Efficacy of radiomics model based on the concept of gross tumor volume and clinical target volume in predicting occult lymph node metastasis in non-small cell lung cancer

Objective: To extract the gross tumor volume (GTV) and clinical target volume (CTV) of lung cancer from computed tomography (CT) images by radiomics method. To establish a predictive model for occult Lymph node metastasis (LNM) in patients with stage Ⅰ-Ⅱ A non-small cell lung cancer (NSCLC) based on contrast-enhanced CT. Methods: : A retrospective analysis of 598 patients in different hospitals with preoperative enhanced CT presentation of stage I-IIA NSCLC was performed. Of all patients, 119 patients (20%) had postoperative pathology suggestive of LNM. On chest-enhanced CT (CECT) arterial phase pictures, GTV and CTV radiomics characteristics were retrieved. The training group included 401 patients (67 with LNM and 334 without LNM), and the external validation group included 197 patients (52 with LNM and 145 without LNM). For choosing radiomics features, the LASSO regression algorithm was applied. Three models (GTV model, CTV model, and GTV+CTV hybrid model) were developed to predict occult LNM in early NSCLC patients using the screened features. The models were implemented by plotting nomogram plots. Results: : Eight optimal radiomics features were finally locked and significantly correlated with occult LNM (all P<0.05). The ROC curves of the three models showed good predictive effects. The AUC values of the GTV and CTV models in the validation group were 0.821 and 0.812. The AUC of the GTV+CTV hybrid model was significantly better than that of the CTV model (the AUC of the training group and the validation group in the GTV+CTV hybrid model was 0.869 and 0.906, respectively). Moreover, the GTV model (the AUC of the training group and the validation group in the CTV model were 0.834 and 0.812, and the AUC of the training group and the validation group in the GTV model were 0. 845 and 0.821), and statistically significant by Delong test. ROC curve and decision curve showed that the radiomics model had clinical application value. Conclusions: : The radiomics prediction model based on GTV and CTV was developed and verified in this study to be able to predict occult LNM in patients with preoperative clinical stage I-IIA NSCLC, and it was found that the combined model is significantly more accurate than either model operating independently.