| Objective:To explore the utility of radiomics based on multi-sequence MRI to preoperatively predict the status of Mucin4(MUC4)in pancreatic ductal adenocarcinoma(PDAC).Methods:This retrospectively study recruited patients with pathologically confirmed PDAC and preoperative multi-sequence MR examinationin our hospital from January 2016 to June 2019.The paraffin sections were collected for immunohistochemical detection the status of MUC4 in PDAC.Three feature sets were extracted from T1-weighted imaging(T1WI),the artery phase(A)and portal phase(P)of dynamic contrast-enhanced MRI,and the corresponding radiomics models were established.Univariate analysis and principal component analysis(PCA)were performed,the principal component features with the strongest discriminative power and cumulative frequency of 90%were selected.The selected features of each dataset were developed by multivariable logistic regressionfor classifiers.The most discriminative features from each dataset were combined to generate a joint radiomics model(M)and multivariable logistic regression was used for modeling.The tumor size and tumor differentiation degree were constructed a clinical model.The performance of different radiomics models were evaluated by the area under the receiver operating characteristic curve(AUC).DeLong test was performed to compare the AUC of M model,single-sequence radiomics models and clinical model.Results:There were 22 PDAC with high expression and 30 PDAC with low expression of MUC4.Each feature set included 350 radiomics features and four principal component features were selected from each feature set for model construction.A total of 12 principal component features were included in the M model.The AUCs of the T1WI,A,P,and M models were 0.612(0.453-0.772),0.665(0.512-0.819),0.662(0.514-0.811),and 0.909(0.821-0.997),respectively.The AUC of the clinical model was 0.666(0.511-0.821).Compared with T1WI,A,P and clinical models,M model achieved the best predictive performance(comparing the AUC of all models:P<0.05).Conclusion:Radiomics model based on multi-sequence MRI achieved good performance and had the potential ability in predicting the status of MUC4 in PDAC preoperatively. |
