Physiological information in near-infrared spectroscopy

Near-infrared spectroscopy (NIRS) has emerged as an important tool for monitoring brain and tissue oxygen levels in neonatal intensive care units (NICUs). Despite the challenges posed by artifacts in NIRS data, these artifacts can be used beneficially. Systemic physiological artifacts, such as heartbeats, can be used to assess signal quality and extract heart rate (HR). Similarly, respiratory patterns in NIRS signals can be analyzed to extract respiratory rates (RR). This approach is particularly beneficial for neonates, as it can potentially reduce the need for multiple sensors by simultaneously analyzing HR, RR, and cerebral oxygen saturation while ensuring data quality. This thesis explores innovative methodologies for NIRS data analysis, focusing on enhancing data quality and extracting vital signs (HR and RR). The research initially concentrated on adults (Part I) due to the easier data accessibility. Insights from these studies informed the development of algorithms later adapted for neonatal use (Part II), demonstrating the adaptability and relevance of these algorithms in neonatal contexts. Part I: Physiological Information in Adult NIRS This section introduces two novel NIRS signal quality assessment algorithms: the Signal Quality Index (SQI) and the Machine Learning-based SQI (MLSQI). SQI evaluates NIRS signal quality on a multi-level scale, surpassing traditional binary assessments and correlating strongly with annotators' ratings. MLSQI builds on SQI, incorporating machine learning to analyze a broader range of signal features, enhancing real-time decision-making capabilities. Innovative approaches for estimating RR from NIRS data in adults are detailed, including methods for extracting RR without reference signals and robust techniques for RR estimation during physical activities. These advancements enhance NIRS's utility, making it a versatile tool for monitoring physiological responses in real-world scenarios. Part II: Physiological Information in Neonatal NIRS This section highlights significant advancements in neonatal monitoring within NICU settings through new algorithms for extracting HR and RR from NIRS signals. The NHR (NIRS HR) algorithm, designed for HR extraction from (pre)term neonates, shows notable improvement over existing algorithms. The NRR (NIRS RR) algorithm accurately extracts RR, addressing neonatal RR measurement complexities and demonstrating strong performance. This research is important in the field of neonatal care, as it could pave the way for more effective and less invasive monitoring techniques. The thesis showcases an application using a single NIRS device to assess sleep stages in neonates, integrating physiological and non-physiological data to analyze sleep patterns. This approach demonstrates the utility of a single sensor in detailed sleep analysis for neonates born preterm. The thesis highlights NIRS's potential in clinical settings, particularly in neonatal care. By developing algorithms for extracting vital physiological measurements (HR and RR) and integrating these with cerebral oxygenation data, NIRS technology could minimize the need for multiple sensors. In the future, this could potentially enhance the management of conditions such as impaired cerebral vascular autoregulation. Combining NIRS with technologies such as electroencephalography and transcranial Doppler might provide comprehensive, real-time monitoring essential for the early detection of cerebral dysregulation, potentially setting a new standard for non-invasive patient monitoring in neonatal care.