Liu Shing-Hong, Wu Bo-Yan, Zhu Xin, Chin Chiun-Li
Department of Computer Science and Information Engineering, Chaoyang University of Technology, Taichung 41349, Taiwan.
Department of AI Technology Development, M&D Data Science Center, Institute of Integrated Research, Institute of Science Tokyo, Tokyo 101-0062, Japan.
Sensors (Basel). 2024 Dec 7;24(23):7830. doi: 10.3390/s24237830.
Blood pressure (BP) measurement is a major physiological information for people with cardiovascular diseases, such as hypertension, heart failure, and atherosclerosis. Moreover, elders and patients with kidney disease and diabetes mellitus also are suggested to measure their BP every day. The cuffless BP measurement has been developed in the past 10 years, which is comfortable to users. Now, ballistocardiogram (BCG) and impedance plethysmogram (IPG) could be used to perform the cuffless BP measurement. Thus, the aim of this study is to realize edge computing for the BP measurement in real time, which includes measurements of BCG and IPG signals, digital signal process, feature extraction, and BP estimation by machine learning algorithm. This system measured BCG and IPG signals from a bodily weight-fat scale with the self-made circuits. The signals were filtered to reduce the noise and segmented by 2 s. Then, we proposed a flowchart to extract the parameter, pulse transit time (PTT), within each segment. The feature included two calibration-based parameters and one calibration-free parameter was used to estimate BP with XGBoost. In order to realize the system in STM32F756ZG NUCLEO development board, we limited the hyperparameters of XGBoost model, including maximum depth (max_depth) and tree number (n_estimators). Results show that the error of systolic blood pressure (SBP) and diastolic blood pressure (DBP) in server-based computing are 2.64 ± 9.71 mmHg and 1.52 ± 6.32 mmHg, and in edge computing are 2.2 ± 10.9 mmHg and 1.87 ± 6.79 mmHg. This proposed method significantly enhances the feasibility of bodily weight-fat scale in the BP measurement for effective utilization in mobile health applications.
血压测量对于患有心血管疾病的人来说是一项重要的生理信息，比如高血压、心力衰竭和动脉粥样硬化患者。此外，老年人以及患有肾脏疾病和糖尿病的患者也建议每天测量血压。无袖带血压测量技术在过去十年中得到了发展，它让使用者感觉舒适。目前，心冲击图（BCG）和阻抗体积描记图（IPG）可用于进行无袖带血压测量。因此，本研究的目的是实现用于血压测量的实时边缘计算，其中包括BCG和IPG信号的测量、数字信号处理、特征提取以及通过机器学习算法进行血压估计。该系统通过自制电路从体脂秤测量BCG和IPG信号。对信号进行滤波以降低噪声，并按2秒进行分段。然后，我们提出了一个流程图来提取每个段内的参数——脉搏传输时间（PTT）。该特征包括两个基于校准的参数和一个无需校准的参数，用于通过XGBoost估计血压。为了在STM32F756ZG NUCLEO开发板上实现该系统，我们限制了XGBoost模型的超参数，包括最大深度（max_depth）和树的数量（n_estimators）。结果表明，基于服务器计算的收缩压（SBP）和舒张压（DBP）误差分别为2.64±9.71 mmHg和1.52±6.32 mmHg，而在边缘计算中分别为2.2±10.9 mmHg和1.87±6.79 mmHg。该方法显著提高了体脂秤在血压测量中的可行性，以便在移动健康应用中有效利用。
