To compare repeatability and reproducibility of four different methods of apparent diffusion coefficient (ADC) evaluation of liver parenchyma. In fact, repeatability and reproducibility assessment is mandatory in quantitative evaluations, however, these have not been accurately investigated in liver MR‐diffusion‐weighted studies.

Diffusion‐weighted sequences, b‐value = 0–1000 s/mm², were acquired on 30 healthy volunteers by a 1.5T scanner whose reliability has been validated by a phantom study. Four sampling methods, evaluating various parenchyma percentages by different‐sized region‐of‐interests (ROIs), were compared by two observers: 70% and 30% of the volume, 4%‐one‐ROI‐per‐segment, and 4%‐one‐ROI‐per‐slice in the right‐lobe. Ninety‐five percent limits of agreement and intraclass correlation coefficient (ICC) were calculated.

Complete measurements on the left lobe could be obtained in less than half of patients. The 4%‐one‐ROI‐per‐slice and 4%‐one‐ROI‐per‐segment yielded lower mean values compared with 30–70% volume methods (1343–1373 versus 1463–1560·10⁻⁶ mm²/s, respectively). Repeatability was acceptable (ICCs ∼ 0.80) whereas reproducibility was low (ICCs ≤ 0.45) for all methods. Averaging at least 3 measurements in middle‐lower sections of the right lobe improved both repeatability (ICCs to ≥0.87) and reproducibility (ICCs to 0.82) for 30–70% V methods.

ADC measurements were repeatable but not reproducible in our study. Reproducibility could be improved by taking averages on the right lobe with large ROI methods. Studies on procedures that standardize ADC measurements using more than two observers are needed. J. Magn. Reson. Imaging 2010;31:912–920. ©2010 Wiley‐Liss, Inc.