Finding representative electrocardiogram beat morphologies with CUR.

Area of Science:

• Biomedical Engineering
• Data Science
• Signal Processing

Background:

• Electrocardiogram (ECG) analysis often requires dimension reduction techniques.
• Traditional methods may use abstract representations, limiting direct clinical interpretation.
• Identifying representative subsequences is crucial for understanding cardiac health.

Purpose of the Study:

• To apply CUR matrix factorization for dimension reduction in ECG time series.
• To identify clinically relevant beat morphologies and representative subsequences.
• To evaluate the utility of CUR for summarizing ECG data and detecting rare events.

Main Methods:

• Utilized CUR matrix factorization for dimension reduction.
• Applied the method to synthetic and real-world ECG datasets (MIT-BIH, MGH-MF, Incart).
• Incorporated discrete empirical interpolation method (DEIM) and incremental QR factorization.

Main Results:

• CUR factorization successfully identified representative ECG beat morphologies, including rare events.
• The method provided a robust summarization of complex ECG data.
• CUR-selected beats for classification yielded results comparable to existing methods.
• Demonstrated CUR's effectiveness in detecting representative subsequences in quasi-periodic physiological time series.

Conclusions:

• CUR matrix factorization is a valuable tool for ECG data analysis and dimension reduction.
• Its ability to use actual data instances enhances clinical relevance.
• CUR offers a robust approach for summarizing physiological time series and identifying significant events.