Research On Prediction Method Of Continuous Blood Pressure Based On Pulse Wave Characteristic

At present,the number of patients with hypertension remains high,and hypertension has become an important problem that seriously endangers global public health.At the same time,hypertension will not only directly lead to cardiovascular diseases,but also lead to renal failure,stroke and other diseases,which will have an irreversible impact on patients’health.The seriousness of the problem of hypertension is not only due to the high prevalence rate and late mortality rate of hypertension,but more importantly,most patients with hypertension do not know their blood pressure in their daily lives,and the awareness rate and control rate of hypertension are very unsatisfactory.Hypertension is usually a process of gradual development over time,and regular blood pressure measurements can help patients identify possible hypertension in time.If blood pressure can be detected in real time,it will be of great significance for early diagnosis of hypertension,reducing the risk of cardiovascular and cerebrovascular diseases,and even improving the quality of life and life expectancy.In this paper,the relationship between pulse wave waveform and blood pressure is studied.First,the original pulse wave signal is filtered by Butterworth filter,and then the processed signal is divided into periods,and two methods for predicting blood pressure are established.The first method is to predict blood pressure based on the morphological characteristics of pulse wave and common machine learning regression algorithms(multiple linear regression model,support vector machine regression model and random forest regression model).The second is to study the pulse wave period data after division of periods combined with deep learning algorithm(TCN-LSTM regression model constructed by combining time convolution with long-term and short-term memory recursive neural network),and make full use of the time information of pulse waves in each period.The test shows that the absolute value of the average error between the predicted systolic blood pressure and the standard value by the TCN-LSTM regression model based on deep learning is 3.553 mmHg,the absolute value of the average error between the diastolic blood pressure and the standard value is 1.822 mmHg,the predicted systolic blood pressure is 79.1%less than or equal to 5 mmHg,and the predicted diastolic blood pressure is 94.8%less than or equal to 5 mm Hg.The results show that the fitting effect of TCN-LSTM regression model is better than that of machine learning model by extracting features,and the prediction effect and stability of the model are also stronger.The test results of the TCN-LSTM regression model fully meet the grade standard set by the British Hypertension Society,and can effectively measure blood pressure through pulse wave signals.