A parent or guardian for each study participant gave informed consent, and children with typical cognitive development gave their assent, to participate in these studies as approved by the Institutional Review Board of the University of California, Davis. Study participants were recruited through local advocacy groups and the M.I.N.D. Institute clinic. Seventy-two male volunteers between the ages of 7.5 and 18.5 years participated in the study. All participants were healthy volunteers who met criteria in one of four diagnostic groups: low functioning autism (LFA, n=19), high functioning autism (HFA, n=19), Asperger syndrome (ASP, n=16), and typically-developing controls (CON, n=18).
Diagnostic assessments were conducted at the M.I.N.D. Institute clinic. The Autism Diagnostic Interview (ADI-R) (Lord, et al., 1994
) and Autism Diagnostic Observation Schedule (ADOS-G) (DiLavore, et al., 1995
, Lord, et al., 2000
) were administered by a clinician (B.L.G.J.), who had previously obtained reliability with an author of these measures (C. Lord). An IQ exam was administered to all participants. Depending on verbal ability, the appropriate test was used from the following: the Wechsler Intelligence Scale for Children, the Wechsler Abbreviated Scale of Intelligence (Wechsler, 1999
) or the non-verbal Leiter International Performance Scale-Revised (Roid and Miller, 1997
). A full scale IQ of 70 divided the high and low functioning autistic groups. Exclusionary criteria included diagnosis of Fragile X, seizure disorder, tuberous sclerosis, a primary diagnosis of obsessive-compulsive disorder, bipolar disorder, or any other major neurological illness. Details of the diagnostic assessments are available in a previous publication that included this cohort of subjects (Schumann, et al., 2004
A parent or guardian for each participant was present throughout the duration of the scan in an adjacent waiting room. Those study participants requiring anesthesia to undergo MRI were imaged at the UC Davis Hospital on a 1.5T GE Signa NV/I system (LFA, n=19; HFA, n=13; ASP=7). All remaining participants were scanned at the UC Davis Research Imaging Center on a 1.5T GE Signa NV/I system (HFA, n=6; ASP, n=9; CON, n=18). These systems were calibrated prior to scan acquisition and similar image acquisition on both scanners was experimentally validated (Lotspeich, et al., 2004
The protocol for scanning each participant included a three-dimensional coronal SPGR series (TR: 35 ms, TE: 6 ms, FOV: 24 cm, matrix: 256×256, section thickness: 1.5 mm, number of slices: 124, total scan time: 14:24 min), which was used for the volumetric assessment of the cerebellum. In addition, a two-dimensional sagittal T1 spin echo, two-dimensional PD/T2 interleaved double echo, and a diffusion tensor sequence were collected on all participants for other analyses.
Upon review of the images, ten participants were excluded from the study due to excessive movement, distorted images resulting from orthodontics, or additional diagnostic information that precluded the series from being used (LFA, n=1; HFA, n=4; ASP=1; CON, n=4). Within each diagnostic group, excluded participants did not differ from included participants with respect to age, IQ, or symptom severity.
Each coronal SPGR series was imported into ANALYZE 6.0 (Robb, et al., 1989
) and converted to cubic voxel dimensions of 0.9375 mm using a cubic spline interpolation algorithm. Images were reoriented along an axis through the anterior and posterior commissures. Measurements of total cerebral volume used in the current study were described in a previous report (Schumann, et al., 2004
). Briefly, each series of images was edited manually to remove non-brain structures, the brainstem, and the cerebellum. Using a Gaussian cluster multispectral thresholding tool, the ventricles were defined and excluded. Total cerebral volume was calculated from a mask of the remaining brain tissue.
Prior to volumetric analyses, the midsagittal area of the cerebellar vermis was measured (). The vermis was outlined on a single section that approximated as closely as possible the midline of the brain. The vermis was subdivided into lobule groups including lobules I–V, VI–VII, and VIII–X along the primary and prepyramidal fissures.
Figure 1 Sagittal series of MRI sections illustrating lobar segmentation of the cerebellum. Panels are arranged from midsagittal (top left) to lateral (bottom right). Lobule groups: I–V, lobules one through five; VI–II, lobules six through seven; (more ...)
The whole cerebellar volume was also measured. The total volume was segmented into a medullary core (the central white matter and deep nuclei of the cerebellum), the hemispheres (cortex and white matter), and the vermis (midline region of the cerebellum) (). The vermis and hemispheres were separately subdivided into lobules I–V, VI–VII, and VIII–X ( and ). All structures were manually defined by a set of raters who achieved greater than 0.96 inter- and intra-rater reliability on each of the structures. The MRI Atlas of the Cerebellum
(Schmahmann, et al., 2000
) and The Human Cerebellum
(Angevine, et al., 1961
) were closely consulted in the development of the region of interest tracing protocols. Detailed protocols for analysis of all of the cerebellar structures are provided in the Supplemental Materials
Coronal sections illustrating segmentation into whole cerebellar structures (A) and lobar structures (B). GRAY includes the cortex of the hemispheres; WHITE includes the medullary core and deep nuclei.
All statistical analyses were conducted with SPSS 16.0 (SPSS Inc., Chicago, Illinois). Prior to analysis of the anatomical data, age and IQ were compared between groups by an analysis of variance (ANOVA) to detect any group differences. Tukey’s post hoc test followed up on any main effects.
Due to the small and uneven sample size of the groups, there was a potential for the data to be distributed non-parametrically or adversely influenced by outliers. Tests of kurtosis and skewness were conducted on each anatomical measure to determine which type of analysis of variance should be used. The data were sufficiently normally distributed and only one measure, medullary core, was skewed with a right-handed tail.
Since all of the data were normally distributed, univariate and multivariate general linear models (GLM) were used to compare anatomical structures between groups. Age and total cerebral volume were entered as covariates in each analysis. Simple contrasts were made with the typically-developing group as the reference category. Tests were conducted at each anatomical level. Univariate GLM was applied to vermis area and a multivariate GLM was applied to the analysis of the area of vermal lobule groups. A univariate GLM was used for the cerebellum volume. Separate multivariate GLM were conducted for major parts of the cerebellum, and the lobule groups on the left and right hemispheres and vermis. A significance level of a two-tailed alpha of 0.05 was selected a priori. These analyses were repeated for comparison of the collective autism spectrum group to the typically-developing group without specific contrasts.
Subregions of the cerebellum were also analyzed as a ratio to the total cerebellar volume (i.e. normalized). Multivariate GLM, with simple contrasts, were repeated for the normalized volumes (the major parts of the cerebellum, and the lobule groups of the left hemisphere, right hemisphere, and the vermis).
The potential relationships between age and IQ and the anatomical measures within each group were evaluated by linear regression. A regression analysis was also performed for vermis volume in which vermis area was a regressor. This regression tested the degree to which the midsagittal area measurement predicted the volume measurement. A Pearson’s correlation analysis between total cerebellar volume and total cerebral volume determined whether the cerebellum was proportional to the cerebrum.