Development and validation of a preoperative pathological risk prediction model for thymic tumors: A retrospective multicenter diagnostic study.

Abstract

8122 Background: The preoperative diagnosis of thymic tumors currently relies on preoperative CT data and the experience of the surgeon, and there is still a lack of highly effective noninvasive radiomic prediction model to assist in clinical decision making. To develop and validate a preoperative CT-based radiomic model to accurately predict the pathological risk staging of thymic tumors. Methods: This is a retrospective diagnostic study that included patients from two independent centers Charité Universitätsmedizin Berlin (N = 74) as well as the First Hospital of Zhengzhou University (N = 143) between September 2003 and July 2024. WHO pathological type A, AB, and B1 of the postoperative thymic tumor were divided into low-risk groups (N = 110), and type B2, B3, and C (N = 107) were classified as high-risk groups. The patients were randomly divided into a training group (N = 130) and a validation group (N = 87) in a ratio of 6:4. Preoperative CT imaging of thymic tumor was used to define the tumor area and the peritumoral area (5 mm) and to extract the radiomic features using 3Dslicer software and the in-house pyradiomic software package. Feature selection and model development were conducted using the least absolute shrinkage and selection operator（LASSO）algorithm and logistic regression analysis. Results: A total of 217 patients undergoing thymectomy were included in this study. After univariate analysis, patients' age, smoking history, and the presence of combined myasthenia gravis (MG) were included in the clinical model, and the clinical model achieved AUC = 0.75 and 0.65 in the training and validation groups, respectively. The combined radiomic model has better performance than the two radiomic models alone, with AUC values of 0.86, 0.73, respectively. The nomogram model conducted by combined radiomic model with the clinical model achieved the best results in training (AUC = 0.89) and validation cohort (AUC = 0.84). Calibration curve and decision curve analysis (DCA) illustrated the clinical usability and reliability of the model. Conclusions: By combining the clinical model with a combined radiomic model, the nomogram model can effectively differentiate pathological risk staging of thymic tumors, providing surgeons with a potential preoperative decision support.