The Relationship Between EGFR Mutation And Spinal Bone Metastasis In Lung Adenocarcinoma Using MR-based Radiomic Signature

Objective:To explore the feasibility of predicting the mutation status of epidermal growth factor receptor(EGFR)gene by analyzing the MR imaging features of spinal bone metastasis in lung adenocarcinoma with radiomics.Methods:The imaging and clinical data of 164 patients with lung adenocarcinoma confirmed by pathology in the China Medical University Cancer Hospital(Liaoning Cancer Hospital)from 2016 to 2018 were analyzed retrospectively.All 164 patients underwent EGFR gene detection and MR routine spinal examination.All patients were confirmed to have spinal bone metastases by comprehensive imaging(CT,MRI or ECT / PET-CT)or pathology.All cases were reviewed and approved by the Ethics Committee of the Liaoning Provincial Cancer Hospital.Wilcoxon signed rank test and ?2 test were used to compare the differences of clinical characteristics(age,gender and smoking status)between the EGFR mutated group and the wild-group.The number of bone metastases in the EGFR mutated group was compared with that in the wild-group by Mann-Whitney U test.Pearson test was used to compare the number of cases and lesions of bone metastasis with soft tissue mass between the EGFR mutated group and the wild-group.Two doctors analyzed the signal characteristics(signal intensity of each sequence)and morphological characteristics(quantity,soft tissue mass)of all lesions.Then,ITK-snap software(version 3.6.0)was used to segment region of interest(ROI).After that,the ROI of bone metastasis was imported into AK software,and six kinds of texture parameters were selected for feature extraction: Histogram feature,Gray level size zone matrix(GLZSM)feature,Formfactor,Haralik feature,Gray level co-occurrence matrix(GLCM)feature and Grey level run-length matrix(RLM)feature.After preprocessing the extracted texture features,they were randomly divided into training group and verification group according to the proportion of 7:3.Rtree(random forest)was used in the mechanical learning method,and the prediction model was constructed.ROC(receive operating characteristics)curve analysis was carried out,AUC(area under ROC curve)was calculated and predicted,and AUC value,accuracy,specificity and sensitivity were used to evaluate the performance of radiomics model in predicting EGFR mutation state.Results: 1.Among 164 patients,103(62.8%)were in mutation group and 61(37.2%)in wild group.EGFR mutations were significantly more frequent in female and non-smokers(p < 0.05).The main mutation types were 19 del and L858 R,which accounted for 66.0% of the total mutations(68/103).2.On T1-weighted imageg,the lesions in the EGFR mutated group and the wild-group were slightly hypo-or hypointensity;on T2-weighted image,the lesions were slightly hypo-or hypointensity;on fat-suppressed T2-weighted image,the lesions were slightly hyperintensity or heterogeneous intensity.The number of bone metastases in the EGFR mutatied group was significantly higher than that in the wild-group,and the difference was statistically significant(p<0.05).There was no significant difference in the number of cases and lesions of bone metastasis with soft tissue mass between the EGFR mutatied group and the wild-group.(p > 0.05).3.Texture features extracted from T1 WI,T2WI and fat-suppressed T2 WI were 396,396 and 396 respectively.After ANOVA+MW,78,9 and 9 texture features were extracted respectively,and then Correlation Analysis was carried out.Finally,11,5 and 5 texture features were extracted respectively.4.On T1-weighted imageg,the AUC value obtained by mechanical learning rtree method was 0.568,the accuracy was 0.638,the specificity was 0.250,and the sensitivity was 0.844;on T2-weighted image,the AUC value was 0.530,the accuracy was 0.681,the specificity was 0.250,and the sensitivity was 0.911;on fat-suppressed T2-weighted image,the AUC value was 0.538,the accuracy was 0.618 and,the specificity was 0.167,and the sensitivity was 0.864.Conclusion:For patients with spinal bone metastases from lung cancer,the application of MR-based radiomics can provide important reference value for clinical understanding of EGFR mutation.