Prediction Of Vertebral Metastases From Lung Cancer Based On CT Radiomics

Objective:To evaluate the effectiveness of predicting vertebral metastasis of lung cancer based on CT radiomics approach.Methods:The data of of 19 patients with spinal metastases from lung cancer in our hospital from January 2016 to July 2019 were retrospectively analyzed.A total of 120 vertebral without obvious positive CT changes were included,including 63 cases of normal vertebral and 57 cases of bone metastases confirmed by other imaging examinations and follow-up examine.These patients were randomly divided into the training cohort(n=80)and the validation cohort(n=40).All patients underwent CT scan of spine before treatment,and had complete clinical data.The 3d-slicer software was used to segment and extract features of volume of interest(VOI).3d-slicer software was used for image segmentation and feature extraction.A total of 107 features were extracted.LASSO regression was used to reduce the amount of the features.After ten fold cross validation,the optimal adjustment parameter(λ)was selected when the mean square error was minimum.The optimal features was select to establish the logistics regression model.The rad-score was calculated.Multivariate regression analysis was used to estimate potential independent clinical predictors,and combined with rad score to establish a clinical-radiomics model.The receiver operating characteristic(ROC)curve was drawn.The area under the curve(AUC),sensitivity,specificity and Youden index of the training cohort and the validationcohort were calculated to evaluate the diagnostic efficacy of the two models.The AUC of the training cohort and the validation cohort were compared by Delong test.Calibration curve and clinical decision curve were used to judge the consistency between the predicted value and the actual value of the model and the clinical application value of the model.Results:Seven radiomics features were selected by LASSO regression,which were used to construct the radiomics model.The model had good diagnostic efficiency in training cohort(AUC=0.786,95%Cl:0.685-0.887,sensitivity 0.634,specificity 0.821,Youden index 0.620)and validation cohort(AUC=0.750,95%CI:0.589-0.91 1,sensitivity 0.875,specificity 0.625,Youden index 0.553).Multivariate logistic regression analysis showed that smoking history and nerve compression symptoms(P<0.05)were independent clinical predictors.Combined with rad-score,a clinical-radiomics model was established.The model had good diagnostic efficiency in training cohort(AUC=0.819,95%CI:0.728-0.91 1,sensitivity 0.780,specificity 0.718,Youden index 0.496)and validation cohort(AUC=0.831,95%CI:0.669-0.992,sensitivity 0.813,specificity 0.917)724).There was no significant difference in AUC between training cohort and test cohort(P>0.05).The results of decision curve analysis showed that the clinical-radiomics model had higher clinical benefit than the imageomics model.Conclusion:The CT-based radiomics can effectively predict the risk of vertebral metastasis of lung cancer,and the clinical-radiomics nomogram can be used as a visual tool to assist in the diagnosis of vertebral metastasis of lung cancer,and provide support for clinical decision-making of bone metastasis of lung cancer,so as to give patients early treatment and clinical prognosis evaluation.