Registration of DCE MR Images for Computer-Aided Diagnosis of Breast Cancer

TL;DR: A registration scheme that employs an elastic model as the deformation model and normalized cross correlation as the similarity term is presented and results indicate that a local similarity metric such as normalizedCross correlation can achieve desirable registration performance.

Abstract: The kinetic features of lesions on dynamic contrast-enhanced (DCE) breast MRI provide important diagnostic information. However, the same coordinates in raw DCE breast MR images at different times in the series may correspond to different physical locations in the subject due to respiratory motion, cardiac motion, and patient movements during image acquisition. In order to extract accurate kinetic features, an image registration step is necessary to spatially align the voxels across sequentially collected breast MR image volumes to ensure accurate time curve signal representation at each spatial location of the lesion. The challenges in registering DCE breast MR images are that the breasts undergo non-rigid motion and that the image intensity changes over time. This paper presents a registration scheme that employs an elastic model as the deformation model and normalized cross correlation as the similarity term. Symmetric consistency is used to evaluate the registration algorithm. Our results indicate that a local similarity metric such as normalized cross correlation can achieve desirable registration performance.