Beat-to-Beat Blood Pressure Estimation by Photoplethysmography and Its Interpretation

Area of Science:

• Biomedical Engineering
• Physiological Monitoring
• Machine Learning Applications

Background:

• Blood pressure (BP) is a critical vital sign.
• Estimating absolute BP using photoplethysmography (PPG) is an active research area.
• Existing methods show variability in features, classifiers, and results.

Purpose of the Study:

• To develop and interpret a machine learning model for absolute BP estimation from PPG signals.
• To critically evaluate the model's performance against current literature.
• To provide a generalizable method for interpreting multivariate prediction algorithms.

Main Methods:

• Utilized data from three sources (273 subjects, 259,986 beats).
• Extracted features from PPG signals and their derivatives.
• Employed XGBoost regression for BP estimation and Shapley Additive values (SHAP) for interpretation.

Main Results:

• Systolic BP estimation achieved a Mean Absolute Error (MAE) of 9.456mmHg (r=0.730) with strict subject separation.
• MAE improved to 6.366mmHg (r=0.874) when data separation was adjusted.
• SHAP analysis identified four key PPG-derived features for BP estimation.

Conclusions:

• The study demonstrates a robust machine learning approach for absolute BP estimation using PPG.
• Data selection and training/testing separation critically impact algorithm performance and comparability.
• The developed interpretation method highlights valuable PPG features for BP estimation.