Research On Cuff-less Blood Pressure Detection Method Based On PPG Signal

Blood pressure is an important physiological indicator in cardiovascular disease diagnosis.Regular measurement of blood pressure can help patients avoid more serious cardiovascular complications.The cuff-based sphygmomanometer needs to put the cuff sensor firmly on the arm,rendering subjects uncomfortable when blood pressure needs to be continously monitored.Results show that the cuff-less blood pressure estimation can be realized by extracting features related to blood pressure from the photoplethysmography(PPG).But the PPG signal is easy to be interfered,which makes it difficult to accurately extract features.In addition,the artificial features can not fully describe the complex relationship between PPG and blood pressure.In this paper,the key technology of cuff-less blood pressure detection is studied.The main research results are as follows:(1)In this paper,the blood pressure estimation method based on the feature parameters of pulse wave is deeply studied,the physiological relationship between pulse wave and blood pressure is analyzed,and a blood pressure estimation method based on the feature parameters of pulse wave is proposed.This method uses wavelet transform and threshold filtering to preprocess PPG signal.Aiming at the problem of inaccurate feature extraction of PPG signal,a feature point extraction method combining differential threshold method and amplitude extremum method is proposed,which can extract features more accurately.The waveform feature of PPG and its first and second derivatives are analyzed and extracted as the input of subsequent blood pressure model.In addition,the performance of several different machine learning methods in blood pressure estimation is analyzed and compared.Finally,gradient boosting regression tree(GBRT)is selected to build a blood pressure estimation model to complete the cuff-less blood pressure estimation.(2)Aiming at the problem that the manually extracted waveform features are difficult to express the complex relationship between blood pressure and PPG signal,this paper proposes a blood pressure estimation method based on latent space features.The purpose of this method is to make up for the shortage of manual feature extraction,and to further mine the information related to blood pressure contained in pulse wave,so as to improve the accuracy of blood pressure estimation.Firstly,the single cycle PPG and its first and second derivatives signals are extracted,and the distorted single cycle waveform is removed by using the feature filtering method.Subsequently,the high-order cross features of single period PPG signal and its first and second derivatives are extracted by using GBRT,and the deep features in the temporal-spectrum graph of PPG signal are extracted by using convolutional neural networks(CNN).Then,the latent space features are input into support vector regression(SVR)to build blood pressure estimation model.The model is validated on 12000 samples of UCI blood pressure database,and the experimental results show that the proposed method is superior to the latest blood pressure estimation method.(3)A physiological parameter detection system based on PPG is designed.The physiological parameters of blood pressure and heart rate can be measured by inputting the collected PPG signal.The system includes four modules: signal reading module,preprocessing module,feature extraction module and physiological parameter detection module,and the software system is built based on python.