Continuous Estimation of Blood Pressure During Physical and Mental Activity using ECG and PPG

Cuffless blood pressure monitoring has emerged as a promising approach for hypertension management. To develop a new noninvasive continuous monitor, a clinical trial collected electrocardiogram, photoplethysmography and blood pressure signals from 15 healthy subjects. A custom physiological measurement platform, designed by CEA-Leti, recorded synchronised ECG and PPG signals at a sampling frequency of 1 kHz. To evaluate the accuracy of blood pressure estimation, a medical-grade blood pressure meter monitored the brachial systolic and diastolic blood pressure every 20 s. A commercial amplifier connected to the previous device provided the blood pressure at 1 kHz. A combination of physical activities (wall sit and pilates ring), stress (mental calculation) and guided relaxation allowed measuring a large range of intra-subject blood pressure variations. To select reliable values for model training, ECG peaks and PPG windows were filtered based on quality and consistency criteria. Descriptive variables were added in order to customize blood pressure estimation with a population model for a given individual. After pre-processing and feature extraction, the resulting dataset included 21751 rows and eight variables. To estimate the blood pressure, several machine learning algorithms were compared, showing promising results with Random Forest and Adaboost methods. Finally, this new dataset highlights the difficulty of estimating occasional high systolic blood pressure values, in particular above 160 mmHg.