WIP 75 - Broadband Optical Spectroscopy for Noninvasive Early Identification of HIE

Publication Number: WIP 75.7475

Imran M. Ilahi, Ann & Robert H. Lurie Children's Hospital of Chicago, Chicago, IL, United States; Seth Goldstein, Ann & Robert H. Lurie Children's Hospital of Chicago, Chicago, IL, United States; Andrea C.. Pardo, Ann & Robert H. Lurie Children's Hospital of Chicago, Chicago, IL, United States; Janine Y. Khan, Ann & Robert H. Lurie Children's Hospital of Chicago, Chicago, IL, United States

Background: Hypoxic ischemic encephalopathy (HIE) is a leading cause of neonatal mortality and disability, and early diagnosis is crucial for improving outcomes. Therapeutic hypothermia (TH) within 6 hours offers neuroprotection for neonates with moderate to severe HIE, but current diagnostics rely on biochemical and clinical criteria. Broadband optical spectroscopy (BOS) is a non-invasive method that uses a handheld probe to measure a wide range of wavelengths with 500 times the granularity of standard oximeters. In our pilot study of 12 neonates (6 HIE, 6 controls), BOS showed promising spectral differences between HIE and controls, particularly in near-infrared and short-wave infrared regions. This study evaluates BOS as a tool for early HIE diagnosis to identify candidates for timely TH in a larger cohort.

Objective: To identify signal differences between neonates with HIE and healthy controls, apply machine learning to distinguish ischemia, and develop algorithms with >90% sensitivity and specificity for early HIE identification.

Design/Methods: The study received IRB approval from Northwestern Medicine Prentice Women's Hospital and Ann & Robert H. Lurie Children's Hospital. Neonates ≥36 weeks GA with moderate to severe HIE were recruited alongside healthy controls. Exclusion criteria included GA < 36 weeks, birth weight < 1800g, and subgaleal hemorrhage. Each neonate underwent 3-5 BOS data collections from the transfontanellar region and thigh within 24 hours of life. Machine learning will identify key wavelengths and classify samples as ischemic or non-ischemic. Principal Component Analysis (PCA) will reduce data complexity, followed by Linear Discriminant Analysis (LDA) for classification. A neural network will refine accuracy, with 75% of the data for training and 25% for validation. Data collection started in August 2023 and will conclude in December 2024, with analysis planned for January 2025. Final results will be available by March 2025.