As shown in Table 7.1, the most common MR sequences used to characterize brain structure are T1- and T2-weighted images, DTI and MTR. We will review these in turn (for a more detailed overview of MR sequences, see Roberts and Mikulis 2007).

Contrast in structural T1- and T2-weighted MR images is based on local differences in proton density⁶ and on the following two relaxation times: (1) longitudinal relaxation time (T1) and (2) transverse relaxation time (T2) (Fig. 7.4).

Thus, longitudinal (T1) relaxation time represents an exponential recovery of the total magnetization over time. In a complex manner, it depends on the local structural pattern (lattice)⁷ assumed by the hydrogen nuclei; in the brain, the more structured the tissue, the shorter T1. Transverse (T2) reflects the local “dephasing” rate for precessing nuclei⁸ due to local magnetic-field non-homogeneities, with a concomitant decay of the total MR signal; in the brain, the more structured the tissues, the more rapid the dephasing and hence, the shorter the T2.

Quantitative measurements of T1 and T2 relaxation times are not taken frequently, however (see Deoni et al. 2008 for a method suitable for a rapid voxel-wise measurement of T1 and T2 times). In the majority of MR studies, some combinations of T1- and/or T2-weighted images are acquired instead (Text Box 7.1).

While proton density reflects the amount of signal-emitting nuclei (i.e. protons) present in the tissue, relaxation times, and therefore tissue “brightness” on T1- and T2-weighted images, depend on a variety of biological and structural properties of the brain tissue. Water content is one of the most important influences on T1 in the brain: the more water there is in a given tissue compartment, the longer T1—and the lower the signal on a typical T1-weighted image—will be in that compartment. In the adult brain, T1 is the longest in cerebrospinal fluid (CSF), intermediate in GM and the shortest in WM. Lipid content in WM may also influence T1, through magnetic interactions with hydrogen nuclei of the lipids, which are hydrophobic. Iron content is another important influence, primarily by changing local magnetic non-homogeneities: the higher the iron content, the shorter the T2. Finally, the anatomical arrangement of axons may influence the amount of interstitial water and, in turn, T1 values; more tightly bundled axons would have shorter T1 and therefore appear brighter on T1-weighted images.

The introduction of diffusion tensor imaging (DTI) in the mid-1990s opened up new avenues for in vivo studies of white-matter microstructure (Le Bihan and Basser 1995). This imaging technique allows one to estimate several parameters of water diffusion, such as mean diffusivity (MD) and fractional anisotropy (FA), in the live brain. The latter parameter reflects the degree of water-diffusion directionality; voxels containing water that moves predominantly along a single direction have higher FA. In WM, FA is believed to depend on the microstructural features of fibre tracts, including the relative alignment of individual axons, how tightly they are packed (which affects the amount of interstitial water; see above), myelin content and axon calibre. Note that the hypothetical differences in myelination are perhaps too hastily considered as an explanation for (group, age) differences in FA, to the exclusion of other possible factors (see Fig. 7.5). In most DTI studies, the signal is coming from the movement of water in the extracellular space, complicating the interpretations of the underlying neurobiology (reviewed in Paus 2010). An interesting alternative is to measure intracellular movement of a metabolite, such as N-acethyl-aspartate, inside the axon. This can be achieved with so-called diffusion tensor spectroscopy (DTS; Upadhyay et al. 2008). Due to the high time requirements of DTS, however, this technique is not suitable—at present—for large population-based imaging studies.

Finally, magnetization transfer (MT) imaging is another MR technique employed in studies of the structural properties of WM.⁹ Contrast in MT images reflects the interaction between free water and water bound to macromolecules (McGowan 1999). The macromolecules of myelin are the dominant source of the MT signal in WM (Kucharczyk et al. 1994). Post-mortem studies revealed a significant positive correlation between myelin content and MTR (Schmierer et al. 2004, 2008). The fastest way to acquire MT data is to use a dual acquisition, with and without a MT saturation pulse. MTR images are calculated as the percent signal change between the two acquisitions (Pike 1996). Mean MTR values, subsequently, can be summed across all white-matter voxels constituting, for example, the four lobes of the brain.

