Blood Pressure Estimation Based On Artificial Intelligence

As we all know,hypertension is the main cause of many serious diseases,such as cardiovascular diseases and kidney disease.Compared with other diseases,cardiovascular disease has always been the main cause of human death.Therefore,how to measure human blood pressure conveniently,continuously,accurately,and safely is of extraordinary significance for us to prevent the occurrence of cardiovascular diseases and to timely discover the important clues of suffering from cardiovascular diseases.In daily life,we can be divided into noninvasive and invasive methods to predict blood pressure.For invasive methods,continuous blood pressure measurement using arterial catheter is the"gold standard"in clinic.However,it is an invasive measurement that predisposes the patient to infection with a potential risk of complications during the measurement process.For non-invasive methods,auscultation is generally used.However,this method is not only inconvenient,but also cannot achieve continuous measurement.Aiming at the problems of traditional blood pressure measurement methods,a non-invasive and continuous blood pressure measurement method based on photoplethysmography(PPG)is proposed in this thesis.Firstly,for the original PPG signal,this thesis sorts the amplitude coefficients of the PPG signal after discrete cosine transform according to the absolute value from large to small,and then selects the amplitude coefficients from large to small through the adaptive threshold.After the selected amplitude coefficients are close to the threshold,the discrete cosine inverse transform is performed to achieve the effect of denoising.Then,the time domain and frequency domain features are extracted from the denoised PPG signal as the input of the subsequent blood pressure model.What's more,the classification models of support vector machine,random forest and k-nearest neighbor method are fused.The majority obeys the minority as the decision rule.Using the fusion model,the feature vectors are divided into two categories:normal blood pressure and prehypertension blood pressure.Finally,three blood pressure values predicted by support vector regression model,random forest regression model and L₁ norm based linear regression model were used in this thesis to derive a final more accurate blood pressure value by L₂ norm based linear regression model.The final experimental results show that the fusion model method in this thesis has higher classification accuracy and better regression accuracy than a single classification model as well as a single regression model respectively.