MRI And Radiomics For Prediction Of Type Ⅱ Endometrial Cancer

Part Ⅰ:The predictive value of MRI and histopathology of curettage sample in type Ⅱendometrial carcinomaObjective To evaluate the diagnostic value of multi-parameter MRI combined with histopathology of curettage sample in the diagnosis of type Ⅱ endometrial carcinoma(EC),and to explore the optimal sequence for clinical application.Materials and Methods Patients with endometrial carcinoma confirmed by operation and pathology from September 2016 to October 2020 were analyzed retrospectively.According to the final surgical and pathological results,the patients were divided into two groups:type Ⅰ(common type)and type Ⅱ(untypical type)endometrial carcinoma.All patients underwent multi-parameter MRI examination before operation,diagnostic uterine curettage was performed after imaging examination,and immunohistochemical examination of ER,PR,p53 and Ki67 was performed on the curettage sample.The age,symptoms and menopause status of the patients were statistically analyzed,and the MRI signs of the patients were collected:tumor size,boundary,cystic-solid condition,relative T2 value,deep myometrial infiltration,ADC value of the solid part of the lesion,and the enhancement rate of dynamic enhancement in early arterial phase,late arterial phase,parenchyma phase and delayed phase,DWI signal,T2WI signal,MRI diagnosis of deep myometrial invasion,positive ascites and pelvic lymph nodes.Independent sample t-test was used for continuous variables,Mann-Whitney U test was used for categorical variables,chi-square test or Fisher accurate test was used for binary variables and proportions.The parameters with statistical differences between the two groups were included in multivariate Logistic regression analysis,the independent risk factors of type Ⅱ EC were screened,the clinical prediction model of type Ⅱ EC was constructed,and the receiver working characteristic curve(receiver operating characteristic curve,ROC)was drawn to evaluate the diagnostic effectiveness of the model.Results A total of 403 patients with EC were collected and statistically analyzed,including type ⅠEC(n=316)and type Ⅱ EC(n=87).There were 197 cases of G1 endometrioid carcinoma(62.3%),119 cases of G2 endometrioid carcinoma(37.7%)in type Ⅰ EC;and there were 46 cases of G3 endometrioid carcinoma(52.9%),21 cases of serous carcinoma(24.1%),7 cases of clear cell carcinoma(8.1%)and 13 cases of carcinosarcoma(14.9%)in type Ⅱ EC.There were significant differences in age,ER,Ki67,p53,lymph node positivity,deep myometrial invasion,ADC value,late dynamic enhancement arterial phase(DCE2),parenchymal phase(DCE3)and delayed phase(DCE4)between the two groups(p=0.021、p=0.032、p=0.007、p<0.001、p=0.011、p=0.024、p=0.028、p=0.036、p=0.032 及 p=0.021).There was no significant difference in clinical symptoms,menopause,tumor size,margin,PR,DWI signal,T2WI signal uniformity,relative T2 value,early dynamic enhancementrate(DCE1)and ascites between the two groups(p=0.85、p=0.558、p=0.119、p=0.915、p=0.934、p=0.376、p=0.382、p=0.384、p=0.429、p=0.839).The diagnostic efficiency of the predictive model based on conventional MRI features is AUC=0.669,95%CI(0.61-0.729),the accuracy is 51.4%,the sensitivity is 86.2%,and the specificity is 62.0%.The predictive model based on clinical and diagnostic curettage tissue immunohistochemical features is AUC=0.772,95%CI(0.72-0.824),the accuracy is 66.7%,the sensitivity is 83.9%,and the specificity is 62.0%.The combined model constructed by combining conventional MRI and immunohistochemical characteristics of diagnostic curettage tissue showed that p53,ki67 and DCE4 were independent predictors of type Ⅱ EC.The combined model constructed by the three factors had the highest diagnostic efficiency:AUC=0.784,95%CI(0.736-0.831),the accuracy is 67.2%,the sensitivity of the model was 88.5%,and the specificity was 59.8%.Conclusion1.The diagnostic efficiency of the predictive model based on conventional MRI features is AUC=0.669,with accuracy,sensitivity and specificity of 51.4%,86.2%and 62.0%,respectively.When combined with preoperative immunohistochemical features,the AUC is 0.784,with accuracy,sensitivity and specificity of 67.2%,88.5%,and 59.8%,respectively.2.ADC value and dynamic enhancement delay enhancement rate(DCE4)are helpful for the differential diagnosis of type Ⅱ EC and type Ⅰ EC.DWI and dynamic enhancement sequence can provide valuable information for the differential diagnosis of the two types of endometrial cancer.Part Ⅱ:The value of MRI-based Radiomics technique for the prediction of type Ⅱ Endometrial CancerObjective To explore the optimal method of constructing