Are pancreatic IPMN volumes measured on MRI images more reproducible than diameters? An assessment in a large single-institution cohort.

Abstract

To assess reproducibility of volume and diameter measurement of intraductal papillary mucinous neoplasms (IPMNs) on MRI images. Three readers measured the diameters and volumes of 164 IPMNs on axial T2-weighted images and coronal thin-slice navigator heavily T2-weighted images using manual and semiautomatic techniques. Interobserver reproducibility and variability were assessed. Interobserver intraclass correlation coefficients (ICCs) for the largest diameter measured using manual and semiautomatic techniques were 0.979 and 0.909 in the axial plane, and 0.969 and 0.961 in the coronal plane, respectively. Interobserver ICCs for the volume measurements were 0.973 and 0.970 in axial and coronal planes, respectively. The highest intraobserver reproducibility was noted for coronal manual measurements (ICC 0.981) followed by axial manual measurements (ICC 0.969). For the diameter measurements, Bland-Altman analysis revealed the lowest interobserver variability for manual axial measurements with an average range of 95% limits of agreement (LOA) of 0.68 cm. Axial and coronal volume measurements showed similar 95% LOA ranges (8.9 cm3 and 9.4 cm3, respectively). Volume and diameter measurements on axial and coronal images show good interobserver and intraobserver reproducibility. The single largest diameter measured manually on axial images showed the highest reproducibility and lowest variability. The 95% LOA may help define reproducible size changes in these lesions using measurements from different readers. • MRI measurements by different radiologists can be used for IPMN follow-up. • Both diameter and volume measurements demonstrate excellent interobserver and intraobserver reproducibility. • Manual axial measurements show the highest interobserver reproducibility in determining size. • Axial and coronal volume measurements show similar limits of agreement. • Manual axial measurements show the lowest variability in agreement range.