Preoperative prediction of thymoma risk classification with machine learning-based computed tomography radiomics
Kai Zhao1, Yiming Liu2,3, Honghao Xu4,3
1Department of Thoracic Surgery, Hainan Hospital of Chinese PLA General Hospital, Sanya, China.
World Journal of Surgical Oncology
|June 19, 2026
Summary
This study developed a radiomics nomogram using non-contrast CT scans to differentiate thymoma risk. The combined model accurately predicts thymoma subtypes, aiding personalized treatment strategies.
Area of Science:
- Radiology
- Oncology
- Medical Imaging Analysis
Background:
- Thymomas are typically assessed with chest computed tomography (CT), but limitations exist in distinguishing histological subtypes.
- Overlap in semantic features between localized thymomas and other subtypes necessitates improved diagnostic methods.
Purpose of the Study:
- To develop a combined radiomics model using non-contrast CT.
- To investigate the clinical utility of this model in preoperative thymoma risk classification.
Main Methods:
- Retrospective analysis of 436 thymoma patients (2010-2023).
- Data divided into training (n=306) and validation (n=130) sets.
- Radiomic features extracted from non-contrast CT, selected using PCA, correlation, and LASSO algorithms.
- A radiomics nomogram integrated clinical factors and radiomics scores.
Main Results:
- A nomogram incorporating two clinicoradiological and 11 radiomics features was constructed.
- The nomogram demonstrated superior diagnostic performance for thymoma risk stratification compared to single models.
- Area under the curve (AUC) was 0.753 in the training cohort and 0.735 in the validation cohort.
Conclusions:
- The nomogram accurately differentiates thymoma histological subtypes by integrating clinical and radiomics data.
- This tool can assist in formulating personalized treatment plans.
- The model shows potential for clinical promotion and application.