For imaging brain “activity,”¹⁰ the most common MR parameter to measure is the blood oxygenation-level-dependent (BOLD) signal detected on T₂*-weighted MR images. This signal relies on the fact that neural “activation” is associated with an oversupply of oxygenated (“red”) blood to the activated brain region; consequently, small veins draining the “activated” region contain some of the “unused” oxygenated blood (“red-veins” phenomenon). In other words, the BOLD signal reflects the proportion of oxygenated (“red”) and deoxygenated (“blue”) blood in a given brain region at a given moment. Variations in the oxygenated-to-deoxygenated blood ratio provide an MR contrast. Because the deoxygenated blood contains deoxyhemoglobin, which is paramagnetic, it therefore increases local non-homogeneities. That, in turn, interferes with T2 relaxation (“dephasing”; see Sect. 7.2.1).

Now what do we measure with the BOLD signal? Although there is no dispute about the relationship between local hemodynamics and brain activity, there is still little agreement about the role various neural events play in driving the hemodynamic signal. At least two issues need to be considered when interpreting the functional significance of BOLD responses to a behavioural challenge: (1) the relative importance of neuronal firing versus synaptic activity occurring in the sampled tissue and (2) the relative contributions of excitatory and inhibitory neurotransmission.

Intuitively, a brain region requires more energy and, hence, more blood flow when neurons located in that region increase their firing rate. This notion is also conceptually attractive; if true, one would be able to take findings obtained with functional imaging of the human brain and directly correlate them with those acquired through single-unit recordings in non-human primates. Several experiments suggest, however, that firing rate is not the best predictor of local changes in hemodynamics; synaptic activity might be a better one (Text Box 7.2).

The second issue of interest, namely the relative contribution of excitatory and inhibitory neurotransmission to changes in local hemodynamics, is less clear cut—with one study suggesting that the former drives the signal (Text Box 7.3.).

Overall, it is likely that changes in local hemodynamics reflect a sum of excitatory postsynaptic inputs in the sample of scanned tissue. The firing rate of “output” neurons may be related to local blood flow (Heeger et al. 2000; Rees et al. 2000), but only inasmuch as it is linked in a linear fashion to excitatory postsynaptic input. Inhibitory neurotransmission may lead to decreases in CBF, indirectly, through its presynaptic effects on postsynaptic excitation.

First, T1-weighted images are linearly registered, from the acquisition (“native”) space to standardized stereotaxic space—such as the average ICBM/MNI-152 atlas (aligned with Talairach and Tournoux space; Evans et al. 1993). The next step involves tissue classification into GM, WM and CSF. A neural-network classifier is trained, using priors¹² indicating locations that are highly likely to contain the three types of tissue (Zijdenbos et al. 2002; Cocosco et al. 2003). The tissue classification step yields three sets of binary 3D images (i.e. GM, WM and CSF). Each of the binary images can be smoothed to generate probabilistic “density” images. These maps are then used in voxel-wise analyses of age- or group-related differences in GM or WM density (see Ashburner and Friston 2000, for a methodological overview).

A deformation field—computed while aligning (registering) a population member to the target¹³—models the extent to which the anatomy of the individual differs from that of the population. Consequently, the set of all deformation fields captures the overall population variability. By comparing inter-population to intrapopulation deformations, we can identify and quantify anatomical differences in local shape between populations.¹⁴ Unlike standard volumetric approaches, this deformation-based technique can detect differences in the morphology of brain regions that do not necessarily coincide with manually defined structures. It is therefore free from a priori considerations about anatomical nomenclature and arbitrary granularity of an atlas. In particular, it can detect changes at the substructure level or across groups of structures.

