Accounting for Changes in Signal Variance in Diffusion Weighted Images Following Interpolation for Motion and Distortion Correction

Introduction: Processing of clinical diffusion weighted images (DWIs) for Diffusion Tensor Imaging (DTI) and other applications requires correction for motion, eddy current distortions [1], and susceptibility induced EPI distortions [2]. These corrections are obtained by image registration techniques followed by interpolation of the original images. Image interpolation, however, alters the noise characteristics of the resulting diffusion image: in particular, the expected signal variance is no longer spatially homogeneous [3]. Not accounting for these changes in the intrinsic variance of the images may introduce spatial artifacts in tensor fitting, produce miscalculation in probabilistic tractography, and affect the outcome of robust tensor fitting methods like RESTORE [4]. In this work, we propose a method for accounting for signal variance changes in diffusion weighted images that underwent a series of distortion correction steps. We show the effects on images of the Chi-squared statistics of tensor fitting. This method is implemented in the software package TORTOISE [5] which is freely available for download (www.tortoisedti.org).