Application Of Magnetic Resonance Texture Analysis And Multiparametric Functional Imaging To Study Pancreatic Ductal Adenocarcinoma

Part Ⅰ: Preoperative prediction of pathologic grade of pancreatic duct adenocarcinoma based on MRI-derived radiomicsObjectives: To investigate the predictive value of contrast enhanced MRI-derived radiomics for preoperatively predicting pathologic grade of pancreatic duct adenocarcinoma(PDAC).Material and Methods: A total of 251 patients who underwent pancreatic MRI imaging examinations and surgical resection,pathologically confirmed as PDAC,were enrolled in this study from three hospitals.All cases were randomly assigned into the training set(n=176)and the validation set(n=75)with a ratio of 7:3 using FAEv0.3.6 software.Radiomics features of target lesions were automatically extracted from the arterialphase T1-weighted images.The MRI radiomics score(Rad-score)was established by using the least absolute shrinkage and selection operator logistic regression(LASSOLR)analysis in the primary cohort.A nomogram combining the MRI Rad-score with CA199 and tumor size was generated.Results: The prediction efficiency of radiomics model based on arterial-phase T1-weighted images could predict the pathologic grade,and the AUC of the MRI radiomics model in the training and validation set were 0.849(95% CI,0.799-0.907)and 0.743(95%CI,0.642-0.844),respectively;The specificity,sensitivity,PPV(positive predictive value),NPV(negative predictive value)and accuracy of the Radscore were 0.879,0.856,0.869,0.793 and 0.867 in the training set.The specificity,sensitivity,PPV,NPV and accuracy were 0.826,0.668,0.778,0.691 and 0.736 in the validation set.The combined nomogram based on the MRI Radscore,tumor size and CA199 could accurately predict the pathologic grade in the training set(AUC=0.864)and was validated in the validation set(AUC= 0.804).The calibration curves demonstrated good agreement.Decision curve analysis indicated that the novel combined nomogram is of clinical usefulness.Conclusion: Both our novel combined nomogram and the constructed radiomics signature based on fat-suppressed T1 WI of arterial phase could serve as a potential tool to predict the pathologic grade in patients with PDAC.Part Ⅱ: Preliminary study on predicting the pathological grade of pancreatic duct adenocarcinoma based on DCE-MRI,IVIM and DKI modelsObjective: To explore the value of different models(dynamic contrast-enhanced magnetic resonance imaging(DCE-MRI),intravoxel incoherent motion models(IVIM)and diffusion kurtosis imaging(DKI))in the prediction of pathological grading of pancreatic ductal adenocarcinoma,and the performance of the indexes was compared.Materials and Methods: A total of 67 patients who underwent dynamic contrastenhanced magnetic resonance imaging,multi-b-value diffusion-weighted imaging and diffusion kurtosis imaging scans and were pathologically confirmed as pancreatic duct adenocarcinoma by surgery or biopsy were collected,including 1 case of high differentiation and 49 cases of moderate differentiation,17 patients with poor differentiation;The diagnostic performance of the volume transfer constant(Ktrans),rate constant(Kep),extravascular extracellular volume fraction(Ve)and semiquantitative parameters(i AUC)in DCE-MRI model,the apparent diffusion coefficient(ADC),true diffusion coefficient(D),perfusion-related diffusion coefficient(D*),perfusion fraction(f)in IVIM model,the mean diffusion value(MD)and mean kurtosis(MK)in the DKI model,were compared by prediction probability.The time-intensity curve of pancreatic ductal adenocarcinoma were recorded.Results: DCE-MRI parameters(Ktrans,Ve,Kep and i AUC values),IVIM model parameters(f,D*)and(MK,MD values)in DKI models were significantly different in pancreatic duct adenocarcinoma with different pathological grades(P<0.05).The Ktrans,Kep and i AUC of the well-moderately differentiated group were lower than those of the poorly differentiated group,while the Ve of the well-moderately differentiated group was higher than that of the poorly differentiated group;the ADC and D values of the different differentiated groups were not significantly different;and MD were higher in the poorly differentiated group;The AUC of Ktrans,Kep,Ve,i AUC,MK,MD,f and D* were 0.714,0.756,0.791,0.710,0.737,0.784,respectively,in the differentiatding pancreatic ductal adenocarcinoma with different pathological grade.The AUCs of different model parameters for the judgment of pancreatic ductal adenocarcinoma with different