Multimodal Adaptive Analysis And Application Research Of Pulse Wave Based On Information Entropy

Pulse wave is formed by the stimulation of heart beat and the reflection of vascular tissue,which contains rich information about human physiological and pathological status,and is an important class of physiological signals.Photoplethysmography(PPG)signal is a commonly used pulse signal that can be used to calculate important physiological parameters such as blood oxygen saturation,respiratory rate,heart rate,and so on.It is widely used in clinical medical monitoring and family health monitoring.With the development of machine learning and deep learning technology,a large amount of high-quality signal data needs to be prepared for relevant experiments to achieve the improvement of learning and training effects.People have higher requirements for the quality of data sets used in experiments,which requires signal preprocessing to become more efficient,more accurate,and have a certain degree of adaptive ability.PPG signals are susceptible to various noises during the acquisition process,and there are differences in pulse morphology among different individuals.Different situations should be considered in the noise reduction and outlier screening of PPG signals,and adaptive preprocessing mechanisms need to be implemented.In this study,ensemble empirical mode decomposition(EEMD)algorithms was used to conduct adaptive noise reduction and outlier screening research for PPG signals,combining the morphological characteristics and noise distribution of PPG signals.This article applies sample entropy(SpEn)algorithm to the analysis of pulse waveform characteristics,combining with traditional Chinese medicine pulse diagnosis theory,to explore meaningful pulse characteristics.The main research results of this article are as follows:1.Fingertip PPG pulse analysis dataset collation.This article divides the abnormal data in PPG signals into three categories:blank invalid signals,amplitude abnormal signals,and periodic abnormal signals.Label the PPG signal for valid and invalid conditions.The effective PPG signals are further divided into PPG signals with standard pulse shapes and PPG signals with concealed pre-dicrotic pulse waves or dicrotic pulse waves according to the morphological differences of pulse waves.2.Adaptive noise reduction and outlier screening methods for PPG signals.EEMD algorithm can adaptively decompose signals into multiple intrinsic mode functions(IMF).In this paper,based on the EEMD algorithm and Pearson product-moment correlation coefficient to determine the boundary between noise and effective signal,an adaptive signal denoising method is implemented,which effectively removes noise while preserving the morphological information of the original signal.The IMF data sequence with the highest correlation with the original PPG signal is used,and an adaptive threshold is set to filter abnormal data segments based on the fluctuation of the original PPG signal peak and period.The average recognition accuracy obtained on the independently collected finger PPG pulse signal and the MIMIC-Ⅲ data set PPG pulse data set is 94.41%.3.Extraction of peak characteristic components of pulse signals based on sample entropy.This paper analyzes the sample entropy of PPG signals with two different pulse wave shapes,and the statistical results show that there are significant differences in the average level of entropy between the two.In further research,based on the distribution of sample entropy and Pearson correlation coefficient in the IMF,this paper achieves the extraction of two signal components that contain important pulse peak feature information.