An individualised radiomics composite model predicting prognosis of stage 1 solid lung adenocarcinoma

Abstract

To develop and evaluate a radiomics composite model for predicting disease-free survival (DFS) in stage I solid lung adenocarcinoma, and compare it to a simple radiomics model. Patients of pathological stage I solid lung adenocarcinoma treated with lobectomy (n = 119) were enrolled retrospectively. Three hundred and ninety-seven radiomics features per lesion were extracted from enhanced chest computed tomography (CT) imaging. Spearman's correlation coefficient and the LASSO (least absolute shrinkage and selection operator) regression model were used to reduce the dimension and select radiomics features. Univariate or multivariate logistic regression was used to build prediction models. A survival curve based on the radiomics composite model was plotted with Kaplan-Meier survival analysis to stratify the risk of recurrence. The confusion matrix, receiver operating characteristic (ROC) curve, and decision curve analysis were used to evaluate the performance of the prediction models. Recurrence occurred in 22.6% of patients. The survival curve of the radiomics composite model could accurately differentiate high-risk from low-risk patients. In the validation sets, the areas under the ROC curves (AUCs) of the pathological TNM stage (8th IASLC), clinicopathological model, radiomics model, and radiomics composite model were 0.587 (95% confidence interval [CI] 0.502-0.650), 0.629 (95% CI 0.558-0.682), 0.726 (95% CI 0.681-0.770), and 0.849 (95% CI 0.783-0.898), respectively. The prognosis of stage I solid lung adenocarcinoma predicted by an individualised radiomics composite model was more accurate than that of the simple radiomics model.