Dti in the Clinic: Evaluating the Effects of Smoothing

The accuracy and interpretation of results obtained by Diffusion Tensor Imaging (DTI) are largely influenced by several experimental parameter settings. In Voxel-Based (VB) analysis images are smoothed, in order to improve their Signal to Noise Ratio (SNR) and to reduce the impact of normalization and artifacts. This is a critical step and care must be taken so that directional information and boundary structures are preserved throughout the smoothing process. Although, Gaussian smoothing is generally applied as a preprocessing step in VB analysis, the choice of kernel dimension seems to strongly affect values and locations of the detected white matter abnormalities (1). Moreover, probably because of the intrinsic anisotropic nature of the diffusion Gaussian filtering is likely to induce consistent partial volume effects. An alternative approach (2) is to perform isotropic smoothing in homogeneous regions and anisotropic smoothing near boundary regions, thus enhancing the SNR of images, but preserving white matter boundaries. This smoothing technique involves an iterative process with a non–homogenous diffusion equation, and is performed individually for each diffusion–weighting direction, after optimisation based on the complete set of diffusion weighted images (2). Aim of the present study was to compare results obtained by anisotropic smoothing and conventional Gaussian smoothing. The two smoothing approaches were evaluated for their capability to separate differential effects between patients and controls in a standard VB analysis, in order to assess the best approach for a clinical setting, with its specific requirements regarding efficient calculation, accuracy, sensitivity and specificity for pathologic alterations.