Application Of CT Radiomics In Predicting Synchronous Brain Metastasis Of Lung Cancer And The Prognosis Of Sclc After Chemotherapy

Part OneCT radiomics analysis for predicting synchronous brain metastasis in patients with lung cancerObjective: To evaluate the feasibility of radiomics based on non-contrast enhanced CT images in predicting synchronous brain metastasis(SBM)in patients with lung cancer.Materials and Methods: 371 patients(147 patients with and 224 without SBM)with lung cancer confirmed by histopathology from October 2012 to August 2020 were retrospectively enrolled.The optimal CT radiomics features were selected by using the least absolute shrinkage and selection operator(LASSO)algorithm.The multivariable logistic regression analysis was used for model construction.Three models(the clinicoradiologic model,radiomics model,and combined model)were constructed to predict SBM in the training set.The combined model was visualized using nomogram.The area under the receiver operating characteristic curve(AUC)and decision curve analysis(DCA)were used to assess the predictive abilities in SBM status using the three models.The results were validated in the testing set subsequently.Differences in AUCs between models were compared by using the Delong test.Furthermore,the predictive values of the three models for oligometastatic(1~3 lesions)or multiple(> 3 lesions)brain metastases in SBM,metachronous brain metastases(MBM),and total(SBM and MBM)groups of lung cancer patients were also investigated.Results: Two clinicoradiologic characteristics and six radiomics features were chosen for predicting SBM.Both the radiomics model(AUC = 0.870 and 0.824 in the training and testing sets,respectively)and the combined model(AUC = 0.912 and0.859,respectively)performed better predictive performance for SBM than the clinicoradiologic model(AUC = 0.712 and 0.692,respectively).All p values of Delong test < 0.05.The DCA indicated the clinical usefulness of the radiomics-based models.The radiomics model can also be used to predict oligometastatic or multiple brain metastases in SBM,MBM,and total groups(all p < 0.05).The clinicoradiologic model and the combined model had no statistical difference in SBM,MBM,and total groups(all p > 0.05).Conclusions: The radiomics model and the combined model can be utilized as valuable imaging biomarkers for predicting patients at high risk of SBM at the initial diagnosis of lung cancer.The radiomics model can also be used as an indicator for identifying oligometastatic or multiple brain metastases.Part twoApplication of delta radiomics in the prognosis of small cell lung cancer after chemotherapyObjective: To investigate the prognostic value of delta radiomics based on thoracic CT to predict local recurrence(LR)and overall survival(OS)for patients with small cell lung cancer(SCLC)after chemotherapy.Materials and Methods: From October 2012 to June 2018,a total of 136 patients with SCLC who underwent platinum-based chemotherapy(etoposide plus carboplatin or cisplatin,EP or EC)were recruited into this retrospective study.Patients were randomly divided into the training(n = 96)and testing(n = 40)cohorts at an approximate ratio of 7:3.Thoracic CT images were obtained at 3 time points(i.e.before chemotherapy,2 cycles after chemotherapy and 4 cycles after chemotherapy).Prediction of LR and OS was first performed using multivariate Cox regression with baseline clinical and radiological characteristics only to serve as the benchmark.Then,radiomics features were extracted from CT images of tumor at 3 single time points consequently.The changes of radiomics features were then calculated between different time points,namely delta radiomics features.Optimal radiomics features were selected from each radiomics feature sets of 3 single time points and 3 delta radiomics feature sets by using univariate Cox regression analysis and the least absolute shrinkage and selection operator(LASSO).Radiomics models were built to predict the LR and OS using multivariate Cox regressions.The predictive performance of the final 6 models for each outcome(LR and OS)were evaluated using the Harrell’s concordance index(C-index).Radiomic risk score(RRS)was calculated from the model of the training cohort of OS and LR which had the best predictive performance,respectively.For stratified analysis of survival outcomes,patients were divided into low and high risk groups according to the median RRS.Stratified analyses were also performed by stage(limited stage vs extensive stage)and chemotherapy regimen according to the median RRS.Results: Prediction performances using only baseline clinical and radiological characteristics,and single time points for LR and OS were not good(all C-indices <0.71).The model of R32(derived from changed features between 4 cycles after chemotherapy and 2 cycles after chemotherapy)performed the best prediction performance for OS and LR in both the training and testing cohorts(C-indices =0.850,95% CI 0.818~0.881 and 0.831,95% CI 0.785~0.883 for OS;C-indices =0.723,95% CI 0.670~0.776 and 0.737,95% CI 0.678~0.796 for LR,respectively).RRS of OS arm was found to be significantly associated with OS in the training cohort(HR = 12.918,95% CI 6.855~24.345,p < 0.001)and testing cohort(HR =8.259,95% CI 2.417~28.220,p < 0.001).Similarly,RRS of LR arm was associated with LR in the training cohort(HR = 3.333,95% CI 2.097~5.298,p < 0.001)and testing cohort(HR = 7.130,95% CI 3.012~16.876,p < 0.001),respectively.For risk stratification analysis,the univariate Cox regression analysis revealed that RRS of LR model was significantly associated with LR in LS-SCLC(HR = 3.461,95% CI2.065~5.802,p < 0.001)and ES-SCLC(HR = 4.677,95% CI 2.427~9.011,p < 0.001),respectively.Similarly,RRS of OS model was significantly associated with OS in LS-SCLC(HR = 9.638,95% CI 5.095~18.232,p < 0.001)and ES-SCLC(HR = 4.089,95% CI 2.137~7.825,p < 0.001),respectively.There were no statistical differences between EP and EC in stratified analysis(all p > 0.05).Conclusion: The baseline clinical and radiological characteristics or radiomics features at single time points were not sufficient for prognostic prediction of outcomes(LR and OS)of SCLC patients after chemotherapy.Delta radiomics derived from R32 can optimize the prediction of survival outcomes,and provided an opportunity for adjusting proper treatment strategy.