Background Non-hypervascular pancreatic neuroendocrine tumors (PNETs) showed slight or iso-enhancement in contrast-enhanced computed tomography (CE-CT), which shared similar imaging findings with mass-forming pancreatitis (MFPs). Purpose To explore the value of CT imaging features in differentiating the two diseases. Material and Methods Fifty-one patients with histologically proved MFPs (n = 27) or non-hypervascular PNETs (n = 24) were included. Two radiologists reviewed CT imaging findings and clinical features. Logistic regression analysis was performed to identify relevant features in differentiating non-hypervascular PNETs and MFPs. Receiver operating characteristic (ROC) curve analysis was used to show the performance of the optimal parameters in differentiating non-hypervascular PNETs and MFPs. Results A well-defined margin was more common in non-hypervascular PNETs ( P < 0.05) than that in MFPs. MFPs often occurred in older people ( P < 0.01) and the head–neck of the pancreas compared with non-hypervascular PNETs ( P < 0.05). Metastases only presented in non-hypervascular PNETs ( P < 0.05). CT values at venous phase and delay phase of MFPs were higher ( P = 0.010 and P = 0.029) than those in non-hypervascular PNETs. Logistic analysis showed gender, tumor margin, CT values at venous phase, and tumor components were independent predictors in differentiating the two lesions. The area under the curve (AUC) was 0.938 with a sensitivity of 87.5% and specificity of 92.6% for combined predicators. Conclusion Gender, tumor margin, CT values at venous phase, and tumor components were useful predicators in differentiating non-hypervascular PNETs and MFPs.
Machine learning based texture analysis predicts postoperative pancreatic fistula in preoperative non-contrast enhanced computed tomography
