Effects of interpolation methods in spatial normalization of diffusion tensor imaging data on group comparison of fractional anisotropy

This study investigated the effects on the measurement of fractional anisotropy (FA) during interpolation of diffusion tensor images in spatial normalization, which is required for voxel-based statistics. Diffusion tensor imaging data were obtained from nine male patients with attention deficit/hyperactivity disorder and nine age-matched control subjects. Regions of interest were selected from the genu of corpus callosum (GCC) and the right anterior corona radiata (RACR), with FA values measured before and after spatial normalization using two interpolation algorithms: linear and rotationally linear. Computer simulations were performed to verify the experimental findings. Between-group difference in FA was observed in the GCC and RACR before spatial normalization (P<.00001). Interpolation reduced the measured FA values significantly (P<.00001 for both algorithms) but did not affect the group difference in the GCC. For the RACR, the between-group difference vanished (P=.968) after linear interpolation but was relatively unaffected by using rotationally linear interpolation (P=.00001). FA histogram analysis and computer simulations confirmed these findings. This work suggests that caution should be exercised in voxel-based group comparisons as spatial normalization may affect the FA value in nonnegligible degrees, particularly in brain areas with predominantly crossing fibers.