radiomic model by extracting a huge amount of radiomic parameters from preoperative multi-parameter MRI,and to explore the diagnostic value of radiomic model based on MRI for the diagnosis of type IIEC.To compare the value of radiomic model with conventional MRI in the diagnosis of type Ⅱ EC.Materials and methods A total of 403 EC patients confirmed by surgical pathology from September 2016 to October 2020 were retrospectively analyzed.According to the final surgical and pathological results,there were 316 cases of type Ⅰ EC and 87 cases of type Ⅱ EC.The radiomic features of the tumor were extracted from the ROI of tumor delineated on the T2WI,ADC,and DCE4 sequences.Multivariate Logistic regression analysis was performed using the optimal parameters to dig out the independent characteristics for the diagnosis of type Ⅱ EC and establish the imaging diagnosis model of type Ⅱ EC.Then the predictive models based on radiomic parameters were established.Seven groups of radiomic parameters were calculated:ADC,T2WI,DCE4,ADC+T2WI,ADC+DCE4,T2WI+DCE4 and ADC+T2WI+DCE4,and then the prediction models were constructed by using six algorithms:binary logisticlinear regression,random forest,self-help sampling synthesis,support vector machine(SVM),artificial neural network and naive Bayesian algorithms,respectively.In total,42 radiomic models were constructed.Receiver operating characteristic curve analysis was used to evaluate the diagnostic efficacy of each model.Delong test was used to compare the predictive performance of radiomic model and the combined model of MRI and IHC.Results The optimal model was a naive Bayesian model based on ADC+DCE4 sequences of the 42 radiomic models,with an AUC of 0.869(95%CI:0.793-0.945),the accuracy,sensitivity and specificity of the model were 86.7%,61.5%and 92.6%,respectively.The Hosmer-Lemeshow goodness-of-fit test showed that the model fits the data well(p=0.23).The predictive efficiency of radiomic model was higher than that of combined MRI and IHC model,and the difference between them was statistically significant.Conclusion1.The 42 predictive models based on 7 MRI sequences and 6 different algorithms showed good performance in the diagnosis of type Ⅱ EC.2.The radiomic model based on ADC+DCE4 sequence is most effective in the diagnosis of type Ⅱ EC,AUC=0.869(95%CI:0.793-0.945),the accuracy is 86.7%,the sensitivity is 61.5%,and the specificity is 92.6%.3.The diagnostic efficacy of radiomic model is higher than that of conventional MRI and preoperative immunohistochemical examination.Part Ⅲ:Comprehensive value of MRI,histopathology of curettage sample and radiomics for the prediction of type Ⅱ ECObjective To construct a predictive model and nomogram which is convenient for clinical application by combining the features selected from conventional MRI,radiomic and preoperative immunohistochemistry.To evaluate the value of this combined model in the diagnosis of type Ⅱendometrial carcinoma.Materials and methods Multivariate Logistic regression analysis was performed to construct a combined model based on the optimal parameters of conventional MRI,preoperative immunohistochemistry and radiomic features.A combined predictive model and nomogram of type Ⅱ EC was established.ROC curve analysis was used to evaluate the diagnostic value,and calibration curve was used to observe the goodness of fit of the model.Decision curve analysis was used to evaluate the clinical validity of the model.Results The AUC in the training set and the verification set of the nomogram were 0.951(0.910-0.993)and 0.915(0.651-0.978),respectively;the accuracy,sensitivity and specificity in the training set were 86.8%,72.1%and 90.2%,respectively;and the accuracy,sensitivity and specificity in the verification set were 85.0%,80.8%and 86.2%,respectively.The H-L goodness-of-fit test indicated that there was no significant difference between the predicted and the actual observed value of the model(p=0.532).The decision curve analysis showed that the model had obvious net benefit in clinical application.Conclusion1.A nomogram model was established by combining conventional MRI,preoperative immunohistochemistry and radiomic features.The model performance was good for the diagnosis of type Ⅱ EC,and is helpful for the treatment planning of EC2.The predictive performance of the nomogram model was higher than that of the models based on conventional MRI,preoperative immunohistochemical and radiomic features.The goodness of fit test showed that the predictive results fitted well with the actual data,and the decision curve analysis showed that the nomogram was valuable for clinical application.