Multi-modal prediction model for thymic epithelial tumors: Enhancing surgical decisions and recurrence risk assessment from CT datasets.
This study presents a multi-task AI model that accurately predicts histology and staging in thymic epithelial tumors, which could aid in surgical decision-making.
Where it sits
this study against the rest of the thymalin (thymus extract) corpusSummary and findings
This study developed a multi-task deep learning model for predicting histology and Masaoka-Koga stage in thymic epithelial tumors (TETs) using 659 chest CT scans. The model achieved high performance with an AUC of 0.9674 for histology classification and 0.9328 for Masaoka-Koga staging. No therapeutic claims are made regarding the use of thymalin.
Abstract
<h4>Background</h4>Precise prediction of histology (TETs vs. Non-TETs), Masaoka-Koga staging, and WHO classification is essential for surgical and postoperative decisions. Artificial intelligence (AI) models have shown promise in addressing these gaps, but existing studies often suffer from small datasets and single-task limitations.<h4>Purpose</h4>This study aims to develop and validate a multi-task deep learning model for preoperative non-invasive prediction of histology and Masaoka-Koga stage in thymic epithelial tumors to support individualized risk assessment and treatment planning.<h4>Materials and methods</h4>We analyzed 659 chest CT scans from TET patients with pathologically confirmed histology, Masaoka-Koga staging, and WHO classification collected between October 2014 and July 2025 from Shanghai General Hospital, Shanghai Pulmonary Hospital, and Huashan Hospital. A deep learning model was developed to classify histology into TETs and Non-TETs groups, Masaoka-Koga stages and WHO risk groups.<h4>Results</h4>The model achieved high performance across tasks in the test set. For histology classification, AUC was 0.9674 (95% CI, 0.9347-0.9895), with accuracy of 91.73% (95% CI, 86.47%-96.24%). Masaoka-Koga staging stratification yielded an AUC of 0.9328 (95% CI, 0.8179-1.0000) and accuracy of 95.38% (95% CI, 89.23%-100.00%). WHO risk prediction reached an AUC of 0.8485 (95% CI, 0.7447-0.9347) and accuracy of 78.46% (95% CI, 67.69%-89.23%). Calibration analysis further showed low Brier scores for the histology and Masaoka-Koga models (0.0695 and 0.0456, respectively), supporting acceptable probability calibration for these two tasks.<h4>Conclusion</h4>This multi-task AI model provides a non-invasive, accurate tool for preoperative TET assessment, potentially optimizing surgical strategies and personalized care.
Background
This paper addresses the need for precise preoperative prediction of histology and staging in thymic epithelial tumors (TETs), which is critical for surgical and postoperative decisions. Previous studies have shown promise in using artificial intelligence (AI) for this purpose, but they often suffer from limitations such as small datasets and a focus on single tasks. This study aims to enhance prediction accuracy through a multi-task deep learning model, which could support individualized risk assessment and treatment planning.
Methods
The study analyzed 659 chest CT scans from TET patients with confirmed histology, Masaoka-Koga staging, and WHO classification, collected between October 2014 and July 2025 from three hospitals in Shanghai. A deep learning model was developed to classify histology into TETs and Non-TETs, as well as to stratify Masaoka-Koga stages and WHO risk groups. The primary outcomes included AUC and accuracy for each classification task.
Results
The model achieved an AUC of 0.9674 (95% CI, 0.9347-0.9895) for histology classification, with an accuracy of 91.73% (95% CI, 86.47%-96.24%). For Masaoka-Koga staging, the AUC was 0.9328 (95% CI, 0.8179-1.0000) with an accuracy of 95.38% (95% CI, 89.23%-100.00%). The WHO risk prediction yielded an AUC of 0.8485 (95% CI, 0.7447-0.9347) and an accuracy of 78.46% (95% CI, 67.69%-89.23%). Calibration analysis indicated low Brier scores of 0.0695 and 0.0456 for histology and Masaoka-Koga models, respectively.
Interpretation
The results indicate that the multi-task AI model demonstrates high accuracy in predicting histology and staging in TETs, which is consistent with prior literature suggesting AI's potential in medical imaging. However, while the AUC values are statistically significant, the clinical significance of these findings remains to be established in broader, multi-center studies. Limitations such as the single-center nature of the study and the potential for overfitting in AI models may confound the conclusions drawn.
Key findings
- AUC of 0.9674 (95% CI, 0.9347-0.9895) for histology classification.
- Accuracy of 91.73% (95% CI, 86.47%-96.24%) for histology classification.
- AUC of 0.9328 (95% CI, 0.8179-1.0000) for Masaoka-Koga staging.
- Accuracy of 95.38% (95% CI, 89.23%-100.00%) for Masaoka-Koga staging.
- AUC of 0.8485 (95% CI, 0.7447-0.9347) for WHO risk prediction.
- Accuracy of 78.46% (95% CI, 67.69%-89.23%) for WHO risk prediction.
Limitations
- Single-center study may limit generalizability.
- Potential for overfitting in AI models.
- No long-term follow-up reported.
- Not reported in abstract.