A non-invasive blood pressure prediction method based on pulse wave feature fusion. (April 2022)

• Record Type: Journal Article
• Authors: Yan, Jianjun; Cai, Xianglei; Zhu, Guangyao; Guo, Rui; Yan, Haixia; Wang, Yiqin
• Is Part Of: Biomedical signal processing and control. Volume 74(2022)
• Publication Date: 2022-04
• Subjects: BP blood pressure -- PWV pulse wave velocity -- GBDT gradient boosting decision tree -- SBP systolic blood pressure -- DBP diastolic blood pressure -- AAMI association for the advancement of medical instrumentation -- BHS British hypertension society -- SVR support vector regression -- MBP mean blood pressure -- PTT pulse transit time -- RF random forest -- LR linear regression -- Ridge ridge regression -- MAE mean absolute error -- STD standard deviation -- ME mean errors; Pulse wave -- Feature fusion -- Gradient Boosting Decision Tree -- Blood pressure prediction; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2022.103523
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlight: In this paper, a feature fusion-based noninvasive blood pressure prediction system is proposed. This system includes pulse signal acquisition, signal processing, feature extraction, feature fusion and blood pressure prediction. The experimental results are fully compliant with international standards. The current study combines the advantages of the Korotkoff sound method and the oscillometry in traditional blood pressure measurement methods. This system is simpler and more comfortable to operate than traditional measurement devices. Pulse wave amplitudes and pressures at three pulse-taking pressures were combined, and fused with time-domain features and PWV features. The results show that feature fusion gives better results than before fusion. Compared multiple machine learning methods, a more suitable blood pressure prediction model with a small sample size was identified. The results were more convincing than using only a single algorithm. The system offers a new approach to non-invasive blood pressure prediction and new ideas for pulse signal researchers. Abstract: To study the non-invasive blood pressure prediction based on pulse wave feature fusion to achieve rapid blood pressure (BP) measurement and improve the measurement accuracy, which provides a new method for the non-invasive blood pressure measurement by wearable devices. From the pulse signals, 82 dimensional features were extracted, including time domain features extracted by the feature point method, ratio features of pulse wave amplitude and pulse-taking pressure fusion, and pulse wave velocity (PWV) features. Feature fusion is performed by feature importance analysis to reduce the dimensionality, and the fused features are used to build blood prediction models based on gradient boosting decision tree (GBDT) regression algorithm. The correlation coefficients between the predicted and actual values of systolic blood pressure (SBP) and diastolic blood pressure (DBP) were 0.93 and 0.92, respectively, which had high correlation, and the mean absolute errors between the predicted and actual values of SBP and DBP were 3.75 mmHg and 3.10 mmHg, respectively, with standard deviations of 5.46 mmHg and 3.93 mmHg, all of which met the overall performance requirements of the association for the advancement of medical instrumentation (AAMI) and British hypertension society (BHS) International electronic blood pressure monitor. This blood pressure prediction model can be better used in the non-invasive blood pressure measurement of wearable devices, which is more convenient and has higher prediction accuracy compared with mercury sphygmomanometer.