Key result
A multi-modal wearable device using a random forest regression algorithm predicted systolic and diastolic blood pressures with a mean absolute error of less than 3.3 mmHg.
Why the study?
Traditional cuff-based blood pressure monitoring is inconvenient, uncomfortable, and does not allow for continuous monitoring to prevent serious cardiovascular complications.
Does a multi-modal wearable device with a machine learning algorithm accurately predict blood pressure without a cuff in human subjects?
Does a multi-modal wearable device with a machine learning algorithm accurately predict blood pressure without a cuff in human subjects?
A novel multi-modal wearable device combining PPG and BioZ sensors with machine learning can accurately estimate blood pressure without a cuff, offering potential for continuous cardiovascular monitoring.
Cuffless wearable BP monitoring shows promise but requires validation; leaves open need for larger trials before clinical adoption.
Blood pressure is a critical risk factor for cardiovascular diseases (CVDs), yet most adults do not monitor it frequently enough to prevent serious complications. This is in part because the traditional cuff-based method is inconvenient, uncomfortable, and does not allow for continuous monitoring. To address these constraints, we developed a unique multi-modal wearable device and used a random forest regression (RFR) algorithm that resulted in a model capable of accurate cuffless blood pressure prediction. This multi-modal device features two photoplethysmography (PPG) sensors and two bioimpedance (BioZ) sensors to measure pulse wave propagation along the radial artery on the wrist. The redundancy in the design enhances prediction accuracy. To validate the device, a novel human subject study protocol was also developed that allows an individual's blood pressure to rise safely and repeatably by more than 40 mmHg (systolic pressure) from baseline measurements. In this study, using multiple pulsatile waveforms from the PPG and BioZ sensors as inputs into the machine learning prediction algorithm, showed that the model had higher accuracy than models using a single sensor. Specifically, the training, validation, and leaving one subject out of data sets all showed mean absolute errors of less than 3.3 mmHg for both systolic and diastolic blood pressures (BPs). While results from this test were promising, a subject-wise evaluation showed variability depending on how well an individual's BP distribution matched the training set. These findings demonstrate the potential for a universal model for cuffless BP estimation, with further validation needed in more diverse populations. Thus, the accompaniment of the RFR model with the multi-modal wearable device offers the potential for robust and continuous blood pressure monitoring, providing a unique and practical solution for long-term cardiovascular health management.
No takes yet. Share an insight, caveat, or question.
Hsiao et al. (2025) studied Blood pressure monitoring. Multi-modal wearable device (PPG and BioZ sensors) with random forest regression algorithm vs. Single sensor models was evaluated on Mean absolute error for systolic and diastolic blood pressures. A multi-modal wearable device using a random forest regression algorithm predicted systolic and diastolic blood pressures with a mean absolute error of less than 3.3 mmHg.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: