Pulse Wave Continuous Blood Pressure Measurement Based On Deep Learning

Nowadays,due to work pressure,irregular life and other factors lead to the high number of patients with cardiovascular disease.In order to prevent and reduce morbidity,it is necessary to study cardiovascular parameters and relations between them,and thus to monitor cardiovascular disease.Blood pressure is an important physiological parameter and can reflect the cardiovascular function of human body.The pulse wave signal contains many human body physiological and pathological information.The method of blood pressure measuring through the pulse wave characteristic parameters of is simple,low cost,high precision,continuous measurement and other advantages,and has broad prospects in application.Based on the theoretical basis of blood pressure measurement,two continuous measurement models of blood pressure were established.One blood pressure model is built by regression analysis according to the traditional method.Another one model is adopted depth learning framework TensorFlow,and used BP neural network training the relationships between blood pressure and characteristic parameters.The blood pressure errors calculated by the two models are within the 3mmHg standard value and the second model error is within 2mm Hg.In line with international standards.The main work of this paper is as follows:Firstly,the pulse wave signals are collected by using the optical heart rate pulse meter(fingertip),and are filtered by wavelet transform method and 5-3 times method.In the extraction of feature points,a hybrid algorithm is proposed to identify feature points which combines threshold difference method,wavelet transform method and differential method.Results show that the algorithm can extract the feature points accurately.Secondly,the model based on the linear regression method is built,which extracting the time domain characteristics of pulse wave parameters,analyzing of the relevance between blood pressure and characteristic parameters,and then using step-wise regression to construct the model.Comparing with the standard blood pressure,the error is within 3mmHg.Finally,a BP neural network model is proposed which takes characteristic parameters as the input of the BP neural network.When constructing the model,in order to eliminate overfitting phenomenon and get optimal model,dropout in TensorFlow was used to make the model slimming.TensorBoard was also used to present the final model of blood pressure neural network so as to clearly describe the blood pressure model constructed in this paper.By estimating BP blood pressure model,the error is less than 2mmHg compared to standard blood pressure,and is more accurate than traditional practice.