaly detection. The analyzed signal was presented as a set of correct ECG structures and outliers (characterizing different types of disorders). In the course of learning we used the modified Method of Optimal Directions (MOD) to find a dictionary that would reflect correct structures of an ECG signal. The dictionary found this way became a basis for sparse representation of the analyzed ECG signal. In the process of anomaly detection based on decomposition of the analyzed signal onto correct values and outliers, there was used a modified Alternating Minimization Algorithm (AMA). Performance of the proposed method was tested using a widely available database of ECG signals - MIT–BIH Arrhythmia Database. The obtained experimental results confirmed the effectiveness of the method of anomaly detection in the analysed ECG signals.