Inter‐ and intraobserver variability in the evaluation of dynamic breast cancer MRI

Abstract

Purpose: To quantify variations within and between observers ascribable to manual region of interest (ROI) placement in patients with breast cancer undergoing dynamic MRI. Materials and Methods:Expert and nonexpert observers independently outlined tumor ROIs on 30 dynamic T 1 -weighted (T 1 W) MRI scans on five occasions over two months. Lesion size (number of pixels) and kinetic parameter estimates, including the transfer constant (K trans ), were calculated for each ROI placement. Inter-and intraobserver variability was assessed with respect to the interval between drawings, lesion morphology, and observer experience.Results: For the nonexpert, the variability reduced with decreasing time intervals between ROI drawings (the coefficient of variance (wCV) values at two months, two weeks, one day, and same-day time intervals were respectively 11.6%, 10.7%, 4.8%, and 2.6% for lesion size, and 8.9%, 9.7%, 6.7%, and 3.2% for K trans ). For the expert observer, the variability was smaller overall and more constant, but improved for same-day ROI placements (region size wCV: 7.5%, 6.2%, 7.1%, and 3.7%; K trans wCV: 5.4%, 5.3%, 5.6%, and 4.5%). Conclusion:Significant observer variability in manual ROI placement occurs in dynamic MRI of breast cancer. For serial patient studies, ROI placements should be outlined at the same sitting to minimize observer error.

A Bayesian hierarchical model for the analysis of a longitudinal dynamic contrast‐enhanced MRI oncology study

“…Hence, a single mode is present in the posterior distribution. This variability, due to ROI definition, has been previously documented (38). The distributions of median K trans for patient 8 show the reverse effect, albeit much more subtle than patient 11, where the post-treatment distribution of median K trans appears to be bimodal but still spans a similar range of values.…”

“…It is important to also view the statistical images overlayed on suitable anatomical images to provide physiological explanations for the observed shapes of the voxel-level posterior distributions. Even when drawn by an experienced radiologist the tumor ROIs may substantially influence the quantitative results (38).…”

Detection and classification of contrast‐enhancing masses by a fully automatic computer‐assisted diagnosis system for breast MRI

“…Besides the morphologic interpretation of breast MRI, the assessment of dynamic curve types may also lead to considerable interobserver variability (12)(13)(14)(15). CAD systems have been implemented to perform an enhancement pattern of an entire lesion and to minimize interobserver variability (7,8,17,36).…”

“…However, for interpreting dynamic as well as morphologic parameters, substantial interobserver variability has been reported (11)(12)(13). The manual placement of a region of interest (ROI) for obtaining kinetic information of a contrast-enhancing lesion is reported to be quite subjective and may lead to inaccuracies if the ROI is placed in a less enhancing and thus less vital tumor part (12,14,15).…”

“…Although recent developments have focused on the automatic, computerassisted analysis of morphologic MRM features, eg, by the use of a classification tool, such as an artificial neural network (ANN), clinical assessment of MRM is still based on the observer's interpretation of morphology (11,13,15,(21)(22)(23)(24). There is clinical interest to integrate observer-independent morphologic analysis into ubiquitously available CAD systems and to link the computer-extracted features to the BI-RADS descriptors.…”