Modified demons registration for highly deformed medical images

In this paper, we propose a modified version of demons algorithm (non-linear intensity based approach) for registration of medical images. We revise demons algorithm and use anisotropic regularization in place of Gaussian smoothing to introduce an efficient mono-modal 2D image registration method. Here, in particular, the velocity field calculation step is improvised to obtain accurate deformation fields for prominently deformed images. Then, deformation fields are regularized by anisotropic diffusion smoothing method to acquire a smooth registered output It is relatively faster and accurate in comparison to other demons extended approaches. Registration accuracy obtained with the proposed registration method is 95.44% in comparison to 74.58% and 76.04 % in case of Accelerated demons method and Usha´s symmetric registration method respectively, when tested for highly deformed brain MRI images.