Abstract: The detection of anomalies in electrocardiograms (ECG) is a crucial technique for identifying irregular heartbeats, facilitating the early recognition of abnormal ECG readings prior to the diagnostic phase. Existing methods for ECG anomaly detection, which range from academic studies to commercial ECG devices, continue to experience a significant rate of false alarms. This issue arises from the inability of these methods to distinguish between ECG artifacts and genuine ECG signals, particularly when the artifacts closely resemble the actual signals in terms of shape and frequency. Consequently, this situation necessitates heightened vigilance among physicians and increases the risk of misinterpretation for those without specialized training. To address this challenge, the present study introduces a novel anomaly detection technique that demonstrates high robustness and accuracy in the presence of ECG artifacts, thereby effectively minimizing the false alarm rate. The design incorporates expert insights from cardiologists alongside motif discovery techniques. Furthermore, each phase of the algorithm aligns with cardiologists' interpretations. This method is applicable to both single-lead and multi-lead ECGs. The results of our experiments, conducted on real ECG datasets, have been analyzed and evaluated by cardiologists. As a result, ECG anomaly detection has gained significant traction among researchers and practitioners, being employed to identify periods of unusual ECG activity. The effectiveness of the anomaly detection method is directly correlated with the outcomes of cardiac disease identification and diagnosis. A network is trained using non-anomalous data and serves as a predictor over multiple time intervals. Determining whether the observed data is anomalous is a critical task that has been extensively explored in the literature. An auto-encoder is utilized to capture patterns, enabling the prediction of anomalies within the data. It is trained exclusively on normal pattern data and subsequently tested with anomalous data.
Keywords: Electrocardiogram, anomaly, auto-encoder, Long Short-Term Memory.
Copyright Form