Unsupervised ensembling of multiple software sensors with phase synchronization: a robust approach for...

Area of Science:

• Biomedical Engineering
• Physiological Signal Processing
• Cardiology

Background:

• Electrocardiogram-derived respiration (EDR) offers a non-invasive method for monitoring breathing.
• Existing EDR algorithms have limitations in signal quality and accuracy.
• Fusion of multiple EDR signals is a potential strategy to enhance performance.

Purpose of the Study:

• To develop and evaluate a novel algorithm for fusing multiple EDR signals into a single, higher-quality respiratory signal.
• To improve the robustness and accuracy of respiratory monitoring using electrocardiogram (ECG) data.

Main Methods:

• Software sensors (EDR algorithms) were identified based on respiratory signal quality index.
• High-quality EDR signals were aligned using phase synchronization with the graph connection Laplacian.
• Fused EDR signals (sync-ensembled EDR) were evaluated on large polysomnogram databases.

Main Results:

• The sync-ensembled EDR algorithm demonstrated statistically significant improvements over existing methods using synchronized correlation (γ-score) and optimal transport (OT) distance.
• Sensitivity analysis confirmed that signal quality selection and phase synchronization are critical for performance.
• The algorithm accurately estimated average respiratory rate.

Conclusions:

• The sync-ensembled EDR algorithm provides a robust and accurate method for extracting respiratory information from ECG.
• Phase synchronization is a theoretically sound and practically effective technique for developing advanced EDR algorithms.