Abstract

Aim: To evaluate the discrimination ability of computed tomography (CT) texture analysis between pancreatic ductal adenocarcinoma (PDAC) and hypovascular pancreatic neuroendocrine neoplasm (PNEN).
Materials and Methods: Preoperative CT examinations of 49 patients with pathologically proven PDAC (35) and hypovascular PNEN (14) were evaluated retrospectively. Images in the portal vein phase of each tumor were assessed by two radiologists. Differences in texture features between PDACs and PNENs were analyzed with Mann Whitney-U test and logistic regression analysis.
Results: Patients with PNEN (50.4 ± 11.7 y) were significantly younger than the patients with PDAC (62.0 ± 11.1y) (p = 0.002). Nine CT texture features were found statistically significant between groups. NGLDM and GLZLM texture features showed the best diagnostic performance on multivariate analysis. Between these parameters, NGLDM_ Busyness showed the best diagnostic performance with the AUC value of 0.865. Combination of NGLDM_ Busyness and GLZLM_LZLGE improved the diagnostic performance at a significant level (AUC = 0.906, 95% CI = 0.747-0.988, p < 0.05).
Conclusion: CT texture features of single portal venous phase images may help to discriminate hypovascular PNEN from PDAC.

Key words: Computed tomography; pancreatic ductal adenocarcinoma; pancreatic neuroendocrine neoplasm;texture analysis