| Objective:To investigate the value of intravoxel incoherent motion（IVIM）imaging and parametric texture analysis in the differential diagnosis of pancreatic cancer（PC）,neuroendocrine tumor（NET）and chronic mass-forming pancreatitis（CMFP）and the prediction of T staging and pathological grading of PC.Methods:From December 2019 to January 2021,we prospectively enrolled 84 patients who underwent IVIM-DWI imaging of pancreas（b=0,50,100,150,200,300,500,800,1000,1500 s/mm2）and were diagnosed as PC or NET by biopsy or surgical pathology and CMFP by pathology or clinical follow-up.Patients were divided into groups according to the pathological results.Regions of interest（ROI）were delineated on the original diffusion-weighted image with a b-value showing the clearest boundary between the lesion and the normal pancreas tissue,and D,D*,f parameter maps and corresponding quantitative parameter values were generated.The differences of IVIM quantitative parameters among PC,NET and CMFP groups were compared,and the receiver operating characteristic curve（ROC）of the patients with statistically significant difference was drawn to get the threshold value,sensitivity and specificity of the corresponding parameters in differential diagnosis of different lesions.According to AJCC staging system,PC patients were divided into T1 group,T2 group and T3-4group.The differences of IVIM quantitative parameters among the groups were compared,and the ROC curve of those with statistically significant differences was drawn.According to the degree of differentiation,the PC patients were divided into poorly differentiated group and moderately-well differentiated group.The differences of IVIM quantitative parameters between the two groups were compared,and the ROC curve of those with significant differences was drawn.Using the Fisher parameter method（Fisher）,the minimum classification error and the minimum average correlation coefficient method（POE+ACC）,the mutual information measure method（MI）and the combined method of the three methods（FPM）of Mazda software to screen the extracted texture feature parameters,30 best texture features were automatically selected,and the differences of texture features among the groups were compared.Raw data analysis（RDA）,principal component analysis（PCA）,linear discriminant analysis（LDA）and non-linear discriminant analysis（NDA）of B11 module were used to discriminate PC,NET,CMFP and PC with different T stages.The results were expressed by misclassification rate.Results:⑴There were 43 patients in PC group,24patients in NET group,and 17 patients in CMFP group.There was no significant difference in age between PC group and NET group（t=0.474,P=0.637）;there was no significant difference in age between PC group and CMFP group as well（t=-0.525,P=0.601）.（2）The average values of parameters among the groups showed that the D value and D*value of PC group were lower than those of NET group and CMFP group,and the differences were statistically significant（P=0.000,0.004）.The threshold value of D value in differentiating PC and NET was 1.210×10-3mm2/s,the area under curve（AUC）was 0.784（sensitivity=0.792,specificity=0.791）.The threshold value of D value in differentiating PC and CMFP was 1.164×10-3mm2/s,the AUC was 0.717（sensitivity=0.588,specificity=0.767）.The threshold value of D*value in differentiating PC and NET was 17.070×10-3mm2/s,the AUC was 0.722（sensitivity=0.875,specificity=0.512）.And the threshold value of D*value in differentiating PC and CMFP was 1.210×10-3mm2/s,25.336×10-3mm2/s,the AUC was 0.689（sensitivity=0.765,specificity=0.605）.（3）The significant texture features of IVIM parameter images in identifying PC,NET and CMFP mainly origin from D image and f image,and are mainly gray level co-occurrence matrix features.The lowest misclassification rate was 17.86%（15/84）,which appeared in the texture features of D*map screened by FPM,and the discrimination method was NDA.（4）With the increase of T stage,D value gradually decreased,and the difference was statistically significant（P=0.015）.As for the comparison between adjacent groups,only the difference between T1 and T2 groups was statistically significant（F=0.195,P=0.047）.The threshold was 1.062×10-3mm2/s,the AUC was 0.760（sensitivity=0.857,specificity=0.727）.（5）The D value of PC in the moderately-well differentiated group was lower than that in the poorly differentiated group（P=0.034）,the threshold value was 0.949×10-3mm2/s,and the AUC was 0.844（sensitivity=1.000,specificity=0.750）.（6）Most of the texture features of IVIM parameter images were statistically significant in distinguishing different T-stage PCs.The main texture features of D and f images were gray level co-occurrence matrix features,and the main texture features of D*images were gray level run matrix features.The lowest misclassification rate was 0%（0/43）,which appeared in the texture features of D map and f map screened by FPM and D*map screened by Fisher coefficient method,with the discrimination method of NDA.And the misclassification rate of D map texture features screened by FPM combined with LDA method was also 0.Conclusion:（1）D value and D*value can be used as reference indexes to differentiate PC from NET and PC from CMFP,and the diagnostic efficiency of D value is slightly higher than D*value.（2）Texture analysis based on IVIM parameter images is helpful to identify PC,NET and CMFP.The gray level co-occurrence matrix features of D and f images can provide more information reflecting the heterogeneity of lesions.（3）D value decreases with the increase of T stage of PC（P=0.015）.D value can be used as a reference index to distinguish T1 and T2 stage of PC.（4）D value decreased with the increase of differentiation degree of PC（P=0.034）,which can be used as a biomarker to distinguish moderately-well differentiated PC from poorly differentiated PC.（5）Texture analysis based on IVIM parameter image is helpful to distinguish PCs in different T stages and the gray level co-occurrence matrix features of D image,f image and gray level run matrix features of D*image can provide more reference information. |
PDF Full Text Request