A blood pressure measurement unit using two photoplethysmography sensors and a Gaussian process regressor model estimated systolic and diastolic BP with R2 scores of >0.88 and >0.62, respectively.
Observational (n=26)
Can blood pressure be accurately estimated using pulse wave velocity derived from two photoplethysmography sensors combined with machine learning?
PPG-based sensors combined with machine learning can estimate systolic and diastolic blood pressure with high accuracy, offering a potential non-invasive point-of-care solution.
Effect estimate: R2 >0.88 (systolic) and >0.62 (diastolic)
In recent years, advances in biosensing devices have greatly increased the ability of sensing human biological vital signs. These advances have allowed physicians to better assess the health status of patients. Among them, blood pressure (BP) sensing has been the dominant one, showing the most potential for growth. Despite much progress, a rapid, robust, and easy-accessed way for BP sensing is still much needed for the emerging point-of-care market. To tackle this challenge, in this paper, we present a BP measurement unit that is developed based on two photoplethysmography (PPG) sensors from which pulse wave velocity (PWV) of blood flow can be derived. A robust time difference of collected two subsequent PPG waveforms between two heartbeats was used to calculate PWV. The systolic and diastolic BPs from 26 participants were estimated by using the derived PWV as the input of machine learning (ML) models. Upon testing multiple ML models, the Gaussian process regressor achieved the highest R2score of >0.88 and >0.62 for the systolic BP and diastolic BP, respectively. These R2scores are among the best that can be achieved with state-of-art non-invasive BP measurement devices. This work demonstrates that PPG-based sensors for PWV and BP estimation, combined with ML, have a great potential of becoming a complementary way to measure biological vital signs.
Byfield et al. (Fri,) conducted a observational in Blood pressure estimation (n=26). Photoplethysmography (PPG) sensors with machine learning was evaluated on R2 score for systolic and diastolic blood pressure estimation (R2 >0.88 (systolic) and >0.62 (diastolic)). A blood pressure measurement unit using two photoplethysmography sensors and a Gaussian process regressor model estimated systolic and diastolic BP with R2 scores of >0.88 and >0.62, respectively.