Related Experiment Video
Updated: Jun 24, 2026
Amplitude-Integrated EEG in Infants at Risk of Hypoxic-Ischemic Encephalopathy: A Feasibility Study in Road and Air Transport in Western Australia
Published on: June 21, 2024
XGBoost prediction of adverse neurodevelopmental outcomes in hypoxic-ischemic encephalopathy neonates
Tae-Young Kim1,2, Heung-Min Park2, Young-Ah Youn3
1Department of Science, Sejong Science High School, Seoul, Republic of Korea.
Insights
An AI model accurately predicts neurodevelopmental impairment in infants with hypoxic-ischemic encephalopathy (HIE) using brain MRI and clinical data. This aids early intervention for better outcomes in treated newborns.
Area of Science:
- Neonatal neurology
- Artificial intelligence in medicine
- Neuroimaging
Background:
- Early prediction of neurodevelopmental (ND) impairment is crucial for timely intervention in neonates with hypoxic-ischemic encephalopathy (HIE).
- Therapeutic hypothermia (TH) is a standard treatment for HIE, but predicting long-term outcomes remains challenging.
- Accurate risk stratification is needed to guide individualized care for HIE infants.
Purpose of the Study:
- To develop and validate an interpretable AI model for predicting ND impairment in term HIE infants treated with TH.
- To utilize neonatal brain MRI volumetrics and clinical data for early risk stratification.
- To identify key predictors of adverse neurodevelopmental outcomes.
Main Methods:
- Developed an Extreme Gradient Boosting (XGBoost) prediction model using neonatal brain MRI volumetrics and clinical features.
- Included 89 full-term HIE infants treated with TH; ND impairment defined as Bayley-III score <85 at 18-24 months.
- Evaluated model performance using 5-fold cross-validation and assessed interpretability with SHapley Additive exPlanations (SHAP).
Main Results:
- The AI model demonstrated strong predictive performance (AUC=0.952, sensitivity=0.800, specificity=0.914).
- Key predictors included MRI-based severity, regional brain volume reductions, low 5-minute Apgar scores, and high Gross Motor Function Classification System levels.
- SHAP analysis provided patient-specific risk profiles, enhancing model interpretability.
Conclusions:
- An interpretable AI model can accurately stratify risk for adverse ND outcomes in term HIE infants treated with TH.
- The model integrates neuroimaging and clinical data for individualized prediction.
- This tool facilitates timely, targeted interventions to improve neurodevelopmental trajectories in high-risk neonates.
Abstract:
Early prediction of neurodevelopmental (ND) impairment in neonates with hypoxic-ischemic encephalopathy (HIE) is essential for timely intervention, particularly in infants treated with therapeutic hypothermia (TH). We developed an Extreme Gradient Boosting-based prediction model using neonatal brain magnetic resonance imaging volumetrics and clinical features in 89 full-term HIE infants treated with TH. ND impairment was defined as a Bayley-III composite score <85 at 18 to 24 months. Model performance was evaluated using stratified 5-fold cross-validation, and interpretability was assessed using SHapley Additive exPlanations. The model achieved strong performance (mean area under the receiver operating characteristic curve of 0.952, sensitivity of 0.800, and specificity of 0.914). Key predictors included magnetic resonance imaging severity, regional brain volume reductions, lower 5-minute Apgar scores, and higher Gross Motor Function Classification System levels. SHapley Additive exPlanations demonstrated patient-specific risk profiles. This interpretable AI model enables accurate, individualized early risk stratification of adverse ND outcomes in term HIE infants treated with TH.
