| Purpose: Microsatellite Instability(MSI)status,KRAS mutation status and KRAS mutation subtypes are key clinical indicators for evaluating patient's treatment strategies and predicting prognosis in the molecular classification of colorectal cancer(CRC).The purpose of this study was to investigate the predictive value of Computed Tomography(CT)-based radiomics models for MSI status,KRAS mutations,and KRAS mutation subtypes in patients with CRC.Materials and methods: 1.A total of 502 CRC patients in the two clinical medical centers were collected retrospectively,with 441 patients from center 1 as the training cohort and 61 patients from center 2 as the external validation cohort.Clinical indicators and radiological characteristics of the tumor were collected,and univariate and multivariate analyses were used to construct clinical model.Radiomics features were extracted from arterial-(AP),delayed-(DP),and venous-phase(VP)CT images,and features closely related to MSI status were retained as radiomics signatures using least absolute shrinkage and selection operator(LASSO)analysis.Based on radiomics signatures,this study constructed AP radiomics model,VP radiomics model,and DP radiomics model,and selected the model with the best predictive performance of MSI status from the three models.The clinical-radiomics fusion model was constructed by fusing the clinical model and the radiomics model with the best predictive performance form triphasic enhanced CT,and presented in the form of a quantitative nomogram to assess the individual probability of MSI status.The receiver operating characteristic(ROC)and the corresponding area under the curve(AUC)value,calibration curve and decision curve were used to evaluate the predictive performance and clinical usefulness of each model.2.This study involved 447 patients who underwent KRAS mutation testing and preoperative triphasic enhanced CT.They were categorised into training(n = 276)and validation cohorts(n = 117)in the ratio 7:3.Clinical indicators and radiological characteristics of the tumor were collected,and univariate and multivariate analyses were used to construct clinical model.Radiomics features were extracted from triphasic enhanced CT images,and features closely related to KRAS mutation status were retained as radiomics signatures using LASSO analysis.Based on radiomics signatures,this study constructed AP radiomics model,VP radiomics model,and DP radiomics model,and selected the model with the best predictive performance of KRAS mutation from the three models.The clinical-radiomics fusion model was constructed by fusing the clinical model and the radiomics model with the best predictive performance form triphasic enhanced CT.The ROC and the corresponding AUC value,calibration curve and decision curve were used to evaluate the predictive performance and clinical usefulness of each model.3.Our study included 447 patients with CRCs,who underwent preoperative enhanced CT examination and performed the KRAS mutations subtypes analysis.The first and second parts of our study has been proved that the DP radiomics model had better performance than the AP radiomics model and the VP radiomics model in predicting MSI status and KRAS mutation.Therefore,only the DP radiomics analysis was conducted in the present study.Radiomics features were extracted from delayed images,and LASSO analysis was used to screen the key features of KRAS mutation subtypes.Random forest analysis was used to construct radiomics models to predict KRAS mutation subtypes in patients with CRC.The ROC curve and corresponding AUC value,calibration curve and decision curve were used to evaluate the predictive performance and clinical usefulness of each model.Results: 1.Clinical variables age,tumor location and CEA level were independent predictors of MSI status,and the AUC values of the clinical model constructed by this predictors were 0.781(95%CI,0.722-0.840)and 0.919(95%CI,0.833-1.000)in the training cohort and the validation cohort,respectively.After removed irrelevant features by multiple steps,finally,6 AP radiomics features,10 VP radiomics features,and 16 DP radiomics features were retained as the final radiomics signatures for predicting MSI status.The AUC values of AP radiomics model,VP radiomics model,and DP radiomics model in the training cohort were 0.775(95%CI,0.715-0.835),0.827(95%CI,0.774-0.880)and 0.887(95%CI,0.847-0.927),respectively.The AUC values of the triphasic enhanced CT models in the validation cohort were 0.786(95%CI,0.644-0.929),0.810(95%CI,0.674-0.946)and 0.953(95%CI,0.903-1.000),respectively.In addition,the clinical-radiomics fusion model that incorporated both clinical model and DP radiomics model showed excellent performance,with an AUC,sensitivity and specificity were 