Functional Near-Infrared Spectroscopy as a Bedside Neuroimaging Tool for Neonatal Brain Monitoring and Injury Characterization
Abstract
Early brain injury in neonates is a common neurological complication associated with high mortality rates and long-term morbidities. Two common types of injury are intraventricular hemorrhage (IVH) and hypoxic ischemic encephalopathy (HIE). Currently used clinical neuroimaging tools for diagnosing or monitoring the injury have limitations, e.g., cranial ultrasound is not for continuous monitoring of the evolution of injury so it could miss the optimal timing for intervention, and magnetic resonance imaging (MRI) is not very accessible for infants in neonatal intensive care unit. Functional near-infrared spectroscopy (fNIRS) can be an alternative imaging technique as it can be used at the bedside for continuous monitoring. Previous research using functional MRI (fMRI) indicated that neonates with brain injury were reported to have altered resting-state functional connectivity (RSFC), which can be a potential biomarker for diagnosis. The current thesis examined and expanded upon the potential of fNIRS as a neuromonitoring tool for newborns with brain injury. Specifically, the goal was to determine if fNIRS can yield RSFC indices comparable to fMRI and whether differences in RSFC can be detected between neonates with and without brain injury. Also, predictive models based on machine learning were developed to address the cortical penetration depth of fNIRS, which is a major drawback limiting wider applications of the technology.
Neonates diagnosed with IVH or HIE, and healthy newborns were recruited and scanned with fNIRS and fMRI. RSFC was calculated from both modalities and compared. Then RSFC patterns were compared between neonates with and without injury. fNIRS and fMRI yielded comparable RSFC indices, and fNIRS also identified altered RSFC patterns in both IVH and HIE groups compared to healthy newborns. Then, graph convolutional networks (a subtype of artificial neural networks) were applied to predict subcortical connectivity from cortical connectivity, using either neonatal or adult fNIRS data, and gave good performance. In general, this thesis indicates that fNIRS has the potential to be a new tool for assessing brain injury and monitoring cerebral hemodynamics in neonates.