Applicability of Non-Rigid Medical Image Registration using Moving Least Squares

Abstract

A drawback of the non-rigid registration is its unpredictable nature of the deformation on the target image. Mapping every point on images can cause deformations even to regions, which are expected to remain rigid. A non-rigid registration is therefore desirable, that produces only local deformations where needed, while still preserving the overall rigidity. This work focuses on one such method called the Moving Least Squares (MLS) transformation and compares the results with Thin Plate Splines (TPS). An intensity based non-rigid registration algorithm is applied apriory, if the input medical images are from two different patients in order to facilitate for the selection of homologous control points in them. We compare the performance of both the techniques by calculating the Target Registration Error (TRE) at certain points and results are encouraging.