Automatic estimation of respiration rate from oronasal pressure using Ensemble empirical mode decomposition, Butterworth filter, and data fusion

Abstract

Respiratory rate is an essential indicator of serious illness. Compared to heart rate or systolic blood pressure, its change is more evident, hence is a better means to discern stable patients from those at risks. In addition, the respiration rate is a crucial indication of sleep quality associated with sleep disorders such as obstructive sleep apnea. Oronasal pressure, as a clinical respiratory signal for sleep analysis, can be used in polysomnography both in labs of hospital and homes. Besides, it is often taken as a reference signal in research as opposed to the estimated respiration rate. This study aims to provide an automated respiratory rate estimation system with signals taken from oronasal pressure transducer that can cope with noises and is adaptive to various respiratory frequencies. A robust approach is presented here that employs Ensemble empirical mode decomposition method to remove signal noise, together with Butterworth band-pass filter to obtain the breathing frequency by means of zero-crossing. Among 97.6% of the test data, the study yields a root mean square error of 1.031. Compared to other methods, the current approach provides a more accurate respiration rate estimation in the application of orinasal pressure to sleep analysis.