Background: Magnetic resonance imaging (MRI) is the gold standard for outcome prediction after hypoxic-ischemic encephalopathy (HIE). Published scoring systems contain duplicative or conflicting elements.
Methods: Infants ≥36 weeks gestational age (GA) with moderate to severe HIE, therapeutic hypothermia treatment, and T1/T2/diffusion-weighted imaging were identified. Adverse motor outcome was defined as Bayley-III motor score <85 or Alberta Infant Motor Scale <10th centile at 12 to 24 months. MRIs were scored using a published scoring system. Logistic regression (LR) and gradient-boosted deep learning (DL) models quantified the importance of clinical and imaging features. The cohort underwent 80/20 train/test split with fivefold cross validation. Feature selection eliminated low-value features.
Results: A total of 117 infants were identified with mean GA = 38.6 weeks, median cord pH = 7.01, and median 10-minute Apgar = 5. Adverse motor outcome was noted in 23 of 117 (20%). Putamen/globus pallidus injury on T1, GA, and cord pH were the most informative features. Feature selection improved model accuracy from 79% (48-feature MRI model) to 85% (three-feature model). The three-feature DL model had superior performance to the best LR model (area under the receiver-operator curve 0.69 versus 0.75).
Conclusions: The parsimonious DL model predicted adverse HIE motor outcomes with 85% accuracy using only three features (putamen/globus pallidus injury on T1, GA, and cord pH) and outperformed LR.
Keywords: HIE; MRI; Machine learning; Neonatal; Outcome.
Copyright © 2023 Elsevier Inc. All rights reserved.