CT-based Radiomics For Prediction Of Neoadjuvant Therapy Outcomes In Locally Advanced Rectal Cancer

Objetive There is a significant individual difference in the neoadjuvant therapy outcomes of locally advanced rectal cancer patients.About 70%of patients have a good response while about 30%of patients are insensitive to neoadjuvant therapy.However,there is no definitive method to predict the neoadjuvant therapy outcomes of locally advanced rectal cancer before treatment.This study is to investigate the value of predictive models based on CT radiomics in predicting the neoadjuvant therapy outcomes for locally advanced rectal cancer before treatment.Methods A total of continuous 168 patients who were diagnosed as locally advanced rectal cancer by enteroscopy and clinical evaluation,then underwent neoadjuvant therapy and radical surgery were retrospectively enrolled.The clinicopathological data including age,sex,CEA level,tumor site,clinical stage,clinical T staging,clinical N staging and CT data before treatment were collected.All patients were divided into a good response group(TRG0?1)and a poor response group(TRG2?3)according to the postoperative pathological tumor regression grade(TRG).All the patients were divided into training set and validation set in a 1:1 ratio by random.The chi-square test was used to compare the count data between the two groups,and the independent sample T test or rank sum test was used to compare the measurement data between the two groups.The A.K.software was used to extract the radiomics features from 3D-ROI which manual sketching on the CT enhanced nephrographic phase of locally advanced rectal cancer,and using lasso dimension reduction.Simultaneously,independent predictors of predictive value of neoadjuvant therapy for locally advanced rectal cancer were screened by multivariate logistic regression.Then,radiomics signature and nomogram model was constructed by logistic regression,which combined 6 radiomics features and all independent predictors respectively.The receiver operating characteristic curve(ROC curve)and calibration curve were used to evaluate the discrimination degree and calibration degree of the prediction model.Then the diagnostic performance was compared between the radiomics signature and the nomogram,as well as the training set and the validation set of the nomogram.Finally,the decision curve was used to evaluate the clinical value of radiomics signature and nomogram.Results There were 106 males and 62 females,average age was 55.58±11.52 years,66 patients with good response,102 patients with poor reaction.There were 396 radiomics features extracted from each patient,and three types of image ensemble features were selected after lasso reduction,including one gray histogram feature,one gray level co-occurrence matrix feature,and four run-length matrix features.Multivariate logistic regression analysis showed that radiomics score,elevated CEA(?3.4 ng/mL),and clinical Tstage(cT4)were independent risk factors of the neoadjuvant therapy efficacy prediction for locally advanced rectal cancer before treatment.The area under the ROC curve(0.881)was higher than the radiomics signature(0.791)(P<0.05),and the model had a high degree of calibration(P>0.05).The area under the ROC curve of the training set was 0.881(95%Cl:0.799-0.963),sensitivity was 0.906,specificity was 0.778,accuracy was 0.838,and the area under the ROC curve of the validation set was 0.817(95%Cl:0.715-0.919),sensitivity was 0.735,specificity was 0.789,and accuracy was 0.761.The decision curve showed that both nomogram and radiomics signature had some clinical application value.Conclusions Elevated CEA(?3.4 ng/mL),clinical T stage(cT4),and CT radiomics are independent risk factors in predicting neoadjuvant therapy of locally advanced rectal cancer.Both the CT radiomics signature and the nomogram model have high predictive efficacy and clinical value in predicting the neoadjuvant therapy outcomes of locally advanced rectal cancer,and they can be used to guide the development of individualized treatment options for clinical partners,and nomogram model prediction Performance is better than radiomics signature.