| Along with social advancement and time’s development,people’s pressure is also increasing.According to the latest data released by the World Health Organization(WHO),since 2005,more than 15 million people have died from cardiovascular and cerebrovascular diseases every year,higher than any other disease.If the current growth rate continues,it is estimated that 25 million patients will die of cardiovascular and cerebrovascular diseases by 2030,and the issue of hypertension has increasingly become a common concern.In recent years,there has also been a large number of portable blood pressure measurement devices on the market,but their prediction process lacks a theoretical sentence,and the accuracy rate is also uneven.In medicine,based on physiological indicators,the process is cumbersome and costly,and it is not convenient.In response to this situation,this article combines traditional data processing methods with deep learning algorithms,proposes relevant data preprocessing schemes based on pulse(finger pulse)signals,and a neural network model based on pulse and blood pressure information.It discovers that pulse contains common features for identifying hypertensive patients,achieving the goal of blood pressure assessment through pulse.The main research results are as follows:1.An improvement has been made to the scheme for solving the residual noise problem in pulse signal preprocessing.This scheme uses wavelet threshold denoising and waveform morphology to jointly process pulse signals.This scheme can remove the excessive width and narrowness of abnormal pulse waves,and eliminate the interference of pulse signals from power frequency noise,EMG noise,and baseline drift.In addition,the scheme also uses a quality detection strategy to filter low quality pulse signals to make their efficiency higher than a specified threshold.Through the implementation of this scheme,the noise of the original pulse signal is removed,and abnormal waveforms are also detected and eliminated,thereby improving the quality and reliability of the pulse signal.2.Establish four commonly used neural network models,and input the processed pulse signals into four neural network models.By comparing the prediction accuracy of the four neural networks,select the most suitable neural network model for blood pressure assessment,namely,LSTM network,and conduct subsequent parameter adjustment.3.Innovatively propose the fusion of convolutional neural network(CNN)and improved LSTM network-GRU to build a topological neural network model of CNN-GRU and use it for blood pressure evaluation of newly collected pulse data.To expand,it is to convert the pulse signal divided by cycle from one-dimensional time domain to two-dimensional time domain through wavelet transform method,completing the enrichment of pulse feature dimensions,and then convert the data into appropriate sizes and successively input them to the convolution module and GRU module,and then input them to the full connection layer,finally achieving the classification of hypertension patients. |
