The Effects of Inter-slice Gap Thickness on Parametric Methods for Diffusion Tensor Imaging

Although high-resolution, whole-brain coverage is desirable for the acquisition of diffusion tensor (DT) images, it is not always acquired clinically due to time and scanner constraints. Although region of interest (ROI) analysis is less affected by this slice undersampling, there are likely more substantial implications for the use of parametric methods. This work will explore the effects of inter-slice gap size, or the absent data between acquired slices, on the sensitivity of statistical parametric mapping for detecting group differences in signal and spatial extent using DTI data. Simulated data sets were generated from a high-resolution, whole brain DTI acquisition with either no mask, a spherical or diffuse white matter mask applied to the unprocessed diffusion data. K-space manipulations were used to create data sets with larger voxels and these were additionally undersampled to simulate gap thickness. These simulated data sets were processed and the fractional anisotropy maps analyzed using statistical parametric mapping to calculate a voxel-wise T-test between masked and non-masked groups. The T-statistic map was then used as a classifier and the original mask as an indicator for true positives in receiver operating characteristic (ROC) analysis. From this analysis, it was shown that undersampled data sets diminish the effectiveness of parameteric mapping differentially based on the type of mask.