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

Introduction: Various tools have been developed to determine the priority of radiography in trauma patients. This study aimed to investigate the role of machine learning models in predicting chest injuries following multiple trauma.

Methods: We used the database of a comprehensive cross-sectional survey conducted in 2015. Eight machine learning models were developed using demographic characteristics, physical exam findings, and radiologic results of 2860 patients.

Results: Area under the receiver operating characteristic curve (AUC) was greater than 0.96 in Random Forest, Gradient Boosting, XGBoost, Decision Tree, Support Vector Machine (SVM), Logistic Regression, K-Nearest Neighbors (KNN), and Neural Network models. The random forest model, XGBoost and Gradient Boosting had the highest accuracy (0.99). Sensitivity was also highest in the Gradient Boosting, XGBoost and KNN models (0.99). The specificity of all of the models in predicting chest radiography outcomes of multiple trauma patients was higher than 0.97, except for logistic regression and SVM (0.912 and 0.885 respectively).

Conclusions: Our study highlights the strong potential of machine learning models, especially Random Forest and Gradient Boosting, in predicting chest trauma outcomes with high accuracy and sensitivity.