As the resolution of MRI imaging technology continued to improve, computerized data-analysis software was also developed, which permitted more precise quantification of imaging data. Voxel based morphometry (VBM) is an approach which measures the specific grey matter content of each volumetric pixel (voxel), thereby permitting automated delineation of grey and white matter. Simultaneously, images can be ‘warped’ onto a common template, in order to more precisely enumerate the differences between cases and control data sets (Hernandez-Garcia and Buschkuehl 2012).

This technology has allowed for a more exact quantification of whole brain and regional volumes, removing the need for manual mapping. In addition, specific subregions can be investigated in greater detail, or the whole brain can be surveyed, in terms of grey matter density by voxel cluster. Alternatively, strategies to correlate structural MRI data from VBM with social and cognitive function have also been employed (Ecker et al. 2012).

There is significant heterogeneity in the literature regarding findings. Differences in image filtering techniques, corrections applied for motion artifacts, statistical thresholds given multiple comparisons, selection of brain templates, as well as demographic and clinical differences in patient samples have complicated the interpretation of study results. Several meta-analyses (Duerden et al. 2012; Nickl-Jockschat et al. 2012; Via et al. 2011; Yu et al. 2011; Cauda et al. 2011) have attempted to synthesize findings, but even results across meta-analyses are variable (see Table 7.1). Two meta-analytic approaches in VBM include the activation likelihood estimate (ALE) (Laird et al. 2005) and the signed-differential mapping (SDM) methods. Both ALE and SDM combine data from published MRI coordinates to generate probability maps identifying grey and white matter clusters where findings converge across studies.

Areas that seem to emerge most consistently in these meta-analyses include volume loss in the putamen (Duerden et al. 2012; Nickl-Jockschat et al. 2012; Yu et al. 2011), volume gains in the caudate (Duerden et al. 2012; Yu et al. 2011) and volume loss in the hippocampus and amygdala (Nickl-Jockschat et al. 2012; Via et al. 2011; Yu et al. 2011). Significant differences also emerge in the prefrontal cortex (see Table 7.1). These findings must be interpreted with the consideration that structural brain changes in ASD may in part be due to comorbid psychiatric symptoms (Juranek et al. 2006), as well as concomitant treatment with psychotropic medications (Chakos et al. 1994). These and other confounding factors cannot be controlled for across meta-analyses. The functional implications of these areas will be discussed in more detail in Sect. 7.6 regarding regions of interest in ASD.

In addition to identifying regions of interests, the meta-analytic approaches described above have helped to elucidate age related differences in this condition. Findings span many heterogeneous regions, but similar to volumetric data, differences seem more pronounced and more numerous in the childhood/adolescent samples compared to adult samples (Duerden et al. 2012; Nickl-Jockschat et al. 2012).

Data from VBM analyses have also helped to inform the decision to condense previous diagnostic categories (i.e. autistic disorder, PDD-NOS, and Asperger’s syndrome) into a common autism spectrum disorder in the DSM-5 , given the absence of discrete biologically based disease categories (Via et al. 2011).

In cortical thickness mapping, the technology from voxel based morphometry is applied specifically to the cerebral cortex , to permit more precise anatomic analysis of this area. Similar to whole brain VBM, in cortical thickness analysis, a computer is programmed to extract the cortical grey matter from underlying structures, and this data is transposed onto a common surface template. In this way, the cortical grey matter can be further quantified in terms of cortical thickness, and cortical surface area. These measurements are of interest in ASD, as they are thought to represent separate embryological processes, which may inform a more specific etiology in neurodevelopmental disorders (Lui et al. 2011). Cortical surface area, for example, is thought to reflect neural stem cell proliferation and migration into the ventricular zone, which occurs early in embryogenesis (Rakic 2009). Cortical thickness, however, fluctuates throughout the lifespan, and is representative of a dynamic process involving cell layering, axonal myelination, synaptic growth and pruning (Shaw et al. 2008). It is important to note that typical cortical development involves a process of gradually increasing thickness during early childhood, which reaches a maximum prior to the onset of puberty, followed by gradual thinning which extends into early adult life (Shaw et al. 2008). Neuroimaging findings suggestive of altered developmental trajectories would be consistent with the conceptual understanding that ASD is a neurodevelopmental disorder.