pathological grade were presented in descending order: Ve > MD > Kep > f > MK > Ktrans > i AUC > D*;There are three types of TIC curves of pancreatic ductal adenocarcinoma including inflow,platform and outflow.Conclusion: DCE-MRI,IVIM and DKI models can provide reliable quantitative/semiquantitative parameters for the preoperative evaluation of pancreatic ductal adenocarcinoma with different pathological grading.The parameters Ve(DCE-MRI model)and MD(DKI model)have the highest diagnostic performance,and the model parameters can help to determine the pathological grade of pancreatic ductal adenocarcinoma.Part Ⅲ : MRI radiomics-based nomogram for preoperative prediction of peripancreatic lymph node metastasis in pancreatic ductal adenocarcinoma: a multicenter studyObjective:To develop and validate a 3-T MRI radiomics-based nomogram from multicenter datasets for pretreatment prediction of the peripancreatic lymph node metastasis(PLNM)in patients with pancreatic ductal adenocarcinoma(PDAC).Material and Methods: We retrospectively enrolled 251 patients with histologically or cytologically confirmed pancreatic ductal adenocarcinoma from 3 hospitals.Quantitative imaging features were automatically extracted from fat-suppressed T1-weighted images(FS T1WI)at the arterial phase.The MRI radiomics score(Rad-score)was established with the least absolute shrinkage and selection operator logistic regression(LASSO-LR)analysis in the primary cohort.A nomogram combining the MRI Rad-score with CA199,differentiation degree and tumor size was generated.The performance of the combined predictive nomogram was assessed in the external validation cohort with the area under the curve(AUC),decision curve analysis,and the calibration curve.Results: The AUC of MRI Rad-score were 0.868(95% CI,0.613-0.852)and 0.772(95% CI,0.659-0.879)in the training and internal validation cohort,respectively.The combined nomogram based on the MRI Rad-score could accurately predict LN metastasis in the training cohort(AUC=0.909)and was independently validated in both the external and internal cohorts(AUC = 0.835 and 0.805,respectively).The calibration curves demonstrated good agreement.DCA indicated that the novel combined nomogram is of clinical usefulness.Conclusion: Both our novel combined nomogram and the constructed radiomics signature based on FS T1 WI of arterial phase could serve as a potential tool to predict PLNM in patients with PDAC.Part Ⅳ: Development of arterial MRI-T1WI-Radiomics model to predict proliferation,invasion and metastasis of pancreatic ductal adenocarcinomaObjective: To develop a radiomics model based on the optimal radiomics reatures which are produced from arterial MRI-T1 weighted imaging(T1WI)to predict expression level of Ki-67 and P53 in pancreatic ductal adenocarcinoma(PDAC).Material and Methods: A total of 132 patients with pathologically confirmed PDAC who had abdominal MRI before operation and Ki-67 and P53 expression level test after resection were retrospectively collected.All cases were randomly divided into the training set(70%)and the test set(30%).The ROIs of PDAC were drawn manually by two radiologist,and the texture feature parameters of volume of interest(VOI)were extracted.Based on the 3Dslicer4.8.1 open-source software,we firstly performed data analysis three times by removing features with low variance,and LASSO to extract the optimal radiomics features form the training set.Secondly using the optimal radiomics features,the Logistic Regression(LR)classifiers used to establish a radiomics-Ki-67 and P53 expression prediction model.The receiver operation characteristic curve(ROC)of the test set and the training set was used for evaluating the feasibility of the prediction model.Results: 132 patients were divided into two groups based on Ki-67 positive rates ≥50% and <50%,P53 positive rates ≥60% and <60%.Through three times data analysis 12 and 15 optimal radiomics features for Ki-67 and P53 were selected,respectively.Using the optimal radiomics features to develop a prediction model for Ki-67,we found that the area under the ROC curve(AUC)of the training and the test set obtained by the LR classifiers were 0.863 and 0.829,respectively.The AUC of the training and the test set for P53 were 0.877 and 0.833,respectively.Conclusion: Radiomics features based on arterial MRI-T1 WI images is correlated with the expression of tumor Ki-67 and P53 in PDAC.Radiomics analysis can be used as a non-invasive predictive method for the expression of Ki-67 and P53 in PDAC.