Autism is a developmental disorder characterized by social deficits, impaired communication, and restricted and repetitive behavior patterns (American Psychiatric Association, 2000
). Postmortem and structural magnetic resonance imaging studies have highlighted the frontal lobes, amygdala and cerebellum as pathological in autism (Amaral et al., 2008
), but there has yet to be agreement on the anatomical extent, timing, and consistency across subjects of the biological abnormalities (Williams and Minshew, 2007
Brain imaging studies of these developmental abnormalities often report an increased total brain volume (Hazlett et al., 2005
) and early acceleration in brain growth in autism, but it is not agreed whether this enlargement is restricted to childhood or continues into adulthood (Nicolson and Szatmari, 2003
Studies examining the differential contributions of gray and white matter to this abnormal growth in autistic patients have not had entirely consistent results, some detecting an increase in only gray matter or only white matter, but others finding it in both tissue types (Nicolson and Szatmari, 2003
). The localization of this brain volume increase is also debated (Bonilha et al., 2008
): frontal areas may contribute disproportionately to the volume increase (Carper et al., 2002
), but some suggest that more posterior brain regions are disproportionately affected (Hazlett et al., 2006
). A recent meta-analysis also found an overall increase in cerebellar volume (which may be proportional to the increase in total brain volume) and in the caudate nucleus, but found consistent reductions in the cross-sectional area of the corpus callosum (Stanfield et al., 2007).
Most traditional volumetric analysis have used region of interest analyses, using manual tracing of structures or automated segmentation (Yushkevich et al., 2006
). Measures of overall structure volumes may fail to detect subtle or highly localized anatomical differences between groups, and may overlook consistent regional differences in anatomical shape. Recently, computational mapping methods have been used increasingly to examine brain structure. Unlike traditional volumetric methods, statistical maps can detect highly localized group differences in brain morphology without the need for manual tracing or prior specification of regions of interest (Thompson et al., 2004a
). These methods have detected regional thinning of the corpus callosum (Vidal et al., 2006
), subtle hippocampal volume reductions (Nicolson et al., 2006
), and ventricular volume reductions (Vidal et al., 2008
) in autism, even when significant volume reductions in the brain as a whole were not detectable. In Vidal et al. (2008)
, surface-based statistical maps of group differences revealed subtle, localized reductions in ventricular size in patients with autism in the left frontal and occipital horns, which may reflect exaggerated brain growth early in life. Ventricular volumes measured using traditional methods did not differ significantly between groups. Other voxel-based anatomical mapping techniques, such as voxel-based morphometry (Ashburner and Friston, 2000
), have been used to detect subtle alterations in the corpus callosum in autism (Chung et al., 2004
). One study suggested that increases in temporal and parietal cortical thickness (Hardan et al., 2006
) may contribute to the volumetric increases in autism and may also relate to anomalies in cortical connectivity. Even so, another voxel-based mapping study had apparently conflicting findings (McAlonan et al., 2005
): children with autism had a significant reduction in total gray matter volume and significant increase in CSF volume. They had significant localized gray matter reductions within fronto-striatal and parietal gray matter and additional decreases in ventral and superior temporal gray matter.
To better understand the distribution and direction of these effects, further voxel-based studies are urgently needed.
Tensor-based morphometry (TBM) is a related structural image analysis technique that can reveal profiles of volumetric gains and deficits in patients versus control populations. TBM has not, to our knowledge, been applied to study autism. In TBM, a fluid image warping approach reshapes a set of brain images to match a common anatomical template. From these fluid deformation mappings, relative volume differences are computed between each individual and the anatomical template, and displayed voxel-by-voxel as a map. These maps may be compared across groups to identify regions with systematic volumetric differences. TBM has been used previously to characterize brain differences in various neurological disorders such as Alzheimer’s disease, semantic dementia, HIV/AIDS (Chiang et al., 2005
; Leow et al., 2006
; Hua et al., 2008
, Leporé, 2008a
), and neurodevelopmental disorders such as Fragile X syndrome (Lee et al., 2007
) and Williams syndrome (Chiang et al., 2007
). A similar approach has been applied to longitudinal scans to study brain changes over time (Thompson et al., 2000
; Chung et al., 2001
; Aljabar et al., 2008
). TBM may also be used to study statistical associations between regional brain volumes and relevant predictors, such as age, sex, or IQ (Chiang et al., 2007
This study had two goals. First we examined the three dimensional (3D) profile of systematic morphometric differences between patients with autism and controls using TBM. While TBM can reveal differences throughout the brain in 3D without a priori specification of regions of interest, we hypothesized that patients with autism would have diffuse volumetric excesses throughout the brain, particularly in the white matter, based on reports of white matter overgrowth in infancy. In line with prior reports, we anticipated localized gray matter abnormalities (either reductions or excesses, as the direction of the effects is not consistent in the literature) in temporal and parietal regions that include classical language processing systems. We also hypothesized that we would detect volume increases in the cerebellum, a region frequently reported as abnormal in autism.
In the original version of TBM, the determinants of the Jacobian matrices are derived from the local deformation field obtained after the nonlinear registration. These encode compressions and expansions, and can be used to map regional volume differences between patients and controls. In this study, we used the more general method described in Pennec (2004)
and Leporé et al. (2008a)
(summarized in ), in which the local deformation tensor field is analyzed statistically to detect local volume and local shape differences in tissue.