Respiration rate extraction from ECG signal via discrete wavelet transform

Abstract

In many vital signs monitoring systems, the ECG signal is detected but not the respiration rate. Nonetheless, the need for continuous, noninvasive, and reliable respiratory rate monitoring has long been recognized. We describe here a signal processing technique based on wavelets that derives the respiratory waveform from ordinary single-lead ECG. The proposed method first decomposes the ECG signal with the DWT up to its 9th or 10th level. Two correction rules are used for determining the decomposition level to be used according to the respiration frequency. Then the signal is reconstructed just one level using the high frequency reconstruction filter of the DWT in order to remove the coefficient Approximations (cA) components. Finally, a simple threshold is applied and a peak detection algorithm is used after obtaining the coefficients Details (cD) reconstruction. The results are interpreted as individual respiration movements. From there the respiration rate is extracted. The algorithm results were compared to the simultaneously recorded ECG and respiration recordings of the PhysioNet/PhysioBank Fantasia database. An absolute average error of 6.8% was obtained, considered highly acceptable for ambulatory patient monitoring.