Abstract
Sleep apnea is a prevalent respiratory disorder that significantly impacts cardiovascular health. Traditional diagnosis via Polysomnography (PSG) is resource-intensive and intrusive. This study proposes a novel, non-invasive approach using ECG-derived respiration (EDR) and complex network theory. By transforming physiological signals into Weighted Limited Penetrable Visibility Graphs (WLPHVG), researchers identified specific network signatures—including characteristic path length and entropy—that effectively distinguish between wakefulness and various sleep stages. These findings suggest a viable path toward low-cost, wearable sleep monitoring for apnea patients.
Introduction
Sleep apnea is characterized by repeated interruptions in breathing during sleep, leading to risks such as hypertension and cardiovascular disease. The current "gold standard" for diagnosis, Polysomnography (PSG), requires patients to undergo overnight monitoring with an array of sensors, which is often expensive and disruptive.
There is a growing interest in Cardiopulmonary Coupling (CPC) analysis, which utilizes simple ECG and respiration data. However, physiological signals during sleep are inherently non-linear and dynamic, making traditional statistical methods insufficient. This research addresses the need for sophisticated mathematical tools to extract meaningful features from these complex signals to enable home-based monitoring.
Methodology
The researchers developed a multi-step pipeline to transform heart signals into mathematical networks:
- ECG-Derived Respiration (EDR): Instead of using a dedicated breathing belt, the respiratory signal is extracted directly from an Electrocardiogram (ECG) by tracking the heart's electrical axis rotation and heart rate fluctuations during the respiratory cycle.
- Weighted Limited Penetrable Visibility Graph (WLPHVG): This core innovation treats a time series as a geometric landscape. A "Visibility Graph" connects data points (nodes) if they have a line of sight over intermediate points. The "Limited Penetrable" aspect allows lines of sight to pass through a set number of obstacles, while the "Weighted" aspect assigns values to edges based on horizontal and vertical distances between points.
- Network Metrics: Three key parameters are calculated from the resulting graph: Characteristic Path Length ($L$), Weighted Clustering Coefficient Entropy ($E_{C\omega}$), and Weight Distribution Entropy ($E_\omega$).
Key Insights and Results
The study utilized the MIT-BIH Polysomnographic Database, which includes overnight recordings from 16 subjects with sleep apnea. By mapping physiological time series to a network, the researchers identified patterns of regularity and chaos that are invisible in raw waveforms.
Statistical Differentiation
The analysis revealed that the Characteristic Path Length ($L$) is the most sensitive metric. It showed a sharp, continuous decrease as patients moved from wakefulness to light sleep, serving as a clear marker for sleep onset. Furthermore, entropy-based parameters ($E_{C\omega}$ and $E_\omega$) peaked during deep sleep (DS), suggesting that cardiorespiratory patterns become more stable and regular during deep sleep compared to the fragmented patterns seen during wakefulness or apnea-interrupted light sleep.
As shown in the distributions above, there is a significant shift in the mean values of these metrics between stages. For instance, $L$ drops from approximately 7 during wakefulness to 5 during light sleep, demonstrating the metric's ability to capture physiological transitions.
Conclusion and Future Work
The WLPHVG-based features successfully characterize the different stages of sleep in apnea patients. While the characteristic path length $L$ outperformed entropy measures in stage differentiation, all three parameters provided distinct signatures for Wake, Light Sleep, Deep Sleep, and REM stages.
Future work will focus on expanding the sample size beyond the initial 16 subjects to account for demographic variations. Additionally, the researchers aim to integrate these WLPHVG features into machine learning models to develop a fully automated, ECG-only sleep staging system suitable for home-use wearable devices.
Summary updated