Segmentation of Breast MRI Scans in the Presence of Bias Fields

Magnetic Resonance Imaging (MRI) is currently considered to be the most sensitive tool for imaging-based diagnostic and presurgical assessment of breast cancer. In addition to their valuable diagnostic contrasts, MRI scans also provide a superb delineation of breast anatomy, which facilitates a number of related clinical applications, among which are breast density estimation and bio-mechanical modeling of breast tissue. Such applications, however, require one to know the disposition of various types of the tissue, thus warranting the procedure of image segmentation. In the case of breast MRI, the latter is known to be a challenging problem due to the relatively complicate nature of measurement noises, which is particularly problematic to deal with in the presence of bias fields. Accordingly, in this work, we introduce a simple method that can be used to “gaussianize” the noise statistic, which allows the problem of image segmentation to be formulated in the form of a simple optimization problem. In this formulation, segmentation of breast MRI scans can be completed in only a few iterations as demonstrated by our experiments involving both in silico and in vivo MRI data.