Structural MRI Overview

Types of structural MRI sequences

Different types of MRI sequences provide sensitivity to various aspects of neuroanatomy. T1-weighted images are probably the most common, and are sensitive to tissue class type (e.g., gray matter vs. white matter).

Diffusion-weighted images reflect the directionality of water diffusion in tissue. In an unconstrained environment, water molecules would move randomly. However, in the brain, their travel is constrained by brain tissue, particularly the fatty myelin found along white-matter tracts. Diffusion-weighted images can be used for various types of diffusion tensor imaging, including tractography (identification of white matter tracts) and assessing fractional anisotropy (FA), often considered a reasonable proxy for white matter integrity.

Types of structural MRI analyses for T1-weighted images

From T1-weighted structural scans, it is possible to perform a number of types of analyses to quantify patterns of regional gray matter. One of the more common approaches, and the one we’ll focus on here, is voxel-based morphometry (VBM) ([AshburnerFriston2000], [Mechelli2005]). VBM provides a measure of gray matter volume (or density) based