Respiration Rate Estimation Using ECG and PPG Data: Open Source Implementations Focused on Wearables

Respiratory rate (RR) is an indicator of various psychological and pathological conditions and needs to be monitored easily and reliably. Emerging trends and recent advancements indicate that wearables are the future of continuous noninvasive vital measurement for healthcare. Numerous recent studies have focused on estimating RR from the electrocardiogram (ECG) or photoplethysmogram (PPG), which can be acquired non-invasively with minimal hardware and without obstructing spontaneous breathing patterns. While easy and flexible in acquisition, such signals may suffer from poor signal quality, which in turn results in unreliable estimation of RR, an issue that can be addressed by evaluating the quality of the obtained estimates. In this work, we propose a novel method for RR estimation based on the cross-correlation function that demonstrates improved estimation properties compared to most existing open-source methods achieving 0-bias and limits of agreement (LOAs) from -5.37 to 5.44 bpm. We propose the use of variance as an estimate quality metric and demonstrate how this approach results in higher accuracy. Two types of fusion are designed and evaluated, stressing the importance of quality evaluation and estimate combination in improving accuracy, reaching LOAs between -1.44 and 0.9 bpm. Finally, we stress the importance of signal duration in estimation accuracy, and demonstrate how longer segments result in lower error, validating this finding on multiple open-source methods. Our results are evaluated on three publicly available datasets, including CapnoBase, a benchmark database on respiration.