Predicting the Presence of Traumatic Chest Injuries Using Machine Learning Algorithm.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Trauma Care

Background:

• Determining radiography priority in trauma patients is crucial for timely diagnosis.
• Existing tools for radiography prioritization require validation and enhancement.

Purpose of the Study:

• To evaluate the efficacy of machine learning (ML) models in predicting chest injuries in multiple trauma patients.
• To compare the performance of various ML algorithms for chest trauma diagnosis.

Main Methods:

• Utilized a 2015 cross-sectional survey database of 2860 multiple trauma patients.
• Developed and assessed eight ML models (Random Forest, Gradient Boosting, XGBoost, Decision Tree, SVM, Logistic Regression, KNN, Neural Network) using demographic, physical exam, and radiologic data.

Main Results:

• All eight ML models achieved an Area Under the Receiver Operating Characteristic Curve (AUC) greater than 0.96.
• Random Forest, XGBoost, and Gradient Boosting models demonstrated the highest accuracy (0.99).
• Gradient Boosting, XGBoost, and KNN models exhibited the highest sensitivity (0.99), with specificity exceeding 0.97 for most models.

Conclusions:

• Machine learning models, especially Random Forest and Gradient Boosting, possess significant potential for accurately predicting chest trauma outcomes.
• These ML tools can enhance the diagnostic process and improve patient management in trauma settings.