Estimating respiratory rate in freely moving users using independent component and multi-resolution analysis

Non-invasive health monitoring technologies increasingly attract attention as they allow for continuous, comfortable vital sign monitoring. However, contactless sensing of vital signs using radar systems has significant challenges in accurately extracting physiological information from noisy signals, especially when subjects move freely. This study introduces a novel framework that combines Independent Component Analysis (ICA) and Empirical Wavelet Transform (EWT) to estimate respiratory rate (RR) from radar signals under free-movement conditions. ICA automatically selected physiologically relevant components from the radar signals. Subsequently, the Empirical Wavelet Transform served as an adaptive Multi-Resolution Analysis (MRA) technique, effectively decomposing and reconstructing respiratory signals to improve peak detection accuracy. We evaluated the proposed framework using experimental data from ten subjects performing activities that mimic daily life in a living laboratory environment. A TMSi MOBi8 system recorded the reference RR signals simultaneously. Performance evaluation using Pearson's correlation coefficient revealed a strong correlation (r = 0.94) for the best-performing method. At the same time, the Bland–Altman analysis showed a mean error of -0.41 breaths per minute, demonstrating the ICA-EWT framework's effectiveness in estimating RR in freely moving real-world settings. However, addressing issues related to radar placement and signal interference is suggested to improve the method's accuracy.