Multi-Source Causal Invariance for Cuffless Blood Pressure Estimation Based on Photoplethysmography Signal Features.

Area of Science:

• Biomedical Engineering
• Physiological Monitoring
• Machine Learning in Healthcare

Background:

• Cuffless continuous blood pressure (BP) monitoring is crucial for personal health management.
• Accuracy challenges arise from diverse and heterogeneous physiological data sources.

Purpose of the Study:

• To develop a multi-source feature selection framework for accurate cuffless BP estimation.
• To leverage Markov blanket theory and causal invariance for robust feature selection.

Main Methods:

• Extracted 218 BP-related photoplethysmography (PPG) features from three heterogeneous datasets.
• Constructed a causal feature set using the Multi-Dataset Stable Feature Selection via Ensemble Markov Blanket (MDSFS-EMB) algorithm.
• Employed Gaussian Copula Mutual Information for robust nonlinear relationship modeling.

Main Results:

• The MDSFS-EMB algorithm demonstrated superior feature selection efficiency and prediction accuracy.
• The selected feature set showed strong generalization capability on an independent external validation dataset.
• Causal relationships between PPG features and BP were innovatively explored across multiple data sources.

Conclusions:

• The proposed framework offers a clinically applicable approach for cuffless BP estimation.
• The MDSFS-EMB algorithm provides an adaptable and accurate solution for BP monitoring.
• This study advances the understanding of PPG-based BP monitoring through causal inference.