Heart Rate Extraction From Neonatal Near-Infrared Spectroscopy Signals

Near-infrared spectroscopy (NIRS) intensity signals provide useful additional physiological information, of which the most prominent one is the pulsatile fluctuation by heartbeats. This allows for the extraction of heart rate (HR), one of the primary clinical indicators of health in neonates. In this study, we propose a novel algorithm, NIRS HR (NHR), for extracting HR from NIRS signals acquired from neonates admitted to the neonatal intensive care unit (NICU). After parental consent, we synchronously recorded NIRS at 100 Hz and reference HR (RHR) at 1 Hz, from ten newborn infants (gestational
$\text {age} =38 \pm 5$
weeks; 3092 ± 990 g). The NHR algorithm consists of two main parts. The first part includes four steps implemented once on the whole NIRS measurement: preprocessing; HR frequency bandwidth determination; interquartile range (IQR) computation; and segmentation. The second part includes three steps implemented on each signal segment: motion artifact detection, signal quality assessment, and HR computation. We compared the NHR algorithm with two existing algorithms. The results showed that the proposed NHR algorithm provides a significantly (
$p < 0.05$
) higher correlation (
$r$
= 99.5%) and lower Bland-Altman ratio (BAR = 3.6%) between the extracted and RHRs, compared to the existing algorithms. Extracting HR from NIRS in a clinical setting of critically ill neonates admitted to neonatal intensive care is feasible. With NIRS and HR combined in a single monitoring system, it is possible to have a perfectly time-synced integrated analysis of cerebral hemodynamics, as well as systemic hemodynamics and autonomic nervous system tone.