AUTHOR=Feng Wenfeng , Lin Runlong , Zhao Wenzhe , Cai Haifeng , Li Jingwu , Liu Yongliang , Cao Lixiu 

TITLE=Identifying small thymomas from other asymptomatic anterior mediastinal nodules based on CT images using logistic regression

JOURNAL=Frontiers in Oncology

VOLUME=Volume 15 - 2025

YEAR=2025

URL=https://www.frontiersin.org/journals/oncology/articles/10.3389/fonc.2025.1590710

DOI=10.3389/fonc.2025.1590710

ISSN=2234-943X

ABSTRACT=PurposeTo develop and validate a logistic regression (LR) model to improve the diagnostic performance of chest CT in distinguishing small (≤3 cm in long diameter on CT) thymomas from other asymptomatic small anterior mediastinal nodules (SAMNs).Materials and methodsA total of 231 patients (94 thymomas and 137 other SAMNs) with surgically resected asymptomatic SAMNs underwenting plain CT and biphasic enhanced CT from January 2013 to December 2023 were included and randomly allocated into training and internal testing sets at a 7:3 ratio. Clinical and CT features were analyzed, and a predictive model was developed based on independent risk features for small thymomas using multivariate LR in the training set. Receiver operating characteristic (ROC) curves and decision curve analysis (DCA) were used to compare the performance of the model and individual risk factors in the internal testing set. An additional prospective testing set (10 thymomas and 13 other SAMNs) was collected from the same institution between 2023 and 2024. The model’s performance was evaluated by area under the curve (AUC) and compared with the results of three radiologists using the DeLong test.ResultsThe LR model incorporating four CT independent risk features (lesion location, attenuation pattern, CT values in the venous phase [CTV], and enhancement degree) achieved an AUC of 0.887 for small thymomas prediction. This performance was superior to CTV alone (AUC = 0.849, P = 0.118) and significantly higher than other individual risk factors in the internal testing set (P < 0.05). DCA confirmed the model’s enhanced clinical utility across most threshold probabilities. In the prospective test set, the LR showed an AUC of 0.908 (95% CI: 0.765-1.00), comparable to the senior radiologist’s performance (AUC = 0.912 [95% CI: 0.765-1.00], P = 0.961), higher than the intermediate radiologist’s performance (AUC = 0.762 [95% CI: 0.554-0.969], P = 0.094), and significantly better than the junior radiologist’s performance (AUC = 0.700 [95% CI: 0.463-0.937], P = 0.044).ConclusionsThe CT-based LR model demonstrated well diagnostic performance comparable to that of senior radiologists in differentiating small thymomas from other asymptomatic SAMNs. CTV played a leading role in the model.