Robust estimation of respiratory frequency from exercise ECGs

A method for robustly estimating the respiratory frequency from exercise ECGs is presented. The special characteristics of these recordings, such as the highly non-stationary noise, the exercise-induced QRS morphologic variations, and the dynamic nature of the respiratory frequency during the exercise test, make the classical estimation of a respiratory signal to break down. Our method is based on least-squares estimation of the rotation angles of the heart electrical axis by aligning successive QRS-VCG loops to an adoptively updated reference loop. The respiratory frequency is estimated by spectral analysis of the series of rotation angles using a reference frequency tracking algorithm. The method was evaluated by means of a simulation study. The respiratory frequency estimation error achieved by this method (0.623%±0.316%, mean±SD) was found to be lower than that obtained by a classical method based on QRS areas (3.220%±3.873%).