RT Journal Article SR Electronic T1 Preoperative prediction of meidastinal node metastasis using Radiomics model based on 18F-FDG PET/CT of the primary tumor in non-small cell lung cancer patients JF Journal of Nuclear Medicine JO J Nucl Med FD Society of Nuclear Medicine SP 1034 OP 1034 VO 62 IS supplement 1 A1 Zheng, Kai A1 Wang, Xinrong A1 Tang, Yongxiang A1 Fang, Zhihui A1 Hou, Jiale A1 Zhu, Zehua A1 Hu, Shuo YR 2021 UL http://jnm.snmjournals.org/content/62/supplement_1/1034.abstract AB 1034Purpose: For patients newly diagnosed with non-small cell lung cancer (NSCLC), the exact imaging evaluation of pathological lymph node (LN) status plays an important role in the therapy regimen's choice. So far, the existing imaging modalities and researches cannot solve the problems caused by false-positive LNs and occult LN metastasis. We assume that a prediction model constructed by radiomic features of the primary tumor has the potential to make up for such deficiencies. The present study aimed to investigate whether a fluorine-18-fluorodeoxy glucose positron emission tomography / computed tomography (18F-FDG PET/CT)-based radiomics model could predict pathological mediastinal LN staging (pN staging) in NSCLC patients undergoing surgery. Materials and Methods: The prediction model was developed in a training cohort that consisted of 501 patients with clinicopathologically confirmed NSCLC. Radiomic features were extracted from 18F-FDG PET/CT of primary tumor. Support vector machine and extremely random trees were used for radiomics model building. Internal validation was assessed. An independent testing cohort contained other 215 NSCLC patients. The performance of radiomics model and clinical node staging (cN staging) in predicting pN staging (pN0 vs. pN1&2) were compared for each cohort and the 634 patients without both obvious enlarged and FDG-avid mediastinal lymph node (cN± group). Results: The radiomics model, which consisted of 25 selected features, was significantly associated with LN metastasis (P < 0.005 for both training and testing cohorts). According to receiver operator characteristic (ROC) curves, the sensibility, specificity and the areas under curves (AUC) of radiomics model and cN staging were 0.704, 0.794 and 0.81 in the training cohort, 0.688, 0.704, and 0.766 in the testing cohort, 0.78, 0.611 and 0.78 in the cN± group, respectively. Conclusions: The radiomics model based on 18F-FDG PET/CT was potential in mediastinal LN staging in patients with NSCLC.