Observer variability in the interpretation of contrast enhanced MRI of the breast

The purpose of this study was to determine observer variability in the interpretation of contrast enhanced breast MRI and to evaluate its effect on the detection and differentiation of breast cancer. 57 women underwent breast MRI using spin echo and dynamic spoiled gradient-recalled sequences. Images were independently reviewed by three radiologists, two experienced and one newly trained in breast MRI interpretation. One of the experienced readers reviewed all examinations twice. Interpretation was based on lesion conspicuity, signal intensity, contour and enhancement pattern. Contrast uptake was assessed using region of interest (ROI) analysis of the dynamic images and calculation of a maximum enhancement index. Sensitivity and specificity in the diagnosis of malignancy irrespective of disease extent, and in the diagnosis of multifocal malignancy were estimated. 113 lesions were reported. Kappa coefficient estimations showed only a moderate agreement between the two experienced readers in rating morphological characteristics; the agreement between the newly trained reader and the experienced readers was even worse. Moreover, there was significant interobserver and intraobserver variation in the enhancement index measurements. Weighted kappa values indicated good agreement between the experienced readers in lesion and overall interpretation, excellent intraobserver agreement, but substantial disagreement between the newly trained reader and both experienced readers. All readers showed good sensitivity in cancer detection, but specificity was substantially lower. There is significant observer variability and a substantial learning curve in the interpretation of breast MRI, and variability in the ROI analysis of dynamic data. Further efforts to improve the reliability of ROI analysis and image interpretation are needed to help MRI realise its full potential in the clinical management of breast cancer.