There is a great deal of evidence that schizophrenia is associated with white matter (WM) impairments. Voxelwise analysis methods have been developed to directly compare anatomical regions across individuals; however, the results of voxelwise analyses between schizophrenia patients and healthy controls have been inconsistent, likely due to methodological variation. This investigation was intended to determine whether registration choices, partial volume effects, and sample characteristics impact group comparisons of WM integrity.

Structural and diffusion-weighted images were collected from 31 schizophrenia patients and 27 healthy controls. Custom templates were created from the structural images of patients, controls, or both. Structural images were aligned to each custom template and a standard template, and correlation ratio (CR), a measure of similarity to the template, was calculated. Diffusion images were used to calculate fractional anisotropy (FA), a measure of white matter integrity. The FA images and WM segments of structural images were aligned to all templates and compared on a voxelwise basis across groups. Within FA images, the effects of age, group, and their interaction were tested with multiple regression. Within patients, the effects of illness duration and medication duration on FA were tested with semipartial correlations. Finally, a simulated dataset with enlarged ventricles was created and used in place of patients in the comparisons of CR and FA.

There was no evidence that CR differed across groups for any template. Group differences in FA and relative WM volume (i.e. smoothed WM segments) were identified at several locations that varied by template, though differences in both measures were consistently identified in the right inferior longitudinal fasciculus, parietal cingulum, and parietal superior longitudinal fasciculus. Covariation of relative WM volume did not have a consistent effect on FA group differences. The main effect of age was significant in widespread regions, including CSF, and the group-by-age interaction was significant in a small number of regions that did not survive correction for multiple comparisons.

Overall conclusions were that registration did not differ across groups, partial voluming likely affects FA comparisons but is difficult to account for, and changes in FA with age do not affect group differences.