Primary breast abnormalities: selective pixel sampling on dynamic gadolinium-enhanced MR images

To evaluate a region-of-interest (ROI) analytic method that involves selective sampling of pixels within predetermined ranges of contrast material enhancement values ("thresholding") on magnetic resonance (MR) images of primary breast abnormalities.

Dynamic gadolinium-enhanced MR images were obtained in 105 women. ROIs were drawn to outline the full extent of lesions. Relative signal intensity increase was determined on a pixel-by-pixel basis on 1- and 2-minute postcontrast images, as was the maximum relative signal intensity increase. Thresholding was used to analyze each ROI, with the upper boundary defined by the highest pixel value and lower boundaries of 0%-100%.

Seventy-one invasive carcinomas and 37 benign lesions were analyzed. Narrower thresholding (i.e., larger percentage) resulted in an increase in all enhancement ratios (P < .0005). The enhancement ratio on 1-minute postcontrast images differed between benign lesions and carcinomas (P < .0005), but there were no significant differences in 2-minute and maximum ratios. Mean enhancement differences between benign and malignant lesions increased with narrower thresholding, but variability also increased linearly. Results of receiver operating characteristic analysis showed that thresholding did not affect the diagnostic usefulness of enhancement ratios.

Contrary to current opinion, selective sampling of the most enhancing areas of breast abnormalities may not provide any diagnostic advantage over the use of easily drawn, lesion-encompassing ROIs.