Noninvasive Continuous Cardiac Output Measurement Based On Peripheral Pulse Wave Signals

With the surge of China’s population,the battle against poverty and the entrance of a well-off society,the incidence of central cerebrovascular diseases in the population has also increased rapidly,and the requirements for the detection and monitoring of cardiovascular functions have become more diversified.Hemodynamic parameters such as Cardiac Output(CO)、stroke output、cardiac index、stroke index、total peripheral resistance and blood pressure have important clinical significance in describing the cardiovascular function of human body.Cardiac output refers to the output of unilateral ventricle per minute,which is an important parameter to characterize the health status of cardiovascular system and an important diagnostic basis for cardiac function and cardiovascular diseases.The accuracy of its measurement is crucial for the treatment of patients with high-risk cardiovascular diseases.At present,the cardiac output testing equipment for more professional and expensive imported medical instruments,does not have portability,measurement accuracy is not high,expensive,also not convenient round-the-clock monitoring,not to facilitate the early screening of CO,this article will give up the traditional cardiovascular system modeling and theoretical formula derivation CO,by extracting the time domain and frequency domain characteristics,and a combination of both to construct the feature vector,implementation is based on deep learning noninvasive cardiac output measurement.This low-cost,easy to operate,routine testing method is of great significance for the prevention and treatment of cardiovascular diseases.The main content of this paper is as follows:(1)Get MIMICII database 202 patients age,gender,cardiac output and extremely corresponding pulse wave form data,such as the waves and corresponding treatment after classification,detection oscillation signal peak,the peak around the peak amplitude of a certain proportion to obtain the corresponding point in time,extraction of the point in time and corresponding increase or decrease the pressure signal,is the systolic blood pressure,average pressure and diastolic blood pressure,systolic blood pressure and diastolic blood pressure difference value of the corresponding related to pulse wave form characteristics of pulse pressure points.(2)Using after extraction of the pulse potter character point are calculated respectively in the time domain characteristics of blood pressure,time characteristics,area features and proportion and so on,and then through the Fourier transform,extract the frequency spectrum characteristics,mainly including several times the frequency of harmonic amplitude,in the end,all the time domain and frequency domain characteristics of combination,build characteristic vector and normalized processing,and weighting matrix multiplication and add the offset,in the end,artificial neural network is used to calculate cardiac output and parameters related to cardiovascular system,The correlation coefficient between the true value and the predicted value predicted by this model is 0.96(on the training set),the correlation coefficient between the true value and the predicted value is 0.88762(on the verification set),the mean of the absolute error between the true value and the predicted value is 0.9963 ml,and the variance is 0.4908 ml.The mean of the relative error was 0.1653 ml,and the variance of the relative error was 0.1553 ml.(3)End-to-end deep learning model is established,combined with a one-dimensional convolutional neural network automatically extract the characteristics of the pulse wave form,to solve due to the differences of the patient’s hand to extract the waveform characteristics of robustness is not strong,and when using the length of memory network to the sequences of regression,finally obtaining CO,the correlation coefficient is 0.70317,but the model on the validation set and test set on poor performance,Later,we used the simulation data set and constantly added the data.After that,we found that the model had good performance in the training set,verification set and test set.(4)The CO were obtained after the long and short time memory network regression and the CO actually measured in the database were investigated through consistency analysis method.The analysis results proved that the consistency between the method and the CO actually measured in the database was good,that is,the method could meet the requirements of cardiac output measurement,The error is within 1ml,which meets the clinical needs.