Based on multiscale computed tomography (CT) radiomics, a better model was established to differentially diagnose benign lesions and lung adenocarcinoma of Lung-RADS 2022 category 4B solid lung nodules.

The retrospective study included 178 patients with solid pulmonary nodules were assigned to the training dataset (n = 124) and the testing dataset (n = 54). Gradient boosting decision tree (GBDT) was used to reduce the dimensionality of data and select the best radiomics features. Four models were developed by logistic regression method, namely the clinical and imaging model (CIM), the plain CT radiomics model (PRM), the enhanced CT radiomics model (ERM), and the combined model (CM). Area under the curve (AUC) evaluates the model performance. Net reclassification improvement (NRI) and the integrated discrimination index (IDI) were calculated to compare the performance of different models to determine the best model.

In the training dataset, the AUC of CIM, PRM, ERM, and CM were 0.795, 0.791, 0.828, and 0.888. The continuous NRI and IDI of CM was better than that of CIM, PRM, ERM (P < 0.001), CM is optimal. In the testing dataset, the AUC of CIM, PRM, ERM and CM were 0.810, 0.689, 0.864 and 0.881. The continuous NRI of CM was better than that of CIM, PRM, ERM (P < 0.050). The IDI of CM was better than that of CIM and PRM (P < 0.050). The AUC values of the Mayo Clinic (Mayo) model, Veterans Administration (VA) model, Peking University People’s Hospital (PKUPH) model and United Imaging Artificial Intelligence (UI AI) model were 0.419, 0.410, 0.676 and 0.675, CM is still optimal.

Radiomics can be used as a non-invasive tool to distinguish between benign lesions and lung adenocarcinoma of Lung-RADS 2022 category 4B solid lung nodules, and CM is the best predictive model.