Intensity standardization in breast MR images improves tissue quantification

Computerized algorithms are increasingly being developed for quantifying breast MRI features for facilitating lesion detection and breast tissue segmentation in various clinical applications. One of the current impediments is the intensity non-standardness of the breast tissue in the acquired MR images across different cases, scanners, and/or patients. This degrades the performance of quantitative image processing. In this work, we investigate the usefulness of post-hoc intensity standardization of breast MR images by using a landmark-based nonlinear intensity mapping algorithm. The standardization algorithm is applied after correction of the images for background bias field non-uniformity. We then quantitatively compare the percentage coefficient of variation (%CV) of image intensity in the fibroglandular (e.g., dense) tissue region before and after standardization to evaluate the standardization procedure. In our experiments, we use 9 representative 3D bilateral breast MRI scans/cases constituting 18 breasts (a total of 504 tomographic breast MRI slices), in which we observe a significant decrease of the %CV in the standardized images, indicating that standardization significantly reduces the intensity variation for the fibroglandular tissue across these cases. Furthermore, we demonstrate for two segmentation methods that the standardization process leads to improved segmentation of the fibroglandular tissue. Our work suggests that intensity standardization following bias field correction may serve as an effective preprocessing step to support improved quantitative breast MR image processing and analysis, particularly for breast density quantification.