Development and validation of a CT-based nomogram to preoperative prediction of pancreatic neuroendocrine tumors (pNETs) grade.

Objective: It is challenging to determine the pancreatic neuroendocrine tumors (pNETs) malignancy grade noninvasively. We aim to establish a CT - based diagnostic nomogram to predict the tumor grade of pNETs.

Methods: The patients with pathologically confirmed pNETs were recruited in two centers between January 2009 and November 2020. PNETs were subdivided into three grades according to the 2017 World Health Organization classification: low-grade G1 NETs, intermediate-grade G2 NETs, and high-grade G3 NETs. The features on the CT images were carefully evaluated. To build the nomogram, multivariable logistic regression analysis was performed on the imaging features selected by LASSO to generate a combined indicator for estimating the tumor grade.

Results: A total of 162 pNETs (training set n = 114, internal validation set n = 21, external validation set, n = 48) were admitted, including 73 (45.1%) G1 and 89 (54.9%) G2/3. A nomogram comprising the tumor margin, tumor size, neuroendocrine symptoms and the enhanced ratio on portal vein phase images of tumor was established to predict the malignancy grade of pNETs. The mean AUC for the nomogram was 0.848 (95% CI, 0.918-0.953). Application of the developed nomogram in the internal validation dataset still yielded good discrimination (AUC, 0.835; 95% CI, 0.915-0.954). The externally validated nomogram yielded a slightly lower AUC of 0.770 (95% CI, 0.776-0.789).

Conclusions: The nomogram model demonstrated good performance in preoperatively predicting the malignancy grade of pNETs, and can provide clinicians with a simple, practical, and non-invasive tool for personalized management of pNETs patients.