Value of Texture Analysis of Intravoxel Incoherent Motion Parameters in Differential Diagnosis of Pancreatic Neuroendocrine Tumor and Pancreatic Adenocarcinoma

Objective To evaluate the value of texture features derived from intravoxel incoherent motion (IVIM) parameters for differentiating pancreatic neuroendocrine tumor (pNET) from pancreatic adenocarcinoma (PAC).Methods Eighteen patients with pNET and 32 patients with PAC were retrospectively enrolled in this study. All patients underwent diffusion-weighted imaging with 10 b values used (from 0 to 800 s/mm 2). Based on IVIM model, perfusion-related parameters including perfusion fraction (f), fast component of diffusion (Dfast) and true diffusion parameter slow component of diffusion (Dslow) were calculated on a voxel-by-voxel basis and reorganized into gray-encoded parametric maps. The mean value of each IVIM parameter and texture features [Angular Second Moment (ASM), Inverse Difference Moment (IDM), Correlation, Contrast and Entropy] values of IVIM parameters were measured. Independent sample t-test or Mann-Whitney U test were performed for the between-group comparison of quantitative data. Regression model was established by using binary logistic regression analysis, and receiver operating characteristic (ROC) curve was plotted to evaluate the diagnostic efficiency.Results The mean f value of the pNET group were significantly higher than that of the PAC group (27.0% vs. 19.0%, P = 0.001), while the mean values of Dfast and Dslow showed no significant differences between the two groups. All texture features (ASM, IDM, Correlation, Contrast and Entropy) of each IVIM parameter showed significant differences between the pNET and PAC groups (P=0.000-0.043). Binary logistic regression analysis showed that texture ASM of Dfast and texture Correlation of Dslow were considered as the specific imaging variables for the differential diagnosis of pNET and PAC. ROC analysis revealed that multiple texture features presented better diagnostic performance than IVIM parameters (AUC 0.849-0.899 vs. 0.526-0.776), and texture ASM of Dfast combined with Correlation of Dslow in the model of logistic regression had largest area under ROC curve for distinguishing pNET from PAC (AUC 0.934, cutoff 0.378, sensitivity 0.889, specificity 0.854).Conclusions Texture analysis of IVIM parameters could be an effective and noninvasive tool to differentiate pNET from PAC.