Similar to the heterogeneity in VBM data, the literature with respect to cortical thickness in ASD is again inconsistent, and seems to vary significantly depending on the age of the study population. Early trials in a small group of children (mean age of 10 years) with ASD compared to controls identified increased cortical thickness, particularly in the temporal lobe, despite comparable brain volumes overall (Hardan et al. 2006). Follow up longitudinal MRI data by the same research revealed that patients with ASD had greater decreases in cortical thickness over time, which correlated with severity of symptomatology. The small sample size, however, resulted in marginal statistical significance once multiple comparisons were accounted for. Despite this, hypotheses regarding exaggerated cortical thinning, excessive elimination of synapses or impaired arborization were put forward from this data (Hardan et al. 2009).

Follow up investigations looking at cortical thickness in ASD have aimed not only to quantify specific differences in larger sample sizes, but also to document the magnitude and direction of aberrant cortical growth rates, taking into account the important effects of age.

Looking at broad age range (10–60 years), Raznahan et al. (2010) examined cross sectional imaging data in males with ASD vs. controls. Contrary to earlier findings, individuals with ASD were found to have smaller cortical volumes and thickness than controls at younger ages. Additionally, the rate of cortical thinning with age was attenuated for affected individuals, particularly in the temporal regions. Surface area did not change with time in either group (Raznahan et al. 2010) (see Fig. 7.2). Scheel et al. (2011) found similar results looking cross-sectionally at patients with ‘high functioning’ autism between ages 20–55 (Scheel et al. 2011). Along these lines, Doyle-Thomas et al. (2013) found attenuated cortical thinning in patients with ASD across a wide age ranges as well (7–39 years) (Doyle-Thomas et al. 2013). Recently, Ecker et al. published a report on cortical thickness, volume and surface area in men with ASD ages 18–42 years. They identified many non-overlapping areas of both increased and decreased cortical thickness and surface area in patients with ASD compared to controls, particularly in frontal and temporal regions, although age correlates were not provided (Ecker et al. 2013a).

On the other hand, analysis of cortical thickness data focusing on more narrow age ranges, and including a greater proportion of younger participants, seems to have yielded findings more consistent with those initially described by Hardan et al. (i.e. exaggerated thinning). Wallace et al. found that males with ASD ages 12–23 years, had thinner cortices overall, and then showed increased rates of cortical thinning during adolescence, with differences again most pronounced in the temporal regions (Wallace et al. 2010). Mak-Fan et al. (2012) looked at cross sectional data in children ages 6–15 years, and identified increased grey matter volume, cortical thickness and cortical surface area in younger children compared to controls, that then also underwent exaggerated thinning, surpassing controls on all measures between ages 10–12 years. Age effects were most pronounced in specific regions, including the medial precuneus and the left inferior frontal gyrus (Mak-Fan et al. 2012b).

In toddlers with ASD, cortical volume and surface area, (but not thickness) were found to be increased compared to controls at 2 years of age. The rate of cortical growth on longitudinal imaging between ages 2–5 years did not differ between groups, however, generating further questions regarding the specific chronology of aberrant neurodevelopment (Hazlett et al. 2011).

To summarize, a few studies suggest cortical overgrowth at very young ages; at age 2 this is marked by increased cortical volume and surface area, but potentially not thickness. School age children seem to display increased cortical thickness that is subsequently met with cortical dysmaturation throughout childhood and adolescence, suggestive mostly of exaggerated and then perhaps attenuated cortical thinning, although the specific magnitude and direction of this change varies depending on the study. The temporal lobe in particular is an area where rate of growth is most altered. Upcoming large scale, prospective neuroimaging trials may help to clarify some of this heterogeneity.

In addition to cortical thickness and cortical surface area, the specific pattern of gyrification in the cortex can also be quantified with mathematical indices. The gyrification index (GI) for example, is the ratio of the total length of the folded cortical surface (extending into each sulcus) divided by the outer cortical surface length (see Fig. 7.3). The rationale for studying these indices is supported in part by early literature where visible structural abnormalities were noted in the gyri of a small group of patients with ASD. Also, as described above with respect to cortical thickness and surface area, specific patterns of gyrification may also be reflective of altered neurodevelopmental processes, or underlying differences in brain cytoarchitecture. For example, variations in neuron layers, cell density, and thalamocortical connections are noted on histology comparing tissue from sulci vs. gyri. Similarly, sulci formation in early life is hypothesized to be driven by tensile pull on areas with greater connectivity to distant regions [summarized in (White et al. 2003)]. Accordingly, altered functional connectivity may yield measurable structural changes in cortical morphology.

