Research On Blood Pressure Measurement And Prediction Method Based On Machine Learning

Blood pressure is one of the important physiological parameters of the human body,the change of blood pressure value can accurately reflect a person’s healthy state of cardiovascular system.Accurate measurement of human blood pressure data and prediction of changes in human blood pressure are of great significance to the prevention and diagnosis of disease in clinical practice.Blood pressure measurement can be divided into invasive blood pressure measurement and non-invasive blood pressure measurement.Non-invasive blood pressure measurement has become an important research direction for continuous blood pressure measurement,because of its"non-traumatic" and "wide application".The fluctuation of human pulse wave is derived from the beating of the heart,and it is an important carrier of the vital signs of human body.Information on human blood pressure extracted from pulse waves can be used as the basis for clinical diagnosis and treatment.In the field of noninvasive continuous blood pressure measurement,studying the method of blood pressure measurement based on pulse wave has become a hot spot.Blood pressure measurement method can be based on pulse wave velocity and pulse wave characteristic parameters.Because method based on pulse wave velocity need to collect two signals at the same time,the measuring devices and environment are high-demand and complex.Therefore,blood pressure measurement based on pulse wave characteristic parameters has a wider application.With the development of big data technology and artificial intelligence,using machine learning technology and lots of clinical data to solve medical problems intelligently is beginning to take shape.Therefore,on the basis of fully understanding the generation process and analyzing the characteristic parameters of pulse wave,in this thesis,machine learning is used to measure and predict blood pressure.In this thesis,the main research contents are the following aspects.First of all,this thesis analyzes the basic characteristics of pulse wave in depth,and explains the principle for extracting blood pressure information from pulse wave signals.At the same time,the knowledge of machine learning and the characteristics of several typical regression models are introduced in detail.Secondly,this thesis introduces how to get pulse wave signals from the MIMIC and how to preprocess the signals,and also compares the denoise effect of the empirical modal decomposition method and the ensemble empirical modal decomposition method.The experiments show that the ensemble empirical modal decomposition method has a better effect on reducing noise.Thirdly,this thesis puts forward the establishment method of blood pressure calculation model based on machine learning.Four typical regression models are used to establish the relationship between pulse wave characteristic parameters and human blood pressure.RMSE and MAE are two indicators for evaluating model performance.Relevant experiments can prove that the blood pressure measurement model based on KNN has the highest accuracy under conditions where the workload is essentially the same.Finally,this thesis presents a blood pressure prediction model based on recurrent neural network.On this basis,a LSTM-BI blood pressure prediction model is proposed which combines LSTM network and background information.After comparing RNN,LSTM,RNN-BI and LSTM-BI,it can be seen that the LSTM-BI prediction model based on background information has the best blood pressure prediction performance.