Reproducibility of MRI-derived liver surface nodularity score: analysis of patients with repeated MRI in various scanners.

Abstract

To assess trans-regional differences, reproducibility across different MRI scanners, and interobserver agreement of liver surface nodularity (LSN) score from routine liver MRI and to evaluate the correlation between LSN score and liver stiffness (LS) value on MR elastography. Ninety patients who underwent gadoxetic acid-enhanced liver MRI twice using different MRI scanners within a year were evaluated. On axial hepatobiliary phase images, right anterior (LSNRT_ANT), right posterior (LSNRT_POST), and left anterior hepatic surface (LSNLT) were chosen for the quantification of LSN score. Repeated-measures ANOVA, paired t test, Pearson's correlation coefficient analysis, and intraclass correlation coefficient (ICC) were used for statistical analysis. LSN scores from high to low were LSNRT_POST, LSNRT_ANT, and LSNLT, representing trans-regional differences (p < 0.001). Reproducibility of LSN measurement across different MRI scanners was high to excellent (ICC = 0.838-0.921). The mean difference between first and second examinations in LSNRT_ANT, LSNRT_POST, and LSNLT were 0.032 (p = 0.013), 0.002 (p = 0.910), and 0.010 (p = 0.285) for reader 1 and 0.051 (p = 0.004), 0.061 (p = 0.002), and 0.023 (p = 0.005) for reader 2. The first and second examinations were highly correlated in all hepatic regions (r = 0.712-0.839, p < 0.001). There was a low to moderate correlation between LSN score and LS value (r = 0.364-0.592, p ≤ 0.001), which was higher in the chronic hepatitis B (CHB) group than in the non-CHB group in all hepatic regions. In our study, LSN measurement on liver MRI showed trans-regional differences and excellent reproducibility across different MRI scanners. To use LSN score more widely, standardization of quantification software and selected hepatic regions is needed.