Objective: In the present work, a comparative study of different breathing rate estimation methods from PPG signal is proposed. The aim of this comparative study was to select the best algorithm, for respiratory rate estimation, among those already proposed in literature.
Methods: The following methods were implemented and tested on the free access CAPNOBASE database, by segmenting the PPG signal in 32s and in 64s windows: empirical mode decomposition (EMD), EMD combined with principal component analysis, wavelets analysis, respiratory-induced intensity variation analysis (RIIV), respiratory-induced amplitude variation analysis (RIAV) and respiratory-induced frequency variation analysis (RIFV). Performances were then compared to other methods (6 differents approaches) already tested on CAPNOBASE and recently published in IEEE-TBME journal.
Results: Among the tested methods, the best performances were reached by using respiratory induced signals over the IMFs and wavelets. The RIAV signal exceeded other methods in both 64s and 32s signal segments. In this case, the median of the absolute breathing rate error was of 0.57 (0.19-1.71 interquartile range 25th-75th) for the 32s window, and of 1.62 (0.09-1.71) for the 64s window in the 0-0.5Hz frequency range. However, the comparison showed only the algorithm proposed by Khreis et al. in IEEE TBME journal, using Kalman filtering and a data fusion approach (KFF), outperformed the RIAV analysis, with 0.5 median absolute error (0.2-1.1 interquartile range 25th-75th) for the 32s window, and 0.2 (0.1-0.9) for the 64s window.
Conclusion: A total of 6 implemented methods were for the first time compared for evaluation. This comparison was accompanied by a further analysis on 6 other methods where performances were already reported on the CAPNOBASE database. Results clearly emerges that KFF is a promising approach for breathing rate estimation from PPG.