Purpose:Occult lymph node involvement is a major issue in the management of non-small cell lung carcinoma (NSCLC), with an estimated prevalence of approximately 2.9–21.6% in 18F-FDG PET/CT series. The aim of the study is to construct a PET model to improve lymph node assessment.Methods:Patients with a non-metastatic cT1 NSCLC were retrospectively included from two centers, one used to constitute the training set, the other for the validation set. The best multivariate model based on Akaike’s information criterion was selected, considering age, sex, visual assessment of lymph node (cN0 status), lymph node SUVmax, primary tumor location, tumor size, and tumoral SUVmax (T_SUVmax). A threshold minimizing false pN0 prediction was chosen. This model was then applied to the validation set.Results:In total, 162 patients were included (training set: 44, validation set: 118). A model combining cN0 status and T_SUVmax was selected (AUC 0.907, specificity at threshold: 88.2%). In the validation cohort, this model resulted in an AUC of 0.832 and a specificity of 92.3% versus 65.4% for visual interpretation alone (p = 0.02). A total of two false N0 predictions were noted (1 pN1 and 1 pN2).Conclusion:Primary tumor SUVmax improves N status prediction and could allow a better selection of patients who are candidates for minimally invasive approaches.