To assess dimension measurement variability of liver metastases from neuroendocrine tumors (LMNET) on different magnetic resonance imaging (MRI) sequences.

In this institutional review board-approved retrospective study from January 2011 to December 2012, all liver MRI examinations performed at our department in patients with at least one measurable LMNET according to response evaluation criteria in solid tumors (RECIST1.1) were included. Up to two lesions were selected on T2-weighted MR images. Three reviewers independently measured long axes of 135 hepatic metastases in 30 patients (16 men, 14 women, mean age 61±11.4 (SD) years; range 28–78 years), during two separate reading sessions, on T2-weighted, diffusion-weighted MRI (DWI) (b; 50, 400, 800 s/mm²) and arterial, portal and late phases after intravenous administration of a gadolinium chelate. Intraclass-correlation coefficients and Bland–Altman plots were used to assess intra-and interobserver variability.

Intra- and interobserver agreements ranged between 0.87–0.98, and 0.88–0.97, respectively. Intersequence agreements ranged between 0.92 [95%CI: 0.82–0.98] and 0.98 [95%CI: 0.93–0.99]. 95% limits of agreement for measurements were −10.2%,+8.9% for DWI (b=50s/mm²) versus −21.9%,+24.2% and −15.8,+17.2% for arterial and portal phases, respectively.

An increase<9% in measurement and a decrease of −10% on DWI should not be considered as true changes, with 95% confidence, versus 24% and −22% on arterial and 17%, −16% on portal phases, respectively. DWI might thus be the most reliable MR sequence for monitoring size variations of LMNETs.