Abstract
Congestive Heart Failure (CHF) is a chronic condition characterized by the heart's inability to pump blood efficiently. Early diagnosis is vital but challenging due to the intermittent nature of early-stage symptoms. This study proposes a novel approach using Heart Rate Variability (HRV) analysis across multiple time scales. By extracting trend features from RR intervals ranging from 5 minutes to 10 hours, the researchers developed a Support Vector Machine (SVM) classifier that achieves 96.7% accuracy. The findings suggest that the dynamic change in HRV measures across time scales offers more robust diagnostic information than traditional single-window snapshots.
Introduction
Heart Rate Variability (HRV)—the physiological variation in time between consecutive heartbeats—is a well-established non-invasive tool for assessing cardiac health. However, clinical standards typically focus on either short-term (5-minute) or long-term (24-hour) recordings. This binary focus overlooks the complex regulatory mechanisms of the autonomic nervous system that operate at intermediate time scales. This research addresses this gap by investigating the optimal time scale for CHF identification and evaluating whether the trend of HRV measures across different scales provides superior diagnostic power.
Methodology
The study analyzed RR interval sequences from two PhysioNet databases: the Normal Sinus Rhythm (NSR) database (54 subjects) and the Congestive Heart Failure (CHF) database (29 subjects).
Processing Pipeline
The analytical framework involves several key stages: preprocessing, feature extraction across seven time scales (5 min, 10 min, 30 min, 1 h, 2 h, 5 h, and 10 h), and classification.
Feature Extraction and Trend Analysis
For each time scale, nine standard HRV measures were calculated, spanning time-domain (e.g., SDNN, RMSSD), frequency-domain (e.g., LF/HF ratio), and nonlinear-domain (e.g., Sample Entropy) metrics. A critical innovation was the introduction of "trend features." The researchers observed that measures like SDNN and the Vector Length Index (VLI) exhibited distinct growth patterns as the observation window increased. These trends were modeled using linear, exponential, and logarithmic functions to capture the "flexibility" of the heart's scaling properties.
Results and Discussion
The study revealed that diagnostic performance varies significantly with the chosen time scale.
The 2-Hour "Sweet Spot"
While 5-minute windows are common, they were found to be insufficient for capturing the full complexity of CHF. The 2-hour scale emerged as the optimal single-scale window, providing a diagnostic accuracy of 94.4%.
Superiority of Multi-Scale Trends
The most significant finding was that trend-based features outperformed exhaustive sets of standard features. By using only three trend parameters (the slope of SDNN and coefficients of the VLI exponent), the SVM model achieved a sensitivity of 93.3%, specificity of 98.3%, and an overall accuracy of 96.7%.
As shown in the figure above, healthy subjects exhibit a much more rapid increase in SDNN as the time scale lengthens compared to CHF patients. This suggests that the heart's ability to scale its variability over time is a primary marker of cardiovascular health.
Conclusion and Future Work
This research demonstrates that multi-scale HRV analysis, particularly focusing on the trends of variability growth, significantly enhances the detection of Congestive Heart Failure. While the results are promising, the study is limited by a relatively small and imbalanced dataset. Future research should validate these findings on larger, more diverse populations and explore the underlying physiological mechanisms—such as circadian rhythms or hormonal cycles—that drive the observed 2-hour optimal window.
Summary updated