In addition to the described subjective two-dimensional morphologic measurements, objective, observer-independent, voxel-based, three-dimensional morphometry was undertaken. To perform this analysis, we transferred all images to a second workstation running image-analysis software (Analyze ver. 7.5; Biomedical Imaging Resource; Mayo Foundation, Rochester, MN). A 50 × 50 × 30-mm subvolume of the whole brain scan was reconstructed, centered on the middle of the optic chiasm parallel to the optic nerves and tracts. This subvolume was interpolated to 0.3 mm
3 isotropic voxel size. Voxel intensities were scaled to the global mean of all images of patients and control subjects to account for intensity differences due to a different general scanner signal. Mean images were calculated for the patient and the control group. A three-dimensional Gaussian filter (full width at half maximum 2 × 2 × 2 mm
3) was applied to the images using statistical parametric mapping (SPM99; Wellcome Department of Cognitive Neurology, University College London, Institute of Neurology, UK). Smoothing was critical, because, with the use of the central limit theorem, smoothing results in a more parametric distribution of the voxel values, allowing the use of standard parametric statistics.
17 Because there is no widely accepted method to normalize head subvolumes, images of the chiasmal region were not normalized and individual differences were thereby preserved. To find voxels that show significant differences between the patients and control subjects, a two-sample
t-test of the scaled intensities was calculated for each voxel by using SPM99. The theory of random Gaussian fields was used to assign significance to the resultant T-fields.
18 19 The statistical T-map was thresholded at
P < 0.05, corrected for multiple comparisons across the volume.