Research On Noninvasive Continuous Blood Pressure Prediction Method Based On PPG Signal

Blood pressure(BP)refers to the lateral pressure acting on the blood vessel wall per unit area when blood flows in the blood vessels,and it is the driving force that promotes the blood flow in the blood vessels.With the improvement of people’s material living conditions,the unhealthy living habits that follow have led to a surge in people with high blood pressure.At present,blood pressure prevention and control have become a national strategy in my country.Non-invasive continuous blood pressure monitoring is of great significance for diagnosing and managing abnormal blood pressure and thus plays a role in early prevention and timely intervention for sudden cardiovascular diseases.The current cuff blood pressure monitoring method requires cuff pressure,which is an intermittent measurement method and cannot meet the needs of long-term continuous blood pressure measurement.With the development of artificial intelligence,machine learning methods have proven their feasibility and potential in various fields.Studies have shown a correlation between pulse waves and blood pressure,which means that blood pressure can be predicted using only a single-channel photoplethysmography(PPG)signal.electrocardiogram(ECG)is used to measure cardiac activity.The records also contain specific blood pressure-related characteristics.Therefore,it is feasible to use machine learning methods to achieve high-precision non-invasive cuffless continuous blood pressure prediction through PPG and ECG signals and can overcome the shortcomings of traditional blood pressure detection methods.However,most of the current non-invasive blood pressure monitoring methods need to improve prediction accuracy.In this thesis,aiming at the problem of insufficient accuracy of blood pressure prediction using PPG and ECG signals,a variety of blood pressure prediction methods are studied and compared.The main work and results of this thesis are as follows:(1)Experiment of blood pressure prediction method based on artificial features.The PPG signal processing method was studied by collecting data from the X3PLUS-HTO2 smartwatch;the PPG,ECG,and Arterial Blood Pressure(ABP)data in the MIMIC-II database published by the Massachusetts Institute of Technology Experiments were carried out,and the artificial extraction of relevant morphological features and time scale features was realized.The machine learning blood pressure prediction method based on artificial features was studied by establishing linear models and various nonlinear models.Among them,the performance of the nonlinear model is higher than that of the linear model,and multiple models meet the standards of the American Association of Medical Instruments(AAMI)and the British Hypertension Society(BHS).(2)Experiment of blood pressure prediction method based on deep learning.Aiming at problems such as information loss in artificially extracted features,a feature extraction method based on deep learning is implemented to predict blood pressure.The blood pressure prediction models using CNN,VGG19,and LSTM were compared and studied.The experimental results show that the prediction performance of VGG19 is higher than that of CNN and much higher than that of LSTM.Moreover,the performance of blood pressure prediction models based on deep learning is generally higher than that of nonlinear models based on artificial features.Aiming at the problem that the prediction accuracy of systolic blood pressure is usually lower than that of diastolic blood pressure,the prediction accuracy of systolic blood pressure is improved by increasing the hidden layer of the convolutional neural network,which proves that the mapping relationship between systolic blood pressure and PPG signal is more high-dimensional and abstract.(3)Experiment on a multi-modal and multi-stage blood pressure prediction method based on deep learning.Considering the advantages of convolutional neural and LSTM networks,a multi-stage blood pressure prediction method is studied.Experiments are carried out in a staged manner so that each neural network can give full play to its advantages.The training set is enhanced by the cutting method.Make the training set better describe possible relationships in the time domain.At the same time,considering the dynamic correlation between systolic and diastolic blood pressure,the prediction accuracy gap between systolic and diastolic blood pressure is narrowed,and the blood pressure prediction performance is further improved.Finally,ECG signal features are added to build a multi-modal and multi-stage blood pressure prediction model.The final experimental results show that this method enhances blood pressure prediction accuracy.In this thesis,by comparing the blood pressure prediction method based on artificial features,the blood pressure prediction method based on deep learning,and the multi-modal and multi-stage blood pressure prediction method based on deep learning,the related techniques suitable for non-invasive cuffless continuous blood pressure prediction are explored.The high-precision prediction of the final systolic and diastolic blood pressure dropped to 3.3922 mm Hg and 2.1551 mm Hg.The results met the AAMI standard and reached the BHS A-level standard.