CT-based nomogram development and validation to predict SSTR2, VEGFR2 and MGMT expression for pancreatic neuroendocrine neoplasms

Objectives To establish and validate a model to predict somatostatin receptor 2 (SSTR2), vascular endothelial growth factor receptor 2 (VEGFR2) and O 6 -methylguanine-DNA methyltransferase (MGMT) expression in pancreatic neuroendocrine neoplasms (pNENs) based on CT images. Methods The data from 85 patients with 88 pathologically confirmed pNENs, who underwent contrast-enhanced CT examination before radical resection, were retrospectively collected. Immunohistochemical detection was performed for SSTR2 ( n = 86), VEGFR2 ( n = 53) and MGMT ( n = 84) expressions. The CT features were evaluated by radiologists. The patients were randomly divided into training and test dataset for each immunohistochemical group. Nomograms were developed based on CT features associated with these immunohistochemical expressions in the training sets, then validated and evaluated. Results Sex, tumour boundary and location were statistically different between SSTR2 positive and negative groups; sex, tumour maximum diameter, boundary, enhanced ratio in venous phase, and CT ratio in arterial and venous phases were statistically different between the VEGFR2 positive and negative groups; tumour maximum diameter, boundary, shape, CT ratio in unenhanced, arterial, and venous phases, and enhanced ratio in arterial phase were statistically different between the MGMT positive and negative groups ( P < 0.05). The nomograms showed good discrimination ability, with AUCs of 0.88 and 0.94 for the training and test sets in the SSTR2 group, 0.96 and 0.84 in the VEGFR2 group, 0.81 and 0.82 in the MGMT group, respectively. Each nomogram exhibited good calibration and clinical usefulness. Conclusion CT-based nomograms effectively predict SSTR2, VEGFR2 and MGMT expression in pNENs and may assist clinicians in pretreatment decisions.