Grid Based Registration of Diffusion Tensor Images Using Least Square Support Vector Machines

In this paper, we present a non-rigid image registration method for DTMR images. This method consists of finding control points using a piecewise affine registration procedure and then estimating final transform between two images by minimizing corresponding Least Squares Support Vector Machine (LS-SVM) function of these control points. In our scheme, a fully symmetric grid points in the reference image is selected and the transformed grid points are computed using the results of piecewise affine registration. These control points are then employed to estimate final transform between images by minimizing the related LS-SVM function. In the transform functions, a finite strain (FS) based reorientation strategy is applied to adopt these methods for DTMR images. The main advantage of this method is that in estimating transform function. it considers all control points. Thus, each point in the reference image is transformed consistently with all other image points.