Prediction Of Immunoscore In Rectal Cancer Using MR-based Radiomics

PurposeTo predict the immunoscore in rectal cancer using MR radiomics,so as to better guide the clinical prognosis and treatment.Material and MethodsThis is a retrospective study.MR T2 WI and ADC images of 74 patients with rectal cancer confirmed by pathology were collected.All patients underwent radical surgery for rectal cancer.The specimens were stained by immunohistochemistry and the immunescore was calculated.There were 34 cases in high immunescore group and 40 cases in low immune score group.At the same time,the clinical data of 74 patients were collected for analysis,including age,gender,tumor location,tumor length,MRI-based T stage,lymph node metastasis,CEA,CA199.Two radiologists of different ages used ITK-SNAP software to manually sketch the lesions layer by layer and generate the region of interest(VOI),A.K software was used to extract image features,a total of 396 features were extracted.The univariate logistic regression and multivariate logistic regression were used to reduce the dimension of the features in turn.Then,machine learning method of logistic regression was used to establish the radiomics signature based on T2 WI and ADC images(RS-T2/ADC),clinical model and a nomogram model based on the combination of radiomics labels and clinical information to predict the immunescore.The prediction efficiency of each model was evaluated by the receiver operating characteristic curve(ROC).ResultsBased on the selected optimal feature subsets of T2 WI images,the radiomics signature was established(RS-T2).In the training samples and the testing samples,the AUC was 0.789(95%CI:0.652-0.891)and 0.75(95%CI:0.528-0.905),the accuracy was 0.706 and 0.826,the specificity was 0.786 and 0.833,the sensitivity was 0.609 and 0.818,the positive prediction was 0.700 and 0.818,the negative prediction was 0.710 and 0.833,respectively.Based on the selected optimal feature subsets of ADC images,the radiomics signature was established(RS-ADC).In the training samples and the testing samples,the AUC was 0.818(95%CI:0.685-0.912)and 0.78(95%CI:0.560-0.924),the accuracy was 0.745 and 0.739,the specificity was 0.893 and 0.667,the sensitivity was 0.565 and 0.818,the positive prediction was 0.812 and 0.692,the negative prediction was 0.714 and 0.800,respectively.Among the clinical factors,MRI-based T stage was correlated with immunescore(P < 0.05).In the training samples and the testing samples,the AUC was 0.769(95%CI:0.656-0.882)and 0.735(95%CI:0.550-0.920),accuracy was 0.784 and 0.739,specificity was 0.929 and 0.833,sensitivity was 0.609 and 0.636,positive prediction was 0.875 and 0.778,negative prediction was 0.743 and 0.714,respectively.Based on the radiomics signatures and clinical information,the combined model was established.In the training samples and the testing samples,the AUC was 0.919(95%CI :0.848-0.991)and 0.879(95%CI:0.727-1.000,accuracy was 0.863 and 0.826 specificity was 0.857 and 0.667,sensitivity was 0.870 and 1.000,positive prediction was 0.833 and 0.733,negative prediction was 0.889 and 1.000,respectively.ConclusionRadiomics signatures based on T2 WI and ADC images of rectal cancer,and the nomogram model combined with the radiomics signatures and clinical information can be used to predict immunescore of rectal cancer,so as to provide a basis for the prognosis judgment and treatment options of rectal cancer.