Effect of embedding dimension on complexity measures in identifying Arrhythmia.

Area of Science:

• Cardiology
• Biomedical Signal Processing
• Data Science

Background:

• Approximate entropy (ApEn) and Sample entropy (SampEn) are standard for Heart Rate Variability (HRV) analysis but rely on fixed parameters (embedding dimension m, tolerance r).
• Fixed parameter assumptions can lead to inaccurate HRV analysis.
• Newer entropy measures like Permutation entropy (PE), Fuzzy entropy (FuzzyEn), and Distribution entropy (DistEn) address limitations with the tolerance parameter (r).

Purpose of the Study:

• To investigate the impact of embedding dimension (m) on the classification performance of various entropy measures for HRV data.
• To evaluate the stability and efficiency of different entropy measures in classifying normal versus arrhythmic heartbeats.

Main Methods:

• Scrutinized four entropy measures (ApEn, SampEn, PE, DistEn) at embedding dimensions m = 2, 3, 4, and 5.
• Utilized normal and arrhythmic RR interval data with lengths from 50 to 1000.
• Assessed classification performance using Area Under the ROC Curve (AUC).

Main Results:

• Distribution entropy (DistEn) and Permutation entropy (PE) demonstrated superior performance in classifying arrhythmic data, with AUCs up to 0.94 and 1, respectively.
• PE performance showed instability with increasing data length (N) for m > 3.
• DistEn exhibited the most consistent and efficient classification performance, with minimal AUC variation (Δ ≤ 0.03) across different m values.

Conclusions:

• Distribution entropy (DistEn) is a highly stable and efficient measure for arrhythmia classification in HRV analysis, irrespective of the embedding dimension.
• Permutation entropy (PE) is effective but can be unstable with varying data lengths, especially at higher embedding dimensions.
• Careful selection of the embedding dimension (m) is crucial for optimizing the stability and efficiency of entropy-based HRV classification.