Other voxel-wise morphological information—namely differences in the local volume—can be estimated from the Jacobian¹⁵ of the deformation field computed between the population members and the population average (Thompson et al. 2000; Chung et al. 2001).¹⁶

In order to estimate volumes of different brain regions (e.g. frontal lobes and the hippocampus), we can take advantage of the nonlinear alignment (registration) of the individual’s T1-weighted images with a labelled template brain; the template contains information about the location of distinct brain regions defined a priori by an expert or generated as region-specific probabilistic maps. Then, we can compute regional volumes in an automatic fashion. Information about anatomical boundaries from the template brain can be back-projected onto each participant’s brain, in native space, and intersected with the tissue classification map. In this way, we can start—for example—by counting the number of grey, white and CSF voxels per lobe (Collins et al. 1994, 1995). Or we can measure the volume of specific brain regions, such as the hippocampus and amygdala (e.g. Chupin et al. 2007; Toro et al. 2009) or the corpus callosum (Nawaz-Khan et al. unpublished observation). Note that it is important to validate automatic segmentations of brain regions against the manual expert-driven segmentation (see Fig. 7.8).

FreeSurfer is the most common suite of automated tools used for reconstruction of the brain’s cortical surface (Fischl and Dale 2000). FreeSurfer segments the cerebral cortex and the WM and then computes triangular meshes that recover the geometry and the topology of the pial surface and the grey/white interface of the left and right hemispheres. The way the local cortical thickness is measured is based on the difference between the position of equivalent vertices in the pial and grey/white surfaces. A correspondence between the cortical surfaces across individuals is established using a nonlinear alignment of the principal folds (sulci) in each individual’s brain with those in the average brain (Fischl et al. 1999).

Approximately two-thirds of the cerebral cortex are buried in folds. To estimate the degree of cortical folding, one can simply compare the total area of the cortical surface to the area of its external surface. A more regional approach has been to compute the ratio between the pial contour and the outer contour in successive coronal sections (Zilles et al. 1988), which allows one to study rostro-caudal variations in folding. We extended these ideas to obtain a local (vertex-based) estimate of the degree of cortical folding; for every point x on the cortical surface, we measure the area contained in a small sphere centred at x (see Fig. 7.9; Toro et al. 2008a). If the brain were lissencephalic, the area inside the sphere would be approximately that of the disc. We estimated the local degree of folding through the surface ratio. The sphere has to be sufficiently large to encompass a few folds but small enough to make the approximation of the lissencephalic area reasonable.¹⁷

The measures of cortical thickness and surface ratio can be compared vertex by vertex (i.e. in the same way voxel-wise analyses of GM/WM densities and deformation fields are carried out; see Sects. 7.3.1 and 7.3.2). Alternatively, global and regional measures can be obtained by calculating mean values across all vertices contained within a particular region of interest (ROI); various anatomical parcellation schemes are available in FreeSurfer, from individual lobes to specific cortical regions (e.g. Destrieux et al. 2010).

Values of FA can be calculated using various tools, such as FDT (http://www.fmrib.ox.ac.uk/fsl) or ExploreDTI (http://www.exploredti.com). Diffusion-weighted images are registered with T1-weighted images and, in turn, with an atlas so that FA/MD values can be calculated for various anatomically defined compartments. An alternative approach, a so-called tract-based spatial statistics (TBSS; Smith et al. 2006), is used to test for local variations in FA/MD in a voxel-wise fashion. Here, individual FA maps are aligned nonlinearly using a method based on free-form deformations and B-splines (Rueckert et al. 1999). The cross-participant mean FA image is calculated and used to generate a white-matter tract “skeleton.” Individual participants’ FA values are warped onto this group skeleton for statistical comparisons by searching perpendicular from the skeleton for maximum FA values.

As mentioned above (Sect. 7.2.1, last paragraph), magnetization transfer ratio (MTR) images are calculated as the percent signal change between the two acquisitions (Pike 1996). The MTR image is registered with the participant’s T1-weighted image and transferred to the standardized stereotaxic space, using the same nonlinear transformation employed to align the T1-weighted image with the (ICBM-152) template. Next, we can calculate the mean MTR values for WM tissue constituting each lobe by merging the WM tissue map with the atlas-based lobar boundaries in native space. For each individual, the mean MTR values are calculated, in native space, across all WM voxels constituting a given lobar volume of WM in that participant.

Maps of MTR and FA values allow us to characterize different properties of WM. By registering these maps with the native T1-weighted image nonlinearly, we can combine information from both modalities. This enables us, for example, to measure the mean MTR and FA values of WM per lobe or fibre tract.