Respiratory rate estimation from photoplethysmogram baseline wandering by harmonic analysis and sequential fusion

Photoplethysmogram (PPG) is typically employed to monitor heart rate and estimate respiratory rate (RR). The majority of PPG-based respiratory rate estimation methods necessitate the identification of individual heartbeat peaks or the decomposition of PPG signals, which is laborious and exhibits restricted performance. This paper presents a novel method for estimating respiratory rate, which employs frequency-domain analysis and sequential fusion over time. The proposed method exploits the harmonics of the PPG baseline wandering induced by respiration, seeking the respiratory frequency that maximizes the signal harmonic power. A sequential fusion between signal windows is designed to regularize the results and suppress possible significant errors while evaluating the quality of each fused output. The algorithm is verified using PPG signals from the CapnoBase and BIDMC datasets and yielded mean absolute errors (MAE) of 0.1 and 2.0 breaths per minute for each dataset, respectively, when providing respiratory rate estimates for all windows. Furthermore, the accuracy of the method can be improved by the exclusion of low-quality windows based on the proposed sequential assessment of quality. The application of harmonic analysis and sequential fusion techniques represents a novel approach for estimating respiratory rate from PPG with enhanced performance. The MATLAB code for the processing of the two datasets is accessible via the following link: https://github.com/Chi1988723/PPG-respiratory-rate-estim ation.git.