A Multi-Parameter Fusion Method for Cuffless Continuous Blood Pressure Estimation Based on Electrocardiogram and...

Area of Science:

• Biomedical Engineering
• Physiological Monitoring
• Data Science in Healthcare

Background:

• Blood pressure (BP) is a critical health indicator, but traditional cuff methods offer limited dynamic insights.
• Cuffless BP monitoring promises continuous physiological data for better health management and BP control evaluation.

Purpose of the Study:

• To develop a wearable device for continuous physiological signal acquisition (ECG and PPG).
• To propose a multi-parameter fusion method for noninvasive BP estimation using selected features.
• To validate the effectiveness of feature selection and fusion techniques in improving BP prediction accuracy.

Main Methods:

• Designed a wearable device for continuous electrocardiogram (ECG) and photoplethysmogram (PPG) signal acquisition.
• Extracted 25 features from processed waveforms and employed Gaussian copula mutual information (MI) for feature selection.
• Trained a random forest (RF) model for systolic BP (SBP) and diastolic BP (DBP) estimation, using MIMIC-III for training and private data for testing.

Main Results:

• Feature selection reduced the mean absolute error (MAE) and standard deviation (STD) for SBP and DBP.
• Post-selection, SBP MAE/STD improved from 9.12 ± 9.83 mmHg to 7.93 ± 9.12 mmHg, and DBP from 8.31 ± 9.23 mmHg to 7.63 ± 8.61 mmHg.
• Calibration further reduced SBP MAE to 5.21 mmHg and DBP MAE to 4.15 mmHg, demonstrating significant accuracy improvements.

Conclusions:

• Mutual information (MI) is highly effective for feature selection in BP prediction.
• The proposed multi-parameter fusion method using ECG and PPG shows great potential for accurate, long-term, cuffless BP monitoring.