Disappointing interobserver agreement among radiologists for a classifying diagnosis of pancreatic cysts using magnetic resonance imaging.

OBJECTIVES

To assess the degree of interobserver agreement of MRI in the diagnostic assessment of pancreatic cysts (PCs).

METHODS

Magnetic resonance imaging sets of images of 62 patients with PCs (32 with histological confirmation and 30 with clinical diagnosis) were reviewed by 4 experienced radiologists. Features scored included septations, nodules, solid components, pancreatic duct communication, and wall thickening (>2 mm). Radiologists were asked whether they considered the PC mucinous and if the PC was suspicious for malignancy. Furthermore, they had to choose a classifying diagnosis. Intraclass correlation coefficient (ICC) was used to measure agreement within the group.

RESULTS

Interobserver agreement for septations and nodules was fair (ICC, 0.36 and 0.23, respectively). Agreement for the presence of solid components was fair (ICC, 0.23), agreement for communication with the pancreatic duct was moderate (ICC, 0.53), and agreement for wall thickening was moderate (ICC, 0.44). There was fair agreement for the discrimination between mucinous and nonmucinous PC (ICC, 0.36). Agreement was fair (ICC, 0.26) for a classifying diagnosis and fair for the presence of malignant features (ICC, 0.33).

CONCLUSIONS

Interobserver agreement was poor to moderate for individual PC features, and there was fair agreement for a classifying diagnosis. Magnetic resonance imaging morphology alone did not allow for a reliable discrimination between different types of PC.