The Value Of Conventional And Texture Features Of CT Images In Pathological Grading Of Pancreatic Neuroendocrine Tumors

Background To evaluate the accuracy of conventional and texture features to predict the pathological grades of pancreatic neuroendocrine tumors(PNET).Patients and methods 102 cases of pancreatic neuroendocrine tumors,confirmed by postoperative pathology,were retrospectively enrolled in our study.Conventional imaging features including maximum diameter,calcification,cystic necrosis,pancreatic duct dilatation,enhancement pattern,CT values were measured and calculated.The region of interest(ROI)were manually drawn by using ITK Snap software and A.K.software was used for texture extraction in the arterial phase and the venous phase.Independent sample t test and chi-square test were used to compare the CT features between the low-grade group and the high-grade group.Binary Logistic regression was used to screen independent predictors.The logistic regression models of CT conventional features,CT texture features,and CT conventional features + texture features were constructed respectively.The predicted probability of each model was calculated.The ROC curve was drawn and the area under the curve was calculated to evaluate the respective and combined effectiveness of the two methods.Results There were 41 cases of pancreatic neuroendocrine tumor in low-grade group and 61 cases of pancreatic neuroendocrine carcinoma in high-grade group.The differences of conventional characteristics between the two groups including maximum diameter,enhancement pattern,cystic degeneration,calcification showed statistical significance(P< 0.05).There were statistically significant differences(P< 0.05)in Relative venous phase(PVP)and venous CT ratio(VCR).The texture parameters including frequency size,variance and valume count in arterial phase and mean deviation,min intensity,variance,frequency size,volume count,voxel value sum in venous phase were significantly different between the two groups(P < 0.05).Binary Logistic regression analysis showed that the maximum tumor diameter(P < 0.05)was an independent predictor.In the model 1 based on the conventional CT characteristics,the AUC value was 0.797,the sensitivity was 61%,and the specificity was 92.5%.In the model 2 based on texture features,the AUC value was 0.784,the sensitivity was 61%,and the specificity was 90%.Combined analysis of model 1 and 2 showed that AUC value was 0.917,sensitivity was 83.1%,and specificity was 90%.Conclusion Maximum tumor size was an independent predictor of pathologic grade of pancreatic neuroendocrine tumors.Conventional CT features and texture features can be used to predict the pathological classification of pancreatic neuroendocrine tumors.Conjoint analysis of the two features can improve the diagnostic efficacy and provide imaging evidence in order to provide options for clinical therapy.