Predictive Features for Anterior Mediastinal Mass Diagnoses.

Objective: We set out to evaluate a set of demographic and computed tomography imaging features for diagnosing anterior mediastinal masses.

Methods: We identified 223 patients with anterior mediastinal masses, which we divided into training and validation sets. One radiologist evaluated computed tomography imaging features on the training set. Then, predictive features were identified, and 3 radiologists evaluated these on the validation set. A naive Bayesian classifier based on the features was compared with the radiologists' first-choice diagnosis.

Results: Internal mammary lymphadenopathy and mediastinal encasement were strongly associated with lymphomas. Low attenuation and midline location were strongly associated with benign lesions, and older age was associated with thymic epithelial neoplasms. The average accuracy of the 3 radiologists' diagnoses was 78%, compared with 71% for the classifier.

Conclusions: Nine demographic and imaging features were found to be helpful in diagnosing anterior mediastinal masses. By using these features, radiologists can suggest the diagnosis with fair accuracy.