The significance of peritumoral 5mm regions features for radiomics model in distinguishing the lung adenocarcinomas and granulomas

To investigate whether features of 5-mm peritumoral regions could significantly improve the predictive efficacy of a radiomics model based on solid pulmonary tumors at distinguishing lung adenocarcinoma(LAC) from granuloma(GR).

We retrospectively evaluated 167 lung tumors pathologically proven to be LAC (96) or GR (71) and divided them into training group (116) and testing (51) group. We delineated each tumor with three different measures using the tumor and its 5-mm peritumoral region. Then, we extracted 465 features from each volume of interest(VOI) and chose the optimal features to build the diagnostic models. We built four different models using different methods. Finally, we compared the performance of the four models in the test set.

The area under the curve(AUC) of each model in the test group was 0.765 (95% confidence interval(CI): 0.620–0.909), 0.797 (95%CI: 0.670–0.924), and 0.784 (95%CI: 0.647–0.920), respectively. Results of the DeLong test showed that the differences between model 2, model 3, and model 1 were not significant. Results of net reclassification improvement(NRI) showed that model 2 and model 3 had better differential diagnostic efficacy than model 1, with accuracies(ACCs) of 0.784, 0.745, and 0.686, respectively, but the differences were not significant (P>0.05). Moreover, the nomogram had good diagnostic and predictive abilities, with an AUC of 0.848 (95%CI: 0.736–0.961) and an ACC of 0.804.

Features of 5-mm peritumoral regions improved the predictive ability of the radiomics model based on the solid pulmonary tumor, but the difference was not significant.