Machine learning models for predicting early hemorrhage progression in traumatic brain injury

Area of Science:

• Neuroscience
• Radiology
• Medical Informatics

Background:

• Traumatic brain injury (TBI) can lead to intracerebral hemorrhage (ICH), a condition requiring careful monitoring.
• Predicting the progression of ICH in mild to moderate TBI is essential for timely intervention and improved patient outcomes.

Purpose of the Study:

• To develop and validate a predictive model for ICH progression in TBI patients using initial CT scans.
• To identify clinical and radiological factors associated with ICH progression in mild to moderate TBI.

Main Methods:

• Retrospective analysis of 650 TBI patients (January 2010 - December 2021) with initial CT scans and clinical data.
• Utilized Random Forest and XGBoost machine learning algorithms, incorporating SHAP values for variable importance and risk assessment.
• Categorized ICH into intraparenchymal hemorrhage (IPH), petechial hemorrhage (PH), and subarachnoid hemorrhage (SAH).

Main Results:

• A 22.2% progression rate of ICH was observed in the study cohort.
• Petchial hemorrhage (PH) and countercoup injury were identified as significant predictors of ICH progression by Random Forest.
• XGBoost model achieved an AUC of 0.9, with a personalized risk assessment diagram achieving an AUC of 0.913.

Conclusions:

• Machine learning models, particularly XGBoost with SHAP analysis, demonstrate high precision in predicting ICH progression in TBI patients.
• Early identification of ICH progression risk through advanced modeling can significantly enhance TBI patient management.
• Petchial hemorrhage and countercoup injury are critical factors to consider in assessing ICH progression risk.