A scalogram tensor decomposition based ECG quality assessment.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Automated diagnosis of cardiovascular diseases relies heavily on high-quality electrocardiogram (ECG) data.
• Noisy ECG records can lead to false alarms and increased physician strain.
• Developing automatic noisy ECG rejection mechanisms is essential for reliable clinical decision support.

Purpose of the Study:

• To develop and evaluate a novel system for automatic rejection of noisy ECG records.
• To improve the reliability of automated cardiovascular disease diagnosis by ensuring data quality.
• To reduce the burden on physicians by filtering out unanalyzable ECG signals.

Main Methods:

• Utilized scalograms derived from 12-lead ECG signals, stacked into a 3-way tensor.
• Applied Tucker tensor decomposition to extract latent features from the core tensor.
• Employed a RusBoost ensemble classifier for ECG record classification and tested using the PhysioNet challenge 2011 dataset with five-fold cross-validation.

Main Results:

• Achieved a high accuracy of 92.4% in rejecting noisy ECG records.
• Demonstrated strong performance with a sensitivity of 87.1% and specificity of 93.5%.
• The combination of scalogram and Tucker tensor decomposition proved effective for ECG quality assessment.

Conclusions:

• The developed system effectively rejects unanalyzable ECG records, enhancing diagnostic reliability.
• Scalogram analysis combined with Tucker tensor decomposition offers a competitive approach for ECG quality evaluation.
• This method shows significant potential for real-world application in automated ECG quality assessment.