Multiparametric MRI-based Radiomics Models For Preoperatively Predicting Rectal Adenoma With Canceration

ObjectiveRectal adenoma is a precancerous lesion of rectal cancer(RC).Rectal adenoma with canceration or not and the stage of canceration have important values for the formulation of treatment plans and the prognosis of patients.Therefore,the aim of our study is to explore the value of multiparametric MRI-based radiomics models for preoperatively predicting rectal adenoma with canceration.Materials and MethodsThis study retrospectively analyzed the clinicopathological data and imaging date of 53 patients with postoperative pathology confirming rectal adenoma(n=29)and adenoma with canceration(n=24)from December 2016 to February 2019 in Shandong Provincial Qianfoshan Hospital.Data collected included the patient’s gender,age,the location of tumor(the distance between the tumor and the anus:<5cm,low;5-10cm,median;>10cm,high),preoperative serological tumor indicators(carcinoembryonic antigen(CEA):normal range is from 0 to 5.0ng/ml;carbohydrate antigen CA-199:the normal range is from 0 to 37.0 U/ml;carbohydrate antigen CA-724:the normal range is from 0 to 6.9 U/ml).Finally.53 patients were enrolled in our study,all patients were divided into a training cohort(n=42)and a test cohort(n=11).All patients underwent preoperative pelvic MR examination,including high-resolution T2-weighted imaging(HR-T2WI)and diffusion-weighted imaging(DWI).Through the RadCloud v2.2(Huiying Medical Technology Co.,Ltd.Beijing,China)platform.draw the largest cross-sectional contour of the tumor as the Region of Intrest(ROI)in the HR-T2WI and DWI sequences respectively.According to RadCloud v2.2,a total of 1396 radiomics features were extracted from the HR-T2WI and DWI sequences respectively.The SelectKBest and least absolute shrinkage and selection operator(LASSO)were utilized for feature selection from the radiomics feature sets from the HR-T2W1 and DWI sequences and from the combined feature set with 2792 radiomics features incorporating two sequences.Five-fold cross-validation and two machine learning algorithms(Logistic Regression,LR;Support Vector Machine,SVM)were utilized for model construction in the training cohort.Three group models were constructed based on different source of features:Model1(radiomics features from HR-T2WI);Model2(radiomics features from DWI);Model3(radiomics features from HR-T2WI and DWI).Diagnostic performance of the models was evaluated by sensitivity,specificity,receiver operating characteristic curve(ROC)and area under the curve(AUC).ResultsTen,8,and 25 optimal features were selected from 1396 HR-T2WI,1396 DWI and 2792 combined features,respectively.In the training cohort,the AUCs of three models by LR algorithm were 0.678±0.196,0.857±0.153,0.976±0.024;the AUCs of three models by SVM algorithm were 0.698±0.138,0.839±0.158,0.976±0.024,Model3 showed better prediction performance than Model1 and Model2.In Model3,there was no significant difference between the LR and SVM algorithms(p>0.05),with AUCs in the test cohort of 0.867(95%CI=0.583-1.000)and 0.900(95%CI=0.583-1.000),respectively.Conclusion1.Multiparametric MRI-based radiomics models have the potential to predict rectal adenoma with canceration.2.The prediction performance of combination of HR-T2WI and DWI is better than that of single sequence.3.Both LR and SVM have equal excellent performance for construction of predictive models for rectal adenoma with canceration.