Interobserver agreement in MRI assessment of pancreatic cystic neoplasms: a retrospective single-center study.

purposeAccurate MRI characterization of pancreatic cystic neoplasms (PCNs) is essential for risk stratification and clinical decision-making. Despite their widespread use, the reproducibility of several MRI features incorporated into international guidelines remains poorly studied, particularly in MRI-based evaluation. We evaluated interobserver agreement among fellowship-trained abdominal radiologists in assessing MRI features of PCNs according to the revised Fukuoka guidelines. MATERIALS AND

methodsIn this retrospective single-center study, three fellowship-trained abdominal radiologists (5-10 years post-fellowship experience) independently reviewed MRI/MRCP examinations of 23 patients harboring 25 PCNs. Readers were blinded to clinical data and each other's assessments. A predefined checklist based on the 2017 revised Fukuoka guidelines was used to evaluate imaging features including cyst size, location, morphology, main pancreatic duct (MPD) diameter, ductal communication, mural nodules, solid components, wall characteristics, and wall enhancement. Interobserver agreement was quantified using Fleiss' kappa for categorical variables (with pairwise Cohen's kappa), and the intraclass correlation coefficient (ICC) for continuous variables. Kappa values were interpreted using the Landis and Koch scale.

resultsSubstantial agreement was observed for lesion location (kappa = 0.75, p < 0.0001), cyst size (kappa = 0.62, p < 0.0001), and morphological form (kappa = 0.60, p < 0.0001). Moderate agreement was noted for MPD diameter (ICC = 0.52, p < 0.0001), wall thickness (kappa = 0.31, p = 0.008), and signs of chronic pancreatitis (kappa = 0.36, p = 0.002). Fair agreement was found for ductal communication (kappa = 0.18, p = 0.041) and wall enhancement (kappa = 0.20, p = 0.089), while solid components showed no agreement beyond chance (kappa = - 0.03, p = 0.812). Mural nodules showed 100% agreement attributable to their absence in this cohort.

conclusionObjective MRI features such as lesion size and morphology showed consistently good agreement between readers, supporting their use as primary criteria in clinical decision-making. However, qualitative features including ductal communication, wall enhancement, and solid components show poor reproducibility, potentially leading to inconsistent risk stratification. Our results highlight the need for standardized interpretation criteria, structured reporting templates, and quantitative imaging tools to improve diagnostic consistency in PCN evaluation.