MRI-based Radiomics In Differentiating Pancreatic Ductal Adenocarcinoma From Mass-forming Chronic Pancreatitis

Objective :To investigate the efficacy of MRI-based radiomics in differentiating pancreatic ductal adenocarcinoma(PDAC)from mass-forming chronic pancreatitis(MFCP).Methods:The clinical and preoperative MRI imaging data of 62 PDAC and 31 MFCP patients confirmed by pathology from January 2015 to October 2022 were retrospectively analyzed.93 cases were randomly stratified sampled according to the ratio of 7:3 as the training set and validation set.The clinical and preoperative MRI imaging data were screened by logistic regression analysis,effective predictive factors were obtained,and a clinical model was constructed.Using 3D Slicer to segment pancreatic lesions one by one on T2 WI,portal venous phase,and delayed phase images.The Pyradiomics toolkit based on the Python environment was used for radiomics feature extraction.Feature selection was performed by interobserver agreement,t-test,or Mann-Whitney U-test,and least absolute shrinkage and selection operators(LASSO).Radiomics models based on different sequences of images were constructed using logistic regression.The receiver operating characteristic curve(ROC)was drawn,and the performance of radiomics models constructed based on different sequences of images was compared by the area under the curve(AUC),and the optimal radiomics model was selected.The optimal radiomics model was again compared with the clinical model.The Delong test was used to compare whether the difference in predictive performance between the models was statistically significant.Results:(1)Comparing the clinical data and MRI imaging data between the PDAC group and the MFCP group,there were statistically significant differences in age and CA199 indexes between the two groups(P<0.05).Compared with the MFCP group,the PDAC group was older,and the CA199 index was often abnormal.(2)T2WI,portal venous phase,and delayed phase images finally retained 8,15,and 8 radiomics features,respectively,and three radiomics models were constructed by applying logistic regression.The AUC values,accuracy,sensitivity and specificity of the T2 WI model in the training set and validation set were 0.9138,0.82,0.88,0.75;0.7589,0.68,0.84,0.50,respectively.The AUC values,accuracy,sensitivity and specificity of the portal venous phase model in the training set and validation set were0.9846,0.92,0.95,0.90;0.9630,0.89,0.89,0.80,respectively.The AUC values,accuracy,sensitivity and specificity of the delayed phase model in the training set and validation set were 0.9609,0.89,0.95,0.89;0.8737,0.82,0.89,0.75,respectively.The performance of the portal venous phase model was the highest,and there was a statistical difference between the T2 WI model in the training set and the validation set(P<0.05),and there was no statistical difference between the delayed phase model in the training set and the verification set(P>0.05).There was no statistical difference between the delayed phase model and the T2 WI model in the training set and the validation set(P>0.05).Combining the radiomics features with the inter-observer consistency test,the portal venous phase model was finally selected as the optimal radiomics model.(3)The AUC values,accuracy,sensitivity and specificity of the clinical model constructed based on effective predictive factors in the training set and validation set were 0.8237,0.75,0.86,0.67;0.8418,0.71,0.84,0.57,respectively.(4)The performance of the optimal radiomics model was better than that of the clinical model,and there were statistically significant differences between the training set and the validation set(P<0.05).Conclusion:The MRI-based radiomics model has high classification efficiency in the identification of PDAC and MFCP,and the performance of the optimal radiomics model is higher than that of the clinical model,which provides an effective and non-invasive identification method for clinical practice.