SVR ensemble-based continuous blood pressure prediction using multi-channel photoplethysmogram

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Continuous blood pressure (BP) monitoring is crucial for cardiovascular health management.
• Existing non-invasive BP prediction methods often rely on single calibration models, leading to potential inaccuracies and overfitting.
• Photoplethysmogram (PPG) signals offer a promising, non-invasive source for BP estimation.

Purpose of the Study:

• To propose a novel continuous, non-occluding blood pressure prediction method using multiple PPG signals.
• To develop an ensemble learning framework that integrates multiple Support Vector Regression (SVR) models for improved BP estimation.
• To reduce mean prediction error and mitigate the risk of overfitting associated with single-model approaches.

Main Methods:

• Utilized an ensemble learning framework comprising multiple Support Vector Regression (SVR) machines for BP prediction.
• Developed a comprehensive feature set from distinct PPG segments, including pulse morphology, heart rate variability (HRV), and pulse wave velocity (PWV).
• Employed calibration models derived from multiple arterial segments, a novel approach for continuous BP estimation.

Main Results:

• Empirical evaluation with 40 volunteers demonstrated superior reliability in estimating both systolic and diastolic BP compared to single-model methods.
• The proposed ensemble method significantly reduced mean prediction error.
• The combined output of the ensemble models exhibited greater stability for both systolic and diastolic BP estimations.

Conclusions:

• The proposed continuous BP prediction method using an ensemble of SVR models and multiple PPG signals offers enhanced accuracy and stability.
• This multi-segment calibration approach represents a significant advancement over existing single-model BP estimation techniques.
• The findings suggest a more robust and reliable non-invasive method for continuous blood pressure monitoring.