Early in the MRI literature, Piven et al. (1990) found significant structural abnormalities in the cortices of several adolescent males with ASD compared to controls. Of 13 patients, 5 displayed noteable polymicrogyria (multiple small cortical folds), and 2 had macrogyria (large cortical folds), as observed by radiologists blinded to diagnosis (Piven et al. 1990). Control participants, on the other hand had typically appearing brains. With enlargement of sample sizes, the more striking structural abnormalities in the brains of patients with ASD were for the most part washed out, necessitating a more precise mathematical approach to quantify specific differences.

Hardan et al. (2004) for example, found preliminary evidence suggesting increased gyrification in the frontal lobes of adolescent but not adult patients with ASD compared to controls, as measured by statistical comparison of GI. Patterns emerged suggesting changes in cortical folding with age in affected patients but not in controls (Hardan et al. 2004). In a population of patients with high functioning autism, increased cortical folding was noted in the frontal lobe (Jou et al. 2010). More recently, high-resolution analysis on 3T scanners described increased cortical folding in the occipital and parietal regions of the brain in patients with ASD compared to controls, but that cortical surface area was comparable between the two groups. Both groups showed a decline in gyrification overtime (Wallace et al. 2013).

Many investigators have sought to identify specific regions of interest (ROI), which may be structurally altered in ASD and serve as a disease biomarker. A variety of investigative approaches have been employed. Early trials were labor intensive, requiring manual tracing and measurement of specific regions. Since the advent of VBM, broad surveys of the brain, as conducted in the studies described in Sect. 7.4, allowed for simultaneous comparison across dozens of regions of interest or thousands of specific voxels, aiming to identify areas associated with the greatest differences between patient and control samples. Alternatively, the a priori specification of a hypothesis regarding structural differences in a particular region of interest has also been used, to allow for conservation of statistical power. Additionally, complex statistical approaches using mathematical models and machine learning algorithms have recently been developed to combine the predictive power of both methods.

Dozens of ROI have been implicated through VBM data (summarized in Table 7.1), but even across meta-analyses, the locations of proposed voxel clusters are variable, and show increases and decreases in grey matter density and volume in similar regions. Meta-analytic approaches reveal some agreement regarding volume loss in the putamen (Duerden et al. 2012; Nickl-Jockschat et al. 2012; Yu et al. 2011) and amygdala (Nickl-Jockschat et al. 2012; Via et al. 2011; Yu et al. 2011), and volume gains in the caudate (Duerden et al. 2012; Stanfield et al. 2008; Yu et al. 2011). Throughout the medical literature, the neuro-anatomical locations most often cited in relation to ASD include the cingulate cortex, limbic system, prefrontal cortex , temporal and parietal lobe structures, cerebellum and basal ganglia (see Fig. 7.4) (Cauda et al. 2011; Amaral et al. 2008). These regions will be discussed along with possible functional implications as they pertain to social communication deficits and repetitive behaviors. Regions of interest within white matter tracts will be discussed separately in subsequent sections.

Earlier studies looking at white matter volume in ASD described a pattern of white matter overgrowth in young children but not adolescents compared to typically developing peers (Courchesne et al. 2004). Subsequent meta-analysis across VBM studies looking at white matter specifically described no difference overall in white matter volume, and no difference between child/adolescent and adult age groups, however data did not include very young children (< 6 years). Some regional differences were noted, with increased white matter volume in the right arcuate fasciculus, the left uncinate and inferior fronto-occipital fasciculus, as well as volume losses in the cingulum and corpus callosum (Radua et al. 2011).

Beyond volumetric analysis, diffusion tensor imaging (DTI) is an approach permitting a more detailed analysis of the white matter microstructure of patients with neurodevelopmental disorders such as ASD. DTI uses MRI technology combined with vector algebra, to describe the magnitude and direction of diffusion of water molecules within each individual voxel in the brain. Differences in diffusivity patterns are represented by several specific parameters, and are thought to reflect subcellular structural changes. It is important to understand these measurements in order to interpret the DTI literature in ASD.

