Abstract:
Objective This study aimed to evaluate the value of CT-based radiomics machine learning models in predicting spread through air spaces (STAS) in lung adenocarcinoma (LUAD) and to determine the optimal peritumoral analysis region.
Methods Data from 378 patients who underwent non-small cell lung cancer surgery at Zhejiang Cancer Hospital between January 2013 to January 2017 were retrospectively analyzed. Logistic regression, random forest, and XGBoost models were constructed using regions extending 0, 3, 6, 9, and 12 mm outward from the tumor margin.
Results The XGBoost model using the 6 mm peritumoral region performed best on the test set, with an AUC-ROC of 0.855 (95% CI: 0.756–0.950), followed by the XGBoost model using the 9 mm region. Decision curve analysis (DCA) indicated that the XGBoost models for the 6 mm and 9 mm regions had higher net clinical benefits. Feature analysis revealed that some wavelet transform features significantly contributed to STAS prediction.
Conclusions This preliminary study suggests that CT-based radiomics machine learning models have predictive value for STAS. The XGBoost model based on the 6 mm peritumoral region demonstrated the best performance, and holds promise in assisting preoperative assessment.
下载: