Construction Of MRI-based Radiomics Model For Prediction Of The Prognosis Of Rectal Cancer Based On Tumor And Pelvic Substructure

Objective:To explore the predictive value of a preoperative radiomic prediction model based on primary tumor,mesorectal fat and pelvic muscle for prognosis of rectal cancer.Methods:A total of 244 patients with rectal adenocarcinoma confirmed by postoperative pathology in the First Hospital of Jilin University were retrospectively analyzed and randomly divided into training set(170 cases)and test set(74 cases)at a ratio of 7:3.Clinical data and preoperative highresolution MR Images of the rectum were collected.Univariate and multivariate COX regression analysis were performed on the clinical characteristics to obtain the characteristics with the strongest correlation with DFS of rectal cancer,and a clinical model was established according to the characteristics obtained after screening.ITK-SNAP software was used to contouring the primary tumor in the axial T2 WI sequence and DWI sequence of MR Images.In the axial T2 WI sequence,the mesorectal fat was delineated,and in the coronal T2 WI sequence,the muscle within the pelvic wall(piriformis muscle,obturator internal muscle and anal sphincter complex)were delineated.Subsequently,RIAS software was used to extract the radiomics features of the above regions of interest(ROI),and stepwise univariate COX regression,self-developed autocorrelation algorithm Corr＿Group,LASSO COX regression,and multivariate COX regression were used to reduce the feature dimension and establish the radiomic signature.Three radiomic models were established based on the radiomic signature of tumor,tumor combined with mesorectal fat,tumor combined with mesorectal fat and muscle within the pelvic wall,referred to as R1,R2 and R3 models,respectively.A combined model was established based on the radiomic signature of tumor,mesorectal fat and muscle within the pelvic wall combined with clinical features,and forest plot and nomogram were established to visualize the combined model.The relationship between the radiomic signature and DFS was verified by Kaplan-Meier survival curve and compared using log-rank test.The cumulative distribution function(CDF)and Restricted mean survival time(RMST)were plotted.The C-index and area under the curve(AUC)were used to evaluate the predictive value of the model,and the calibration curve was used to determine the agreement between the predicted probability and the actual result of the model.Results:For the clinical information of patients,p N stage,CA-199 and depth of tumor invasion were identified as clinical factors significantly related to the prognosis of rectal cancer(P<0.05).After screening the radiomics features extracted from preoperative MR Images,eight radiomic features based on tumor,five radiomic features based on tumor combined with mesorectal fat,and eight radiomic features based on tumor combined with mesorectal fat and muscle within the pelvic wall were obtained,and three radiomic signatures were established.All three radiomic signatures were significantly associated with 2-year DFS in the training cohort(P<0.001),while tumor-based and tumor combined with mesorectal fat-based radiomic signatures were not significantly associated with 2-year DFS in the testing cohort(P=0.42).The radiomic signature based on tumor combined with mesorectal fat and muscle within the pelvic wall was significantly associated with 2-year DFS(P=0.0037).Among the three radiomic models developed in this study,the R3 model had the highest prediction performance,with a C-index of 0.697(95%CI: 0.470-0.846)and an AUC of 0.73(95%CI: 0.52-0.93)in the test set.The combined model incorporating R3 and clinical features had the highest performance among all the five models,with a C-index of 0.724(95%CI: 0.493-0.875)and an AUC of 0.75(95%CI: 0.56-0.94)in the test set.The calibration curve analysis showed that the R3 model was superior to R2 model and R1 model and the combined model was superior to the clinical model in terms of consistency between model prediction probability and actual results.Conclusion:The MRI multi-objective radiomic features based on the primary tumor,mesorectal fat and pelvic muscle have good predictive value for the2-year DFS of patients with rectal cancer.The combination of radiomic signature and clinical features can further improve the performance of the prognostic prediction model.This study can predict the prognosis of rectal cancer patients before surgery and help to guide the individualized treatment of patients.