Research On Cuffless Blood Pressure Monitoring Algorithm Driven By Machine Learning

The increase in the prevalence of cardiovascular disease has sounded the alarm bell for medical and researchers in various countries.As a significant cardiovascular physiological parameter,blood pressure can reflect the basic state of cardiovascular system function.The accuracy of blood pressure estimation can greatly improve the probability of predicting cardiovascular disease.The real-time monitoring method of sleeveless belt has the advantages of no harm,good portability and high accuracy,which can provide a very effective means for the early prevention,detection and treatment of cardiovascular disease.Based on the signal quality evaluation model and model and eigenvector model,this paper has proposed a SVM regression blood pressure model based on fusion genetic-contraband mixed strategy and a blood pressure estimation model based on 1 DCNN and LSTM without feature extraction.Summarize the main work and features of this paper are as follows:(1)The theoretical basis of blood pressure estimation,the mechanism of arterial blood pressure formation,the influencing factors,the relationship between PPG、ECG and BP are expounded in detail,and Data sources、model evaluation metrics used in experiments provide a theoretical basis for subsequent model building.(2)Evaluation of blood pressure signals using the pSQI,gSQI,bSQI three indicators proposed by the heuristic fusion based blood pressure signal quality assessment model,EEMD denoising is carried out to improve the accuracy of blood pressure estimation model.In order to solve the problem of time delay in ECG and pulse signal,this paper introduces the pulsatile starting point algorithm to locate the starting and stopping points accurately,and then extract the feature points respectively.PCA are used to construct eigenvectors to determine 5 eigenvalues with a contribution rate of 85-95%standard.(3)a machine learning-based cuff-free blood pressure estimation model is proposed using PCA constructed feature vectors.Aiming at the parameter selection problem of regression network,the genetic-tabu hybrid strategy is used to optimize the model parameters,and the relationship model between feature vector and continuous blood pressure based on support vector regression is established.the estimated standard errors of the model for systolic and diastolic blood pressure were 6.54 and 4.98 mmHg,respectively,which reached the A grade of the BHS standard.(4)To solve the problem that artificial feature extraction is still needed in the previous blood pressure estimation model,based on ECG and pulse signal,This paper constructs a LSTM blood pressure estimation model with the ability of feature self-learning and the advantage of processing timing features to realize the effective estimation of blood pressure.The error of systolic blood pressure and diastolic blood pressure estimation is 1.05 ± 1.02,0.87±0.69 mmHg,which meets the A grade of the BHS.The blood pressure estimation model constructed in this paper realizes the high accuracy of blood pressure estimation results,provides a new way of prevention and treatment of cardiovascular disease,and provides a theoretical basis and technical support for the application of sleeveless blood pressure monitoring technology.