Study On Multi-temporal Series Fusion For Wearable Respiration Rate Estimation

Respiratory rate(RR)is one of the important vital signs in the human body,and noninvasive and continuous RR monitoring is of great significance for evaluating clinical vital signs deterioration and respiratory diseases.Weak respiratory signals and susceptibility to motion interference have prevented respiratory monitoring algorithms from being widely integrated into wearable devices.At present,Electrocardiogram(ECG)and Photoplethysmographic(PPG)can be more easily obtained through wearable sensors,so this thesis developed a wearable RR estimation algorithm based on respiratory information in ECG and PPG.The main work contents are as follows:Firstly,the respiratory signal is obtained indirectly based on ECG and PPG,and three types of respiratory modulation signals(RMSs)were extracted from the two by feature.Secondly,RMS was evaluated and screened by the respiratory quality index(RQI)and dynamic time warping(DTW).Among them,RQI evaluated RMS based on the proportion of respiratory components,while DTW evaluated RMS based on the similarity between different RMS.Then,variable mode decomposition was innovatively used to filter the filtered RMS.The maximum peak frequency in the decomposed intrinsic mode functions were compared with the peak frequency of the input RMS,and the intrinsic mode function with the smallest difference was selected as the filtered signal.According to RQI and DTW,the RQI-principal component analysis(PCA)fusion strategy and the DTW-PCA fusion strategy were designed respectively.The RR was estimated by these two fusion strategies,and RR estimated effect was verified using the Capno Base dataset.The results showed that the correlation coefficient between the RQIPCA fusion strategy and the reference RR was 0.88,and the correlation coefficient based on DTW-PCA was 0.86.Compared with single RMS and other methods,multi-time series temporal fusion method can obtain higher accuracy,more likely to perform continuous RR estimation,and potentially contain richer respiratory change characteristics,which provides more effective method for wearable RR detection.