Machine learning models from computed tomography to diagnose thymic epithelial tumors requiring combined resection

Area of Science:

• Thoracic surgery
• Medical imaging
• Artificial intelligence in medicine

Background:

• Minimally invasive surgery is standard for thymic epithelial tumors (TETs).
• Complex TET cases may necessitate resection of adjacent structures.
• Predicting the need for combined resection preoperatively is crucial for surgical planning.

Purpose of the Study:

• To develop and validate machine learning models for predicting combined resection in TETs.
• To utilize preoperative contrast-enhanced computed tomography (CT) for prediction.
• To assess the clinical utility of these predictive models.

Main Methods:

• Radiomics features extracted from contrast-enhanced CT scans of 212 patients with TETs.
• Models trained and validated using nested cross-validation with feature selection and machine learning classifiers.
• Clinical utility evaluated using decision curve analysis (DCA).

Main Results:

• The eXtreme Gradient Boosting (XGB) classifier achieved an AUC of 0.797 (training) and 0.817 (validation).
• High predictive performance was observed for Random Forest (RF) and Gradient Boosting (GB) classifiers.
• DCA confirmed model validity across a threshold range of 15-72%.

Conclusions:

• Machine learning models using preoperative CT can accurately predict TETs requiring combined resection.
• These models demonstrate significant potential to assist in surgical approach selection.
• The validated models offer a valuable tool for improving preoperative planning in complex TET surgeries.