Anomaly detection in ECG based on trend symbolic aggregate approximation.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Cardiology

Background:

• Electrocardiography (ECG) anomaly detection is crucial for diagnosing heart conditions.
• Unsupervised learning methods are preferred for ECG anomaly detection due to the scarcity and difficulty of labeling anomalous data.
• Existing unsupervised methods often neglect the distinct shapes of ECG signals associated with different diseases.

Purpose of the Study:

• To propose a novel unsupervised method for ECG anomaly detection that incorporates signal shape.
• To enhance the accuracy and robustness of detecting abnormal heart signals.

Main Methods:

• A simple trend aggregate approximation method is introduced.
• Relative binary trend representation is utilized to capture ECG signal shape features.
• Anomaly detection is performed through similarity comparison of these representations.

Main Results:

• The proposed method demonstrates robust performance in detecting ECG anomalies.
• Experimental results show high accuracy using sensitivity, specificity, and false alarm rate measures.
• The approach effectively utilizes ECG shape information for anomaly detection.

Conclusions:

• The novel method offers a promising approach for unsupervised ECG anomaly detection, considering signal morphology.
• This technique can improve the early detection of heart diseases by identifying subtle anomalies.
• The method provides a valuable tool for analyzing ECG time series data in clinical settings.