Multimodal CT Model: Enhancing Lung Cancer STAS Prediction vs. Unimodal CT

Abstract Background Lung cancer is one of the leading causes of cancer-related death worldwide. Tumor spread through air spaces (STAS), a unique invasive pattern of lung cancer, has a direct effect on patient prognosis. Currently, accurately identifying the presence of STAS in patients before surgery remains an unresolved challenge in clinical practice. This study aimed to construct a model for predicting the STAS status of lung cancer patients on the basis of radiomic features, clinical information, and genetic features. Methods A total of 333 lung cancer patients who underwent surgical treatment were enrolled in this study, and they were randomly divided into a training set and a test set at a ratio of 7:3. All the subjects received preoperative computed tomography (CT) scans, with clinical information and genetic data collected simultaneously. After delineating the volume of interest (VOI) of lung tumors, the research team extracted intratumoral radiomic features, 3-mm peritumoral radiomic features, habitat features, and deep learning features from the VOI via open-source Python packages. Multiple radiomic models were constructed via Python by integrating clinical features and genetic features, and the performance of each model was subsequently evaluated via indicators such as the area under the receiver operating characteristic curve (AUC), decision curve analysis (DCA), and calibration curves. Results Fusion Model 1, which was constructed by integrating intratumoral radiomic features and incorporating 3-mm peritumoral radiomic features, habitat features, and deep learning features, demonstrated excellent predictive performance: its AUC value reached 0.96 in the training set and 0.933 in the validation set. Moreover, the model showed favorable net benefit in decision curve analysis (DCA) and extremely strong stability in the calibration curve, indicating high consistency between the predicted and actual probabilities. In addition, the results of this study revealed no significant associations between genetic features and STAS status; among the clinical features, a tumor diameter > 2 cm was strongly correlated with STAS. Conclusions The multimodal model "Fusion Model 1" has demonstrated excellent predictive performance and stable performance, providing important support and new hope for promoting preoperative personalized medical practice and improving the prognosis of lung cancer patients.