A fast and robust algorithm to assess respiratory frequency in real-time

In hospital environment, the real-time monitoring of a patient's respiration can provide valuable information to the physicians. Two examples are given by measuring the load effort during re-educative exercises or by detecting sleep apnoea. Using an oronasal sensor previously demonstrated performs a non-invasive recording of the breath activity. The sensor operates on condensation/evaporation cycles and provides an electrical signal representative of the respiration phase. We present in this paper a robust data acquisition and treatment algorithm that copes with the wide variations seen by the sensor in order to determine the respiratory frequency. To reduce both the computational load and the noise on the estimated frequency, the Median Absolute Deviation (MAD) was used in order to detect spurious signal segments. A real-time estimation of the respiratory state of elderly patients in hospital demonstrates the efficiency of the presented algorithm. Currently, a observation window of 5 seconds is processed in around 0.5 second.