Heart Rate Detection using a Contactless Bed Sensor: A Comparative Study of Wavelet Methods

Contactless monitoring of vital signs, e.g., heart rate (HR) attracts the researcher's attention due to its convenience and affordability. Among others, under-mattress ballistocardiogram (BCG) sensors have proved effective for contactless remote monitoring of HR. Nevertheless, HR detection from BCG sensors is a challenging task because the signal morphology can vary between and within-subjects. In this paper, we studied the potential of two wavelet-based methods, i.e., the multiresolution analysis of the maximal overlap discrete wavelet transform (MODWT-MRA) and continuous wavelet transform (CWT) for HR detection via a microbend fiber optic sensor (MFOS). BCG signals were gathered from ten sleep apnea patients during overnight polysomnography (PSG) study. The MFOS was placed under the bed mattress and the PSG electrocardiogram (ECG) signals were used as a reference to evaluate the proposed HR detection algorithms. Overall, CWT with derivative of Gaussian provided (Gaus2) slightly better results compared with the MODWT-MRA, CWT (frequency Bspline), and CWT (Shannon). Across the ten patients, the mean absolute error, mean absolute percentage error, and root mean square error metrics were as follows: 4.71 (1.22), 7.58% (2.17), and 5.58 (1.20), and 4.71 (1.07), 7.61% (1.65), and 5.59 (1.02), for Gau2 and MODWT-MRA, respectively. That said, the total precision for MODWT-MRA, i.e., 80.22 (19.01) was higher than Gaus2, i.e., 78.83 (17.84).