Radiomics Approach to Prediction of Occult Mediastinal Lymph Node Metastasis of Lung Adenocarcinoma.

Objective: The purpose of this study was to evaluate the prognostic impact of radiomic features from CT scans in predicting occult mediastinal lymph node (LN) metastasis of lung adenocarcinoma.

Methods: A total of 492 patients with lung adenocarcinoma who underwent preoperative unenhanced chest CT were enrolled in the study. A total of 300 radiomics features quantifying tumor intensity, texture, and wavelet were extracted from the segmented entire-tumor volume of interest of the primary tumor. A radiomics signature was generated by use of the relief-based feature method and the support vector machine classification method. A ROC regression curve was drawn for the predictive performance of radiomics features. Multivariate logistic regression models based on clinicopathologic and radiomics features were compared for discriminating mediastinal LN metastasis.

Results: Clinical variables (sex, tumor diameter, tumor location) and predominant subtype were risk factors for pathologic mediastinal LN metastasis. The accuracy of radiomics signature for predicting mediastinal LN metastasis was 91.1% in ROC analysis (AUC, 0.972; sensitivity, 94.8%; specificity, 92%). Radiomics signature (Akaike information criterion [AIC] value, 80.9%) showed model fit superior to that of the clinicohistopathologic model (AIC value, 61.1%) for predicting mediastinal LN metastasis.

Conclusions: The radiomics signature of a primary tumor based on CT scans can be used for quantitative and noninvasive prediction of occult mediastinal LN metastasis of lung adenocarcinoma.