Anisotropic Kernel Smoothing of DTI Data

J. E. Lee, M. K. Chung, T. R. Oakes, A. L. Alexander Medical Physics, University of Wisconsin, Madison, WI, United States, The Waisman Laboratory for Functional Brain Imaging and Behavior, University of Wisconsin, Madison, WI, United States, Statistics, University of Wisconsin, Madison, WI, United States Introduction DTI measures, such as FA, trace and the eigenvector orientations, are very sensitive to noise. As the spatial resolution of DTI applications is increased, these noise effects are increased. Therefore, it is desirable to use image processing methods, such as regularization or smoothing, that will reduce noise effects while preserving the structural organization of DT field. Typical isotropic smoothing methods are effective for increasing SNR. However, isotropic smoothing also reduces the high spatial frequency image content and blurs the image features. The diffusion tensor describes a Gaussian distribution that tends to be oriented preferentially in the direction of white matter structures. Thus, we implemented an anisotropic Gaussian kernel smoothing method based on the diffusion tensor for preserving structural features while significantly reducing the noise. Theory