Midsagittal T1-weighted proton magnetic resonance images (General Electric Signa at 1.5T, TE=13ms, TR=600ms, 2 excitations, slice thickness 4mm, in-plane spatial resolution 0.625mm) were acquired from 41 male and 6 female age-matched pairs of autistic and normal subjects ranging in age from 5 to 42 years. Informed consent was obtained from all subjects, and from the parents of all subjects under eighteen years of age. All autistic subjects satisfied the diagnostic criteria for autism of DSM-III-R [American Psychiatric Association 1987], the Childhood Autism Rating Scale [Schopler & al. 1988], the Autism Diagnostic Interview [LeCouteur & al. 1989], and the Autism Diagnostic Observation Schedule [Lord & al. 1989]. All autistic subjects were negative for fragile-X. All normal subjects had no history of neurologic or psychiatric disease. The outline of the corpus callosum was traced on an enlarged (2.1 times life-size) image by hand five times, and the area inside the tracing was computed, as per [Egaas & al. 1995]. The tracing whose area was the median of the five was used for intensity measurements. The corpus callosum as delimited by the tracing was automatically partitioned into five regions with boundaries at the sixth, twelfth, eighteenth, and twenty-fourth divisions as specified by Clarke & al. [1989], and the ratio of the median intensity of the set of pixels inside each of regions 1, 2, 4, and 5 to the median intensity of region 3 was computed, thus forming a normalised intensity measure suitable for comparison across scans. These normalised intensities from each region were subjected to a repeated-measures analysis of variance with age as a covariate.
Medians instead of means were used as a measure of central tendency in order to prevent perturbation by any extracallosal tissue at the margin of the tracing. Both short-T1 (cortical grey matter, and the septum and lateral ventricles) and long-T1 (the fornix) substances bound the corpus callosum in the midsagittal section that we studied, so such perturbation was an important consideration. To test the robustness of our median method, we intentionally misalligned some of our tracings with the callosal boundary and compared the medians obtained from this skewed measure with those obtained from the more exact one. We found that the median values were perturbed by at most one digital intensity step. In comparison, differences in median intensity between callosal regions were on the order of ten steps.