Axial diffusivity (AD), for example, quantifies diffusion of water molecules occurring parallel to white matter fibers. Increased AD has been found in diseases involving axonal degeneration; in autism, AD is thought to reflect both the integrity and density of axons . Radial diffusivity (RD) is a measurement of diffusion traveling perpendicular to the white matter fibers. RD is increased in demyelinating diseases, and serves as a surrogate measure of white matter myelination across neurodevelopment. Mean diffusivity (MD) (also known as the apparent diffusion coefficient, ADC) summarizes the total average diffusion occurring in a region, in the absence of any directional gradient (Alexander et al. 2007; see Fig. 7.5).

A summary vector, termed “fractional anisotropy” (FA), describes the overall spherical shape of the other vectors combined. For example, an area where water molecules move freely in all directions would be described as “isotropic.” An isotropic solution would yield in a three-dimensional diffusion vector that appears perfectly spherical, represented by an FA value close to zero. On the other hand, areas where diffusion is restricted in certain directions (e.g. in a white matter tract) are considered “anisotropic.” The resulting three-dimensional vector would be more ellipsoid shaped, represented by an FA value closer to 1.

In neuroimaging of white matter tracts, higher FA is thought to be a sensitive but not specific marker of white matter integrity, including axonal myelination, fiber diameter, and the integrity of cell membranes (Kubicki et al. 2007). In typically developing individuals, FA increases and MD decreases with age, as white matter gradually matures (Faria et al. 2010). The specific pathological correlates of DTI in ASD remain to be fully elucidated, however. From mouse models, induction of demyelination resulted in decreased FA and increased RD, while reduced axonal caliber correlated with lower AD values (Harsan et al. 2006). In mouse models of autism, impaired sociability of the BALB/CJ was associated with lower FA and higher MD on DTI. On pathology, the BALB/CJ mouse had more scattered, less densely packed and more poorly myelinated white matter fibers. The authors attributed higher MD values to the reduced axonal density in affected mice, while more poorly myelinated white matter fibers were thought to result in lower FA values (Kumar et al. 2012).

Dozens of articles have been published looking at DTI in patients with ASD compared to controls across age groups. Across the whole brain, most studies observe reduced FA and decreased MD [reviewed by (Travers et al. 2012)]. A recent systematic review and meta-analysis combines data from 14 DTI studies using regions of interest methods (with mean ages of participants between 3 and 32 years). Their results highlight some consistency in findings and are compared to volumetric data in Table 7.2. Fractional anisotropy seems to be lower in corpus callosum, left uncinate fasciculus and left superior longitudinal fasciculus in affected patients vs. controls, suggestive of reduced white matter integrity and possibly reduced myelination in these regions. Fewer studies have published data on MD, but significant increases emerge in the corpus callosum and the superior longitudinal fasciculi as well, which may suggest reduced axonal density of the fibers in these tracts (Aoki et al. 2013).

These findings may have functional implications related to the core features of ASD. The superior longitudinal fasciculus, for example, is a large white matter bundle coursing in a ‘front to back’ direction across the brain, relaying information between the frontal, temporal, parietal and occipital lobes. It is divided into several segments, which help to coordinate and relay information needed for movement, spatial navigation, occulomotor function, and importantly, language. Increased MD in the superior longitudinal fasciculus correlated with severity of language impairment in affected children and adolescents (Nagae et al. 2012), and lower FA scores in this region correlated with greater communication impairment on the ADOS and ADI-R (Poustka et al. 2012).

The uncinate fasciculus is an association fiber, which connects structures in the prefrontal cortex (specifically the orbital frontal cortex) with anterior temporal lobe structures as well as the amygdala. These regions are thought to be important for emotional regulation and processing, decision-making, and reward valuation and anticipation. In a study of behavioral interventions, increased adherence to therapy and improved clinical outcomes correlated with higher FA values in the uncinate fasciculus (Pardini et al. 2012). In infants, higher FA in the uncinate fasciculus at 6 months of age was predictive of greater social communication at 9 months of age (Elison et al. 2013).

The corpus callosum is the largest white matter structure in the brain, connecting the left and right cerebral hemispheres. It is involved in many cerebral functions including integration of sensory and motor information, as well as many higher cognitive tasks such as problem solving, executive function, planning, language, and social cognition and communication (Bellani et al. 2013a). Congenital absence of a corpus callosum (called ‘agenesis of the corpus callosum’) is a condition often associated with developmental delay and motor impairment, although at times can be asymptomatic as well. A screening survey on children with agenesis of the corpus callosum found that 45 % were over the cut-off for autism spectrum disorder (Lau et al. 2013).