Multi-Frequency Model Fusion for Robust Breathing Rate Estimation

Breathing rate (BR) is an important physiological indicator monitored for a variety of chronic diseases. Since direct measurement devices are often cumbersome to wear, we hence aim to obtain an accurate estimation of BR using other monitored signals, such as PPG or ECG. However, derived modulations from these signals are highly dependent on patient and activity type, making the task difficult as to switching among the modulations. We have previously proposed respiration quality index(RQI) based selection method to update the optimal modulation in a realtime manner. A fusion strategy has also been proposed by coupling the RQI with a Kalman smoother to further exploit the sinusoidal waveforms observed from different modulations. In the current study, we further investigate the enhancement of model complexity of the Kalman smoother by introducing multiple frequency dynamics. Performances are compared to reference methods (Pimentel2016, Karlen2013) on the Capnobase Benchmark dataset. In particular, our enhanced KS method achieves a median absolute error and 25−75 percentile range of 0.22(0.16 − 0.64) bpm, as compared with 0.35(0.28 0.89) bpm from our previous KS fusion method and 1.1(0.3 2.6) bpm from the best reference method in the literature (Karlen et al. 2013).