Classifying cardiac biosignals using ordinal pattern statistics and symbolic dynamics.

Area of Science:

• Biomedical signal processing
• Time series analysis
• Machine learning for healthcare

Background:

• Effective biomedical signal classification relies on selecting appropriate features (parameters/biomarkers).
• Established methods often use heart rate variability (HRV) parameters derived from beat-to-beat intervals.

Purpose of the Study:

• To compare the discriminative power of ordinal pattern statistics and symbolic dynamics against traditional HRV parameters.
• To evaluate features for distinguishing patients with congestive heart failure (CHF) from healthy controls using beat-to-beat time series.

Main Methods:

• Analysis of beat-to-beat time series data from CHF patients and a healthy control group.
• Evaluation of individual and paired features, including ordinal pattern statistics (with time lag) and symbolic dynamics.
• Comparison with established heart rate variability parameters.

Main Results:

• Ordinal pattern statistics, especially when sampled with an additional time lag, demonstrate significant discriminative power.
• These novel features show comparable or superior performance to traditional HRV parameters in the CHF classification task.

Conclusions:

• Ordinal pattern statistics represent a promising and efficient feature set for biomedical signal classification.
• The inclusion of a time lag in sampling ordinal patterns enhances their utility for distinguishing physiological states.