A computerized image analysis framework for dynamic contrast enhanced magnetic resonance imaging (DCE-MRI) with applications to breast cancer.

摘要

Dynamic contrast enhanced magnetic resonance imaging (DCE-MRI) provides a wealth of information about the anatomy of the breast, particularly in the setting of breast cancer diagnosis. In addition to the images it provides regarding the architecture of breast tissue, it also provides functional information about blood flow by means of the DCE study. The sensitivity of DCE-MRI has been reported at close to 100%, so the difficult tasks for the radiologist in reviewing breast DCE-MRI are: (1) discerning between which lesions are benign and which are malignant; and (2) doing so for a patient study that involves hundreds of images and is 4-dimensional. Because of the great detail and volume of information DCE-MRI provides, computational methods for both extracting and analyzing information derived from the images are useful in distilling the entire patient study down to the most salient images and features for the radiologist to examine. In this dissertation, computer-based methods developed for analyzing the data acquired in a breast DCE-MRI patient study are described.;In the first part, pre-processing methods used for aligning the images of the time-dependent DCE study are explained. Since segmentation is important for describing the morphology of the lesion as well as the region of interest for any subsequent quantitative analysis of a lesion, as a second step to pre-processing, a spectral embedding based active contour (SEAC) method for segmentation of lesions is developed and tested. A feature developed for extracting the spatiotemporal characteristics of breast lesions, termed textural kinetics, is then described, and its utility is demonstrated for distinguishing benign from malignant lesions as well as in identifying triple negative breast lesions, a lesion type that is extremely aggressive and has no targeted therapies. Finally, these quantitative methods are summarized in a computer aided diagnosis framework that provides insight into the biologic nature of breast lesion subtypes as well as for directing treatment and determining prognosis.