Online anomaly detection in ECG signal using Hierarchical Temporal Memory

Anomaly detection in time series is a well-studied subject, and it is well-documented in the literature such as ECG signal. Many successful algorithms for analyzing ECG signals are proposed and applied successfully to detect anomalies. However, ECG signal is a complex signal since it is contextual time series. The heartbeats changes over time and each user has his own heartbeats morphology. Consequently, there is no a specific model for this kind of signals. In this paper, we propesed to apply Hierarchical Temporal Memory (HTM). HTM is compatible with the changeable pattern because it is able to integrate a contextual properties to make prediction. A real dataset called MIT-BIH Arrhythmia is applied to evaluate HTM anomaly detection in ECG signals. Compared with Deep Learning, HTM is more efficiently to anomalies detection in ECG heartbeats.