Non-hyperfunctioning neuroendocrine tumours of the pancreas: MR imaging appearance and correlation with their biological behaviour

OBJECTIVE: To describe MR imaging features of non-hyperfunctioning neuroendocrine pancreatic tumours by comparing them to histopathology and to determine the accuracy of MR imaging in predicting biological behaviour. MATERIALS AND METHODS: After institutional review board approval, we retrospectively reviewed 45 patients with pathologically proven NF-NET of the pancreas and ≥1 preoperative MR/MRCP examinations. Of the NF-NETS, 29/45 (64.4 %) were G1 and 16/45 (35.5 %) were G2. Image analysis included the lesion maximum diameter, vascular encasement, extrapancreatic spread, signal intensity on T1- and T2-weighted, contrast enhancement features, and presence of metastases. Tumour vessel density was calculated on the histological specimen using a grid. RESULTS: The median maximum diameter of NF-NETs was 20 mm (range 5-200 mm). Eighty per cent of the NF-NETs were hypointense on T1-weighted images, 82.2 % were hyperintense on T2-weighted images, and 75.6 % were hypervascular. Overall MRI accuracy showed a mean AUC of 0.86 compared to pathology. Lesions with a maximum diameter of 30 mm irregular margins, absence of a cleavage plane with the main pancreatic duct, vascular encasement, extrapancreatic spread and abdominal metastases were significantly associated with malignant NF-NETs. No correlation was found between the tumour vessel density and contrast-enhanced MR imaging pattern. CONCLUSIONS: Hyperintensity on T2-weighted images and iso-/hypervascularity occurred in 27/45 (60.0 %) of NF-NETs. MRI identifies malignant NF-NETs with a sensitivity of 93.3 % and a specificity of 76.9 % (AUC = 0.85). KEY POINTS: • Non-hyperfunctioning neuroendocrine pancreatic tumours (NF-NET) pose a difficult diagnostic challenge. • On T2-weighted MRI, 82.2 % of neuroendocrine tumours appeared hyperintense. • MR imaging showed 0.94 sensitivity and 0.77 specificity in predicting biological behaviour. • The hyper-/isointensity during dynamic MRI did not correlate with vessel density at pathology