Evaluation of anisotropic filters for diffusion tensor imaging

Diffusion tensor imaging (DTI) measures, such as fractional anisotropy (FA), and trace are very sensitive to noise contained in the acquired diffusion weighted images. Typical isotropic smoothing methods reduce the high spatial frequency image content and blur the image features. We hypothesized that the diffusion tensor would be an approximate anisotropic Gaussian filter function because the blur will tend to be oriented parallel to the white matter structures. Thus, we implemented and evaluated an anisotropic Gaussian kernel smoothing method based on the diffusion tensor for preserving diffusion tensor structural features while significantly reducing the noise. We compared the diffusion tensor anisotropic filtering with isotropic Gaussian filtering, and a Perona-Malik (PM) filtering algorithm, which was derived from the intensity gradients of diffusion weighted images. Human brain DTI data with high SNR was used as a gold standard for evaluation. Overall, the anisotropic filters performed similarly, with slightly better performance using the DT anisotropic filter across the whole brain.