Constrained free form deformation based algorithm for geometric distortion correction of echo planar diffusion tensor images

In order to differentiate between normal and abnormal variations in brain diffusion tensor images, it is necessary to develop medical atlases. Atlas creation requires removal of spatial distortions in individual subject diffusion weighted images. In this paper we suggest a new approach using non-linear warping based on optic flow to map both baseline and diffusion weighted echo planar images to the anatomically correct T2 weighted spin echo image. The method is readily implemented and does not require a pre-processing step of rigid alignment. A global histogram matching precedes the base line EP image correction. A Markov random field based classification algorithm was implemented to cluster T2 weighted images into four different tissue type classes. This information was then used to synthesize diffusion based image models used in the warping algorithm to correct the geometric distortions in the diffusion weighted EP images.