Single versus multi-step non-rigid medical image registration of 3d medical images

Multi-model medical image registration is very important in medical image analysis and computer assisted surgery. Accuracy and speed are the two crucial factor of any registration algorithm. A fast radial basis function algorithm for non-medical image registration with improved accurac is presented in this article. The accuracy of the technique is improved by converting the one-step registration algorithm to multi-step i.e. three step registration algorithm. The global transformation accuracy of this technique has been evaluated by using two different anatomica landmarks sets. The former is to calculate the model parameters and the later is used to assess registration. Finally, we demonstrate that the multi-step technique yield better accuracy (using NMI) as compared to the one-step approach and target registration errors of about 2.91 mm on the registration of CT with its synthetically deformed version obtained from the venderbilt database. Our study shows that the multi-step fast RBF based registration is more effective in recovering larger deformation and do kept transformation smoothness than the one-step fast RBF based registration. (author)