Purpose. To improve the efficiency of CT in the preoperative diagnosis of grade 1 and grade 2/3 hypervascular pancreatic neuroendocrine tumors (PNETs) using a combination of texture features and contrast enhancement features.

Materials and methods. 81 patients with 82 morphologically proven pancreatic NETs and preoperative contrast enhanced CT were retrospectively enrolled. Patients are divided into Grade 1 and Grade 2/3 groups. For each group, the ratio of the densities of NET and unchanged pancreatic tissue, the relative tumor enhancement ratio of NET (RTE) in the arterial and venous phases of the study, 52 texture features for each phase of the CT-study were calculated and compared between the groups. The selection of predictors in the binary logistic model was performed in 3 stages: 1) selection of predictors using one-factor logistic models and C-index (AUC), under the conditions of padj <0.05 and AUC> 0.5; 2) selection using the Akaike information criterion (AIC); 3) selection of predictors using regularization (LASSO-regression after standardization of variables). The selected predictors were included in a binary logistic regression model without interactions.

Results. There were significant differences in 18, 28, 35, 16 texture features out of 52 for the native, arterial, venous, and delayed phases, respectively (p <0.05). After selection, the RTE and GLZLM_ZLNU features in the arterial and SHAPE_Compacity in the venous phases of the study were included in the final diagnostic model, which showed an accuracy of 84% in the prediction of grade 2/3 NETs.

Conclusion. We have developed a diagnostic model that includes texture and contrast enhancement features which increases the accuracy of CT in predicting the grade of hypervascular pancreatic NETs.