Abstract #0343
Evaluation of Anisotropic Filtering for DTI as a Function of SNR
Lee J, Chung M, Alexander A
University of Wisconsin, Waisman Laboratory for Functional Brain Imaging and Behavior
The sensitivity to noise in DTI measures increases as the higher spatial resolution of DTI is gained. An anisotropic Gaussian filter based on the diffusion tensor was recently proposed for preferential image blurring along the white matter tracts directions, which minimizes partial volume averaging artifacts. In this study, we evaluated the performance of powered anisotropic Gaussian kernel smoothing using higher resolution human brain DTI data with different SNR levels. The study demonstrates that for SNR levels between 10 and 20, filtering does improve the overall accuracy of DTI brain measurements and both the optimal type and amount of filtering depends upon the SNR of the data. However, for moderately high SNR, filtering will likely introduce more error than is removed.
