Study on radiomics to predict the aggressiveness of multiple primary invasive lung adenocarcinomas

Abstract

Abstract Our aim is to establish a model for predicting the aggressiveness of pathological subtypes of multiple primary invasive lung adenocarcinomas based on CT radiomics to provide a reference for preoperative planning in affected patients. Clinical data and CT images of patients who were diagnosed as having multiple primary invasive lung adenocarcinomas through postoperative pathological analysis from January 2016 to December 2020 in the Third Affiliated Hospital of Kunming Medical University were retrospectively analyzed. 3D Slicer software were used to perform the focal segmentation and feature extraction of the CT images. Five classification learners were employed to establish models for predicting the aggressiveness of pathological subtypes of the lung adenocarcinomas, and evaluate the performance of the prediction model based on the area under the Area Under the subject operating characteristic Curve (AUC). 204 patients were enrolled, surgical intervention was applied to 408 nodules. Of them, 36.8% nodules were Acinar-type.The analysis of the CT radiomics-based aggressiveness prediction model demonstrated that the training group showed that the AUC values of logistic regression (LR), random forest (RF), decision tree(DT), support vector machine(SVM), and adaptive boosting(AdaBoost) models were all within 0.7–1, and the testing group showed that the AUC values of the LR and RF models were all within 0.7–0.9. Our results indicate that acinar-type is the main pathological subtype of multiple primary invasive lung adenocarcinomas. LR and RF models presented a certain level of accuracy performance in predicting the aggressiveness of pathological subtypes of multiple primary invasive lung adenocarcinomas and thus facilitate preoperative planning in cancer patients.