Establishment of a MRI prediction model for solid pseudopapilloma of pancreas and nonfunctional neuroendocrine tumor / 中华胰腺病杂志

ABSTRACT

Objective:To analyze the MRI findings of solid pseudopapilloma of the pancreas (SPTs) and nonfunctional pancreatic neuroendocrine tumors (PNETs), and to establish and verify the prediction model of SPTs and PNETs.Methods:The clinical and MRI data of 142 patients with SPTs and 137 patients with PNETs who underwent surgical resection and were confirmed by pathology in the First Affiliated Hospital of Naval Medical University from January 2013 to December 2020 were collected continuously. Age, gender, body mass index (BMI), lesion size, location, shape, boundary, cystic change, T 1WI signal, T 2WI signal, enhancement peak phase, whether the enhancement degree was higher than that of pancreatic parenchyma in the enhancement peak phase, enhancement pattern, whether pancreatic duct and common bile duct were dilated, whether the pancreas shrank, and whether it invaded adjacent organs and vessels were recorded. According to the international consensus on prediction model modeling, patients were divided into training set (106 SPTs and 100 PNETs between January 2013 and December 2018), and validation set (36 SPTs and 37 PNETs between January 2019 and December 2020). The above characteristics of patients in training and validation set were analyzed by univariate and multivariate logistic regression, and a prediction model was established to distinguish SPTs and PNETs, and then visualized as a nomogram. The receiver operating characteristic curve (ROC) of the nomogram of training set and verification set was drawn, and the area under the curve (AUC), sensitivity, specificity and accuracy were calculated to evaluate the prediction efficiency of the model, and the clinical application value of the prediction model was evaluated by decision curve analysis (DCA). Results:Univariate regression analysis showed that there were significant differences on age, gender, lesion size, shape, cystic change, T 1WI signal, peak phase of enhancement, degree of enhancement in peak phase, pattern of enhancement and invasion of adjacent organs between SPTs group and PNETs group (all P value <0.05). Multivariate regression analysis showed that the older age, male patients, the smaller lesion, no high signal on T 1WI, the enhancement peak phase located in arterial phase or venous phase, and the enhancement degree in peak phase higher than that of pancreatic parenchyma were the six independent predictors of PNETs. The prediction model was established by using these six factors and visualized as a nomogram. The formula for predicting PNETs probability was 4.31+ 1.13×age+ 1.31×tumor size-1.29×female-4.18×high T 1WI signal+ 1.28×the enhancement degree higher than that of pancreatic parenchyma -4.69 ×enhancement peak in delay phase. The prediction model was visualized as a nomogram. The AUC values in the training set and validation set were 0.99(95% CI0.977-1.000) and 0.97 (95% CI 0.926-1.000), respectively. The sensitivity, specificity and accuracy in the training set are 98.00%, 94.34% and 96.12% and in the validation set were 86.49%, 97.22% and 91.78% respectively. The results of decision curve analysis show that the prediction model can accurately diagnose SPTs and PNETs. Conclusions:The prediction model established in this study can accurately differentiate SPTs from PNETs, and can provide important information for clinical decision and prognosis.