A Predictive Model for Perinatal Brain Injury Using Machine Learning Based on Early Birth Data.

Area of Science:

• Neonatal Medicine
• Machine Learning in Healthcare
• Medical Imaging Analysis

Background:

• Predicting perinatal brain injury is challenging, often relying on subjective clinical suspicion.
• Brain magnetic resonance imaging (MRI) is crucial but its timing and necessity can be difficult to determine.
• Developing reliable predictive methods using early data is essential for timely intervention.

Purpose of the Study:

• To develop and validate a machine learning model for predicting perinatal brain injury.
• To utilize readily available early birth data for predictive modeling.
• To improve the accuracy and efficiency of diagnosing neonatal brain injury.

Main Methods:

• A gradient boosting machine learning model was trained on early birth data from 179 neonates.
• Synthetic minority over-sampling techniques (SMOTE) and adaptive synthetic sampling (ADASYN) were employed to handle class imbalance.
• Model performance was assessed using accuracy, F1 score, and ROC curves, with feature importance and SHAP values analyzed.

Main Results:

• The gradient boosting model combined with ADASYN demonstrated superior performance in predicting perinatal brain injury.
• Key differentiating factors included mode of delivery, Apgar scores, capillary pH, lactate dehydrogenase (LDH) levels, and therapeutic hypothermia.
• The 1-minute Apgar score was identified as the most influential predictor, while LDH levels showed the highest SHAP values.

Conclusions:

• The developed machine learning model, particularly with ADASYN oversampling, offers an effective tool for predicting perinatal brain injury.
• This predictive capability can enhance early detection, potentially leading to better long-term outcomes for neonates.
• The model may help reduce the frequency of unnecessary MRI scans, thereby decreasing healthcare expenditures.