Estimation Of Respiratory Rate Based On Data Fusion Using Electrocardiogram And Pulse Wave

OBJECTIVE:The monitoring of respiration signal plays an important role both in clinical and home monitoring. The respiratory sensors we often use nowadays, such as strain gauge transducer, temperature sensor, flow sensor, capacitive transducer, would make the subjects feel uncomfortable. It is very likely to estimate respiratory rate by continuously monitoring electrocardiogram (ECG) and pulse wave (PW), because both of them contain certain respiratory component. In this paper, we try to propose a reliable algorithm to estimate respiratory rate using electrocardiogram (ECG) and pulse wave (PW), based on multiple sensor data fusion with Kalman filter and discuss the reliability of this algorithm. It is provided a more convenient and efficient way to detect sleep apnea by continuously monitoring respiratory rate using this method.METHODS:1. Data acquisition. Two group signals were sampled from14volunteers. One group is normal respiration; the other is a simulation of apnea by breath control.Powerlab data recording and analyzing system, manufactured by AD Instruments Company, was used to sample three physiological signals:electrocardiogram (ECG), pulse wave (PW) and abdominal respiration, among which abdominal respiration is considered to be a reference signal. When sampling, silver chloride electrodes were used to detect ECG, the semiconductor piezoresistive HXH-1respiratory sensor (around chest or belly) was used to detect respiratory signal while the piezoelectric MP100pulse transducer was used to detect PW. The sampling frequency is1000Hz in every channel.The sleep apnea data in MIT-BIH Polysomnorgraphic Database of Physionet (on http://www.physionet.org/) are also analyzed in the research. This database was built by Massachusetts Institute of Technology, in USA.2. Data processing. Data processing is performed largely by MATLAB7.0software, included beat detection of ECG and PW, the estimation of respiratory rate from the RR interval series and the R absolute amplitude of ECG, as well as from the beat cycles of PW. After that, multiple sensor data fusion based on signal quality indices and Kalman filter is required, then compare the results with the reference respiratory rates. Performance evaluation is presented, lastly.RESULTS:The results indicate that the fused respiratory rate performs better than those derived from ECG and PW directly. Compared to the reference of a piezoresistive sensor in normal group, the estimation error of data fusion method is (-0.03±2.78) breaths/min. It is much less than both ECG-derived rates (RR interval series:(0.62±3.30) breaths/min; R absolute amplitude:(0.42±3.47) breaths/min) and PW-derived rate (-0.17±2.69) breaths/min). In conclusion, the multiple data fusion method is noise-resisted and suitable for respiratory rate estimation.In the research, we found that respiratory rate can be estimated more accurately in normal group, while it can not follow the tracks of changes of respiratory rate in apnea simulating group. Why ECG and PW can not reflect the information of respiratory rhythm is that ECG and movement of thorax are not influenced by respiratory movement when apnea happens. In addition, when simulating apnea in the second group, people are not easily to keep a status of being peaceful and calm because of blood oxygen desaturation. For these influences on estimation of respiratory rate, the algorithm needs to be improved.Further study could be done on patients instead of healthy people, then discuss the reliability of detection of sleep apnea by this method.CONCLUSION:The paper presented an algorithm to extract respiratory signals from ECG and PW and estimate respiratory rate by AR model. Signal quality indices (SQIs) were acquired based on signal waveform, rhythm and spectral features. Then the respiratory rate was fused based on SQIs and residuals of Kalman filter. The study indicates that the fused respiratory rate shows a high relevance with reference respiratory rate. In conclusion, the multiple data fusion method is noise-resisted and suitable for respiratory rate estimation.