Blood Pressure Estimation Based on Photoplethysmography: Finger Versus Wrist

Abstract

The photoplethysmogram (PPG) signal is an attractive candidate for unobtrusive blood pressure (BP) monitoring, as it is widely measured by wrist-worn devices. However, most studies of PPG-based BP estimation techniques have used finger PPG signals. This study compares PPG-based BP estimation when using finger and wrist optical sensors. Subject-specific linear regression models employing pulse transit time, PPG intensity ratio and heart rate as features for BP estimation were trained and tested using PPGs and reference continuous BP values obtained from 22 healthy participants performing 2 cold pressor tests. <tex>$Mean\pm stan$</tex> - dard deviation of differences, and the mean absolute difference between reference and estimated systolic BP values were: <tex>$0.47\pm 10.44$</tex> mmHg and 7.78 mmHg for finger PPG signals; and <tex>$1.05\pm 12.86$</tex> mmHg and 9.69 mmHg for wrist PPG signals. Increases and decreases in systolic BP of at least 10 mmHg were detected with F1 scores of: 0.81 and 0.76 for finger PPG; and 0.75 and 0.61 for wrist PPG. Models performed better with finger PPG signals. Different signal processing approaches were required for finger and wrist signals, indicating that finger-based BP estimation models should not be generalized directly to wrist PPGs.