A nomogram based on radiomics for predicting the high-grade histologic patterns in pure-solid clinical stage IA lung adenocarcinoma

Abstract

Abstract Background: High-grade histologic subtypes of lung adenocarcinoma (LUAD) are associated with poor prognosis, and preoperative identification of it may influence the decision of treatment strategy. Methods: In this retrospective study, data of 352 patients who underwent surgery for clinical stage IA LUAD from December 2019 to February 2022 were collected, in which 297 patients were from center 1 and 55 patients were from center 2. According to the percentage of solid(SOL) and micropapilary (MIP) histologic subtypes composition, all patients were divided into 2 groups (high-grade and low to median grade). Radiomics features were extracted from preoperative CT images by Python, Least absolute shrinkage and selection operator (LASSO) were used for radiomics factors selection and rad-score calculation. A final classification model was developed by multivariate logistic regression analysis. Results: A rad-score consisted of 6 features selected from 1130 radiomics showed considerable predictive performance in the internal training set (Area under the curve, AUC=0.76, 95% confidence interval [CI]:0.69~0.82). In contrast, the AUC of the model consisting of tumor diameter, lobulation sign and emphysema was only 0.67 (95% CI: 0.60~0.75). The nomogram based on radiomics and conventional imaging morphology features showed better performance on discrimination in the training set (AUC=0.79,95%CI:0.73~0.86), validation set (AUC=0.79, 95%CI: 0.69~0.89) and external validation set (AUC=0.77, 95CI%:0.63~0.90). Conclusions: A nomogram based on radiomics and conventional imaging morphology features can help to identify the worse prognosis of pure-solid clinical stage IA LUAD.