Machine Learning Models for Predicting Abnormal Brain CT Scan Findings in Mild Traumatic Brain Injury Patients.

Area of Science:

• Neurology
• Medical Imaging
• Artificial Intelligence

Background:

• Traumatic Brain Injury (TBI) is a major global cause of death and disability.
• Predicting CT scan abnormalities in mild TBI is crucial for timely intervention.

Purpose of the Study:

• To develop and optimize machine learning (ML) algorithms for predicting abnormal brain CT scans in mild TBI patients.
• To identify key clinical predictors of abnormal CT findings.

Main Methods:

• Retrospective analysis of 424 mild TBI patients.
• Feature selection using univariate analysis and SMOTE for class imbalance.
• Evaluation of ML models (XGBoost, Random Forest, Naive Bayes) using accuracy, F1-score, and AUC-ROC.
• SHAP analysis for interpreting feature importance.

Main Results:

• XGBoost achieved the highest performance (AUC 0.9611, accuracy 0.8937).
• Abnormal CT findings were associated with older age, male sex, lower GCS, fractures, and hematomas.
• Key predictors identified by SHAP analysis included lower GCS scores, decreased SpO2, and tachypnea.

Conclusions:

• XGBoost and Random Forest models demonstrate high accuracy in predicting abnormal brain CT scans.
• Clinical factors like GCS, SpO2, and respiratory rate are significant predictors.
• These ML models may help optimize CT scan usage and resource allocation, pending further validation.