The Value Of CT Radiomics Features In Predicting ALK Mutation And Distinguishing ALK Mutation From EGFR Mutation In Lung Adenocarcinoma

Objective: To establish radiomics models based on enhanced CT to non-invasively predict anaplastic lymphoma kinase(ALK)mutation status and distinguish ALK from epidermal growth factor receptor(EGFR)mutation in lung adenocarcinoma.Methods:(1)Two hundred and ten patients with lung adenocarcinoma confirmed by surgical resection or puncture biopsy from April 2017 to September 2019 were selected as subjects.Predicting ALK mutation status and distinguishing ALK mutation from EGFR mutation were designed as two separate tasks.In each task,dataset was randomly divided into training set and testing set in a 7:3 ratio.In task one,147 cases were selected as training set and 63 cases were selected as testing set.In task two,109 cases were selected as training set and 47 cases were selected as testing set.(2)ITK-SNAP software was used to manually delineate the region of interest(ROI)on the arterial phase and venous phase images of CT in all patients.And then extracted quantitative radiomic features from the ROI using A.K.software.Features were selected from the training group using the minimum redundancy maximum correlation algorithm(m RMR)and the minimum absolute contraction and selection operator algorithm(LASSO),and then generated a radiomics score(Radscore).(3)Chi-square test and independent sample t-test were used to evaluate general clinical features.Univariate analysis was performed on the features of P<0.1 to find out the potential predictive factors.Then the independent risk factors were selected by multivariate logistic regression analysis.(4)The radiomics prediction model and radiomics nomogram prediction model were constructed according to the independent predictive factors of the selected training set.The accuracy of the prediction model in training group and validation group was evaluated by the area under the curve(AUC)of the receiver operating characteristic(ROC)curve.Decision curve analysis(DCA)was used to whether the radiomics nomogram model was robust enough for clinicians use.Results: In task 1:(1)There were 50 cases of lung adenocarcinoma in the ALK mutant(ALK Mut)group.The age range was 30 to 81 years,with a mean age of 56.3±10.2 years.In ALK wild(ALK WT)group,the age range was 38 to 81 years,with a mean age of 61.3±8.8 years.There was a significant difference in age,smoking history,TNM stage and metastasis between ALK Mut and ALK WT(P<0.05),but no statistical difference was found in sex,maximum diameter and location(P>0.05).(2)A total of 396 radiomics features were extracted from the enhanced CT images.Then 19 features highly related to ALK mutations were selected by m RMR and LASSO algorithms to construct the enhanced CT radiomics prediction model.AUC for radiomics prediction model was0.89 in the training group and 0.79 in the validation group.(3)The results of multivariate logistic regression analysis showed that age,smoking history and Radscore were independent predictors for the ALK mutation in lung adenocarcinoma.The AUC of nomogram model constructed by independent predictive factors had a little better predictive performance(0.80)than the radiomics model(0.79).DCA showed that when the threshold probability was in the range of 0.1-1.0,the performance of nomogram was better than that of clinical model.In task 2:(1)50 patients of lung adenocarcinoma were in the ALK mutant(ALK Mut)group,and the mean age was 56.3±10.2 years(range from 30 to 81 years).There were 106 cases of lung adenocarcinoma in the EGFR mutant(EGFR Mut)group,and the mean age was 60.1±8.6 years(range from 38 to 79 years).Compared with EGFR Mut,patients with ALK Mut were younger(P=0.009)and more likely to metastasize(P=0.019).(2)A total of 396 radiomics features were extracted from the enhanced CT images.Then the optimal 15 features were screened to construct radiomics model for the differentiation of ALK Mut and EGFR Mut.AUC for radiomics model was 0.94 in the training set and0.88 in the validation set.(3)The results of multivariate logistic regression analysis showed that age and Radscore were independent predictors for the differentiation of ALK Mut and EGFR Mut in lung adenocarcinoma.The AUC of radiomics nomogram constructed by independent predictive factors had the same predictive performance(0.88)with the radiomics model(0.88).DCA showed that when the threshold probability was in the range of 0.1-1.0,the net benefit of radiomics nomogram was higher than clinical model.Conclusions:(1)There were significant differences in age,smoking history,TNM stage and metastasis between ALK Mut and ALK WT patients.The radiomic model we established based on the enhanced CT had good performance for predicting ALK gene mutation in lung adenocarcinoma.(2)Compared with EGFR Mut,patients with ALK Mut were younger and more likely to metastasize.The prediction model established based on the enhanced CT radiomic features had high predictive value for the classification of ALK Mut and EGFR Mut.