Noise detection on ECG based on agglomerative clustering of morphological features

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Biosignals like electrocardiograms (ECG) are frequently corrupted by artifacts from physiological or electrical sources.
• Existing noise detection methods, including thresholding and adaptive filtering, have limitations in comprehensively identifying diverse artifact patterns.

Purpose of the Study:

• To introduce a novel time series clustering algorithm for robust noise and artifact detection in ECG signals.
• To evaluate the performance and generalizability of the proposed clustering method across various datasets.

Main Methods:

• Feature extraction to characterize ECG signal morphology and temporal behavior.
• Application of an agglomerative clustering approach to group signal samples based on extracted features.
• Validation using multiple diverse ECG datasets.

Main Results:

• The algorithm demonstrated high performance in detecting noisy patterns and artifacts.
• Achieved a sensitivity of 88%, specificity of 92%, and accuracy of 91%.
• The method proved independent of specific record characteristics, indicating broad applicability.

Conclusions:

• The proposed time series clustering method offers an effective approach for noise and artifact detection in ECG.
• This technique enhances signal denoising accuracy and can be adapted for signal classification tasks.
• Morphological clustering provides a robust foundation for biosignal artifact identification.