Research On Non-invasive Blood Viscosity Intelligent Detection Technology Based On PPG Signal

At present,with the accelerating pace of work and life,more and more people have developed unhealthy work and rest and eating habits.Under such circumstances,the number of people suffering from cardiovascular disease is increasing year by year.Clinical studies have found that the occurrence of cardiovascular disease is accompanied by abnormal changes in blood viscosity,therefore,the detection of blood viscosity is of great significance for the prevention and diagnosis of cardiovascular diseases.In this paper,a non-invasive blood viscosity detection method based on photoplethysmographic pulse wave（PPG）and a machine learning algorithm are used to achieve continuous real-time detection of blood viscosity.In order to solve the problem of baseline drift and high-frequency noise interference in the PPG signal waveform,first,the moving average filtering algorithm is used to remove the high-frequency noise interference in the PPG signal,and then the baseline drift interference in the PPG signal is removed by the wavelet transform,thereby obtaining a high-quality PPG signal.After comparison,the variable window long sliding window detection method with better extraction effect is used as the feature extraction algorithm in this paper to extract the maximum blood volume₈)and the minimum blood volume₈)of the PPG signal,and calculate the average blood volume₈).Then use the extracted eigenvalues to calculate the comprehensive characteristic parameters’of the PPG signal,and fit the parameters with the true value of blood viscosity to construct a preliminary prediction model of blood viscosity,so as to calculate the preliminary predicted value of blood viscosity.In order to solve the problem that the initial estimated value of blood viscosity calculated by the comprehensive characteristic parameter’of the single-parameter PPG signal is easily affected by individual differences,resulting in low accuracy,this paper further proposes the machine learning algorithm of PSO-ELM to improve the prediction accuracy of blood viscosity.First select other blood viscosity influencing factors as:BMI,age,gender,height,weight,these parameters and the initial estimated value of blood viscosity were imported into the ELM network model as input parameters,and an ELM-based blood viscosity prediction model was constructed through training.Then,the ELM network was optimized by the PSO optimization algorithm,and a blood viscosity prediction model based on PSO-ELM was finally constructed,which further improved the prediction accuracy.Compare the prediction results of the PSO-ELM model with other algorithm models,results shows that the prediction result error（MAE）of the blood viscosity prediction model based on PSO-ELM in this paper is 0.14 m Pa·s,the error is smaller and better than other algorithm models.The Bland-Altman consistency analysis method was used,and the experimental results found that most of the points were located in the 95%consistency interval,it shows that the blood viscosity prediction method based on the PSO-ELM model has good consistency with the standard invasive blood viscosity detection method.In order to further improve the work and make the research content of this paper put into practical application,then the network system design of non-invasive blood viscosity detection is carried out in this paper.First,the overall design framework of the system is explained,and then the main hardware and software parts of the networked system are introduced.The hardware part mainly includes PPG signal acquisition module and Wi Fi communication module,and the software part mainly includes data communication,algorithm Implementation,software interface design.Finally,through experiments,it was found that the errors of the blood viscosity prediction results of the four testers were all less than 0.13m Pa·s,which verified the feasibility of the system in practical applications.