Causal Decomposition of PPG Signals for Cuffless Blood Pressure Estimation

Area of Science:

• Biomedical Engineering
• Physiological Signal Processing
• Causal Inference

Background:

• Photoplethysmogram (PPG) signals are widely used for cuffless blood pressure (BP) estimation but suffer from limited reliability due to inter-individual variability.
• Existing methods often struggle to isolate the specific PPG signal components that have a direct causal link with BP.

Purpose of the Study:

• To develop a novel framework for decomposing PPG signals to identify components with a strong causal relationship with BP.
• To construct more reliable and interpretable cuffless BP estimation models by leveraging these causally identified PPG components.

Main Methods:

• A causal decomposition framework combining Ensemble Empirical Mode Decomposition (EEMD) with counterfactual inference was proposed.
• PPG signals were decomposed into intrinsic mode functions (IMFs), and counterfactual sequences were generated to quantify causal relationships with BP.
• Causality-based models were constructed using IMFs with significant causal effects and benchmarked against conventional methods (PAT, GBRT, CNN).

Main Results:

• Mid-frequency PPG components (IMF4-6) demonstrated the strongest causal relationship with BP.
• A Convolutional Neural Network (CNN) model utilizing these components achieved superior BP estimation accuracy.
• The CNN model using IMF4-5 components resulted in mean absolute errors of 5.55/3.45 mmHg for systolic/diastolic BP, a significant improvement over using the original PPG signal.

Conclusions:

• Specific frequency bands within PPG signals possess physiologically meaningful causal relationships with BP.
• The proposed causality-driven approach enhances both the accuracy and interpretability of cuffless BP estimation.
• This study establishes a theoretical framework for causal feature selection in BP estimation and a new paradigm for physiological signal analysis.