Feature point detection for non-rigid registration of digital breast tomosynthesis images

We present an automatic approach to feature point detection that is well-suited to the non-rigid registration of digital breast tomosynthesis images The approach combines the scale saliency and the continuous intrinsic dimensionality of image structures in order to detect key feature points along the breast boundary and within the breast These feature points can be used as the control points for polyaffine transformation regularisation Experimental results show that non-rigid registration driven by such feature points yields good spatial alignment.