| Objective: To evaluate the ability of CT radiomics to differentiate multiple lung adenocarcinoma into multiple primary lung adenocarcinoma and intrapulmonary metastasis of lung adenocarcinomaMethod:This study retrospectively analyzed 87 patients with multiple lung adenocarcinoma who underwent surgical treatment in Shenzhen People’s Hospital from January 2010 to October 2020,including 61 patients with multiple primary lung adenocarcinoma and 26 patients with intrapulmonary metastasis of lung adenocarcinoma.Multiple lesions in the lung were divided into primary lesions and secondary lesions according to the size of the lesions,and then the secondary lesions were manually segmented in CT images and radiological omics features were extracted.A machine learning model was built based on the radiological omics features screened by Lasso.At the same time,a machine learning model was constructed based on clinical features.AUC value was calculated by ROC curve to evaluate the classification performance of the two models.Result:From the CT images of each secondary lesion,851 radiomics features were extracted,and7 radiomics features were selected.Univariate analysis showed that preoperative clinical data（age）was significantly different between the group with multiple primary lung adenocarcinoma and the group with intrapulmonary metastasis of lung adenocarcinoma（P=0.011）.In the test group,the AUC,sensitivity and specificity of the model based on the radiomics characteristics were 0.79,0.76 and 0.83,while the AUC,sensitivity and specificity of the model based on the radiomics characteristics combined with age were 0.82,0.80 and 0.86 in the test group.Conclusion:This study demonstrated the feasibility of differentiating multiple primary lung adenocarcinoma from lung adenocarcinoma with intrapulmonary metastases,and the model constructed with radiomics combined with clinical features（age）had better classification performance than the model with radiomics alone. |
PDF Full Text Request