Radiomics Study For Predicting Bone Metastases And EGFR Mutation Status In Non-small Cell Lung Cancer

Objective: Based on computed tomography(CT)and contrast-enhanced T1-weighted magnetic resonance imaging(CET1 MRI),combined intratumoral and peritumoral regions,combined radiomics and deep learning methods,models were established for predicting whether bone metastases occurred in non-small cell lung cancer(NSCLC)patients and the epidermal growth factor receptor(EGFR)mutation status of NSCLC patients with bone metastases.Methods: This experiment contained 198 patients with NSCLC,including 88 patients with bone metastases and 110 patients with no bone metastasis.Before treatment,CT and CET1 MRI were performed on primary tumor and bone metastases of patients,respectively.Firstly,the features of intratumoral and peritumoral regions were extracted from CT images.For intratumoral,peritumoral,combined feature groups,the features were screened by Mann-Whitney U test,max-relevance and min-redundancy(m RMR),least absolute shrinkage and selection operator(LASSO),and then logistic regression models were established for predicting the occurrence of bone metastasis,respectively.In the further study,for 88 NSCLC patients with bone metastases(48 patients with EGFR mutation and 40 patients with EGFR wild-type),in order to predict the EGFR mutation status,the intratumoral,peritumoral,combined feature groups were extracted from CT and screened,and then used to establish models,respectively.Based on the CET1 MRI of NSLCL patients with bone metastases,deep learning features based on Mobile Net V3-Large network and radiomics features were extracted.The models based on radiomics features,deep learning features,combined features were established,respectively.In order to improve the predictive ability,features from CT and features from CET1 MRI were combined to establish a model.The receiver operating characteristic(ROC)curves were plotted and the area under the receiver operating characteristic curve(AUC)values were calculated to analyze and evaluate the models.Results: In the training and testing set,the model combined intratumoral and peritumoral features had the best performance in predicting bone metastasis(AUC values: 0.913,0.883,accuracy values: 0.879,0.864),for the prediction of EGFR mutation status,the model fused with CT features and CET1 MRI features had the best predictive ability(AUC values: 0.919,0.885,accuracy values: 0.881,0.862).Conclusion: This study analyzed the ability of radiomics in predicting bone metastasis and EGFR mutation status in NSCLC patients.The established models all showed good predictive effect and had the potential of auxiliary diagnosis,which could be used to assist personalized diagnosis and treatment.