Automated Neuroprognostication Via Machine Learning in Neonates with Hypoxic-Ischemic Encephalopathy

Area of Science:

• Neuroimaging and Machine Learning
• Neonatal Neurology
• Developmental Pediatrics

Background:

• Neonatal hypoxic-ischemic encephalopathy (HIE) poses significant risks for mortality and long-term neurodevelopmental impairments.
• Current neuroimaging (MRI) methods for prognostication in HIE are subjective and lack precision.
• Objective and automated analysis of newborn brain MRI is needed to enhance outcome prediction.

Purpose of the Study:

• To develop an automated MRI analysis approach for objective prognostication in neonates with HIE.
• To improve the accuracy of predicting neurodevelopmental outcomes compared to existing methods.
• To integrate quantitative MRI features with clinical data for enhanced predictive modeling.

Main Methods:

• Anatomic MRI template creation from 286 infants undergoing therapeutic hypothermia.
• Extraction of quantitative shape and radiomic measures from deep gray-matter structures.
• Training an elastic net model using MRI measures, demographic/laboratory data, or both, to predict Bayley Scales of Infant and Toddler Development scores at 18 months.

Main Results:

• MRI-based measures significantly predicted Bayley scores for cognitive, language, and motor outcomes.
• MRI-based predictions demonstrated higher correlations, explained more variance, and had smaller errors than clinical data alone.
• Combined MRI and clinical data provided similar or marginally improved predictive performance.

Conclusions:

• Machine learning models incorporating MRI-based features can accurately predict neurodevelopmental outcomes in neonates with HIE.
• These models are effective across all neurodevelopmental domains and the full spectrum of outcomes.
• This automated approach offers a more objective and reliable tool for HIE prognostication.