0.898,0.821,and 0.840 in the training cohort,while0.964,1.000,and 0.904 in the validation cohort,respectively.The calibration curves showed that the clinical-radiomics fusion model had a good goodness of fit.The decision curve shows that the clinical-radiomics fusion model had more clinically practical than a single clinical model or radiomics model in predicting MSI status.2.Age,CEA level and clinical T stage were independent predictors of KRAS mutation status,AUC values of clinical models based on these predictors were 0.654(95%CI,0.593-0.714)in the training cohort and 0.575(95%CI,0.478-0.672)in the validation cohort,respectively.After rigorous feature screening,4 AP radiomics features,3 VP radiomics features and 7 DP radiomics features were retained as the as final signatures for predicting KRAS mutations.The AUC values of AP radiomics model,VP radiomics model,and DP radiomics model in the training cohort were0.711(95 % CI,0.654-0.767),0.692(95 % CI,0.634-0.750)and 0.752(95 % CI,0.699-0.806),respectively.AUC values of the triphasic enhanced CT models in the validation cohort were 0.723(95%CI,0.637-0.809),0.673(95%CI,0.582-0.764)and0.746(95%CI,0.662-0.830),respectively.The clinical-radiomics fusion model that incorporated both clinical model and DP radiomics model showed excellent performance,with an AUC,sensitivity and specificity were 0.772,0.792 and 0.646 in the training cohort,while 0.755,0.724 and 0.684 in the validation cohort,respectively.The calibration curves showed that the clinical-radiomics fusion model had a good goodness of fit.The decision curve shows that the clinical-radiomics fusion model had more clinically practical than a single clinical model or radiomics model in predicting KRAS mutation status.3.After removed irrelevant features,8 radiomics features were retained as the final signatures for the prediction of KRAS G13 D mutation,while 9 radiomics features were final signatures for the prediction of other KRAS mutation.Moreover,3features demonstrated a statistically significant in the discrimination of KRAS G13 D mutation and other KRAS mutation.Based on these relevant features,this study constructed 3 radiomics models for analyzing KRAS mutation subtypes.The constructed radiomics models showed promising performance in predicting KRAS mutation subtypes.The AUC values of the radiomics model for predicting KRAS G13 D mutation were 0.759(95CI,0.673-0.846)and 0.790(95CI,0.679-0.901)in training cohort and validation cohort,respectively.The AUC values of radiomics models for predicting other KRAS mutations in the training cohort and validation cohort were 0.753(95CI,0.697-0.810)and 0.715(95CI,0.614-0.815),respectively.The AUC values of the model for differentiating KRAS G13 D mutation from other KRAS mutation in the training cohort and validation cohort were 0.785(95CI,0.693-0.876)and 0.773(95CI,0.639-0.907),respectively.Conclusions: 1.The radiomics models constructed in this study had good performance for the prediction of MSI status in patients with CRC.The clinical-radiomics fusion model combined with clinical model and DP radiomics model has the best performance in predicting the MSI status of CRC,and the constructed model has been effectively verified by external validation cohort,showing good predictive and generalization ability of the model.The clinical-radiomics fusion model can be used as an auxiliary tool for preoperative detection of MSI status,and can be used in individualized therapeutic strategy planning and prognostic prediction.2.The radiomics model constructed in this study has good performance for the prediction of KRAS mutation status in patients with CRC.The clinical-radiomics fusion model that combines clinical model and DP radiomics model has the best predictive performance,and the constructed model has been effectively verified by internal validation cohort.The clinical-radiomics fusion model can be used as a potential imaging marker for preoperative detection of KRAS mutation status,and guide the selection of molecular targeted drug therapy for CRC.3.The radiomics model based on enhanced CT has promising performance for predicting KRAS mutation subtypes of CRC,and the constructed model has been effectively verified on the internal validation cohort.Our models can be used as an auxiliary tool for preoperative non-invasive detection of KRAS mutation subtypes,to achieve precise targeted therapy stratification,and help to develop effective personalized treatment strategies for patients with CRC.4.CT radiomics method is expected to effectively supplement the existing evaluation system of CRC molecular classification,assist the clinical development of precise treatment plans for CRC,and ensure the greatest clinical benefit for patients. |
PDF Full Text Request