Research On Noninvasive Blood Pressure Intelligent Detection Method Based On PPG And ECG Signal

Blood pressure is an important indicator of cardiovascular disease diagnosis and daily health monitoring.Mastering the fluctuation state of blood pressure can prevent and treat diseases in time.Hypertension is a key risk factor for cardiovascular diseases such as cerebral infarction and heart failure,and it shows a younger development trend.Conscious blood pressure measurement in daily life,early detection and intervention for abnormal changes in blood pressure,can reduce the risk of related diseases.Therefore,this paper designs a networked non-invasive blood pressure detection system.The system collects the photoplethysmography(PPG)and electrocardiogram(ECG)of the human body’s buttocks,and combines human body characteristic data to establish a blood pressure optimization algorithm based on machine learning methods,to achieve continuous non-invasive blood pressure detection.The main researches are as follows:(1)Firstly,in order to improve the utilization of signal and obtain high-quality physiological signal,this paper proposes a high-quality signal extraction algorithm based on preprocessed signal.For the PPG signal,a Butterworth band-pass filter is used to eliminate the high-frequency interference in the PPG signal,and then the Variational Mode Decomposition(VMD)algorithm is used to eliminate the baseline drift in the PPG signal.For ECG signal,VMD algorithm and wavelet transform are combined to denoise.The ECG signal is decomposed based on the VMD algorithm,the baseline drift modal components are removed,and the effective signal component and the noisy signal component are divided according to the correlation with the original signal,and then the noise signal component is denoised by the wavelet transform,and finally the finite signal component is denoised.Reconstructed with the denoised signal to obtain a denoised ECG signal.The preprocessed signal is used for high-quality signal extraction,and the algorithm performs waveform shape and threshold judgment based on the detected signal feature points,and obtains a high-quality signal with a certain period.In order to avoid the influence of differences in human characteristics on the accuracy of blood pressure prediction,the correlation between human characteristic parameters(age,gender,height,weight,body mass index,heart rate)and blood pressure was analyzed,and a multivariate linear blood pressure prediction model was initially established combined with pulse wave transit time(PWTT).(2)Next,the time-domain characteristic parameters of the PPG signal are extracted from the high-quality signal,and the physiological characteristic parameters of the human body,the preliminary blood pressure prediction value and the blood pressure standard value measured by the mercury sphygmomanometer are added to construct a sample data set of the blood pressure prediction model.To reduce the computational complexity of the model,a random forest method is used for feature selection.On this basis,a regression blood pressure prediction model based on XGBoost was established.In order to further improve the prediction accuracy of the model,Bayesian Optimization(BO)algorithm was used to optimize the hyperparameters of XGBoost and compared with other machine learning models.analyze.The model performance was evaluated using mean absolute error(MAE),coefficient of determination(R~2),and accuracy rate(AR).The experimental results show that the results of systolic and diastolic blood pressure obtained in the BO-XGBoost model for the test set are:3.16mm Hg and 3.32mm Hg.The mean absolute error is less than 5mm Hg,which meets the standard requirements of the American Association for the Advancement of Medical Devices(AAMI).(3)Finally,the blood pressure detection algorithm is combined with specific practice to establish a networked non-invasive blood pressure detection system,which is explained from the hardware and software parts.The hardware part mainly includes three modules:PPG and ECG signal acquisition module,WLAN communication module.The software part mainly introduces three aspects:data communication,algorithm realization and client display.In order to verify the blood pressure detection performance of the system,based on the model experiment simulation verification using the sample data set,4 testers with different physiological characteristics were randomly selected to use the system to measure.The mean absolute errors of systolic and diastolic blood pressure obtained by the four testers were 2.42mm Hg and 1.96mm Hg,respectively,which reached the error range established by AAMI.At the same time,the effectiveness and practical application of the networked non-invasive blood pressure detection system proposed in this paper for blood pressure detection are illustrated.