Fast Non-Rigid Medical Image Registration using a Parameterized Surface and Anatomical Landmarks

The paper presents a series of experiments which involve the use of the Fast Radial Basis Function algorithm for non-rigid medical image registration. The algorithm is a point-based registration technique which enables sub-second registration during the evaluation stage of standard-sized MR or X-ray CT datasets without loss of accuracy as compared to standard methods. In this paper we illustrate that the accuracy of the registration improves when using increasingly more salient feature points (i.e. landmarks and regular surfaces) without affecting the speed of the algorithm. Initially, a set of curves are extracted using a combined watershed and active contours algorithm, then tiled and converted to a regular surface using a global parametrization algorithm. Numerical results exhibit target registration errors less than 2mm on intra-subject registration of MR image datasets from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) Database whilst preserving sub-second performance for the Fast Radial Basis algorithm whereas competing algorithms exhibit slower performances.