Abstract
Clinical feasibility nomograms were developed to facilitate the differentiation between thymic epithelial tumors (TETs) and mediastinal lymphomas (MLs), aiming to minimize the occurrence of non-therapeutic thymectomy. A total of 255 patients diagnosed with TETs or MLs underwent pre-treatment 18F-FDG PET/CT. Comprehensive clinical and imaging data were collected, including age, gender, lactate dehydrogenase (LDH) level, pathological results, presence of myasthenia gravis symptoms, B symptoms, mass size, location, morphology, margins, density, and metabolic parameters derived from PET/CT. Radiomic features were extracted from the region of interest (ROI) of the primary lesion. Feature selection techniques were employed to identify the most discriminative subset of features. Machine learning methods were utilized to build candidate models, which were subsequently evaluated based on their area under the curve (AUC). Finally, nomograms were constructed using the optimal model to provide a clinical tool for improved diagnostic accuracy. The performance of the radiomic models was evaluated by their calibration, discrimination, and clinical utility. Several independent risk factors were identified for distinguishing TETs from MLs, including average standardized uptake value (SUVavg), LDH, age, mass size, and radiomic score (rad-score). Significance was observed in differentiating the two types of tumors based on these factors. The best clinical efficacy was demonstrated by the combined model, with an impressive AUC of 0.954. Decision curve analysis and calibration curves indicated that the combined model was clinically advantageous for discriminating TETs from MLs. Besides, the results of external validation showed a sensitivity of 0.8 and a specificity of 0.78. Preoperatively, the differentiation of TETs from MLs can be facilitated by the utilization of the combined clinical information and radiomics model. This approach holds promise in minimizing the occurrence of unnecessary anterior mediastinal surgeries.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have