Non-rigid medical image registration using adaptive knot selection in P-spline

Abstract

In medical imaging application, nonrigid registration is an important step which requires transformation to align the deformed floating image grid points spatially with reference image grids. The transformation includes interpolation which estimates the intensity values of floating image other than the grid points. For nonrigid registration, generally the transformation is modeled using B-spline basis function. As the acquired images are deformed and contains bias field, it is very time consuming to reform the grid at each level. Considering the registration process as an optimization problem, the grid points are adaptively chosen and updated iteratively with a new transformation in each step. In this paper, an adaptive knot placement algorithm is presented for P-spline interpolation method to achieve accurate alignment of local deformation with less computation time. The proposed algorithm is validated with 3 sets of simulated brain images with a known distortion and 2 real brain MR image data sets. Different performance measures such as MSE, RMS, NAE are used to evaluate the qualitative measure of the proposed scheme. Estimated transformation grid and evaluated performance measures show the improvement in the proposed algorithm as compared to the other existing state-of-arts.