Thin-plate spline based automatic alignment of dynamic MR breast images

We propose a thin-plate spline based dynamic MR breast image registration approach under the assumption that only small alignment differences exist after global registration. Control point pairs between reference and sensed images are automatically and adaptively selected according to image gradient and local correlation of the image under translation and rotation. These landmarks are not homogeneously distributed, but are only located on skin lines, at edge corners and on contour ridges which visually demonstrate the structural deformation due to patient motion, normal breathing and cardiac motion. For MR breast sequences which have intensity changes after contrast injection, this method shows robust results and has less computational cost because of its dependence on fewer, but more effective, control points as compared to traditional iterative registration algorithms.