Anatomical landmark localization in breast dynamic contrast-enhanced MR imaging
In this article, we present a novel approach to localize anatomical features—breast costal cartilage—in dynamic contrast-enhanced MRI using level sets. Current breast MRI diagnosis involves magnetic-resonance compatible needles for localization . However, if the breast costal cartilage structure can be used as an alternative to the MR needle, this will not only assist in avoiding invasive procedures, but will also facilitate monitoring of the movement of breasts caused by cardiac and respiratory motion. This article represents a novel algorithm for achieving reliable detection and extraction of costal cartilage structures, which can be used for the analysis of motion artifacts, with possible shape variations of the structure caused by uptake of contrast agent, as well as a potential for the registration of breast. The algorithm represented in this article is to extract volume features from post-contrast MR images at three different time slices for the analysis of motion artifacts, and we validate the current algorithm according to the anatomic structure. This utilizes the level-set method  for the size selection of the region of interest. The variable shape of contours acquired from a level-set-based segment image actually determines the feature region of interest, which is used as a guide to achieve initial masks for feature extraction. Following this, the algorithm uses a K-means method for classification of the feature regions from other types of tissue and morphological operations with a choice of an appropriate structuring element to achieve reliable masks and extraction of features. The segments of features can be therefore obtained with the application of extracted masks for subsequent motion analysis of breast and for potential registration purposes.
Yin, X. X., Ng, B. W.-H., Yang, Q., Pitman, A., Ramamohanarao, K., & Abbott, D. (2012). Anatomical landmark localization in breast dynamic contrast-enhanced MR imaging. Medical and Biological Engineering and Computing, 50(1), 91-101. doi:10.1007/s11517-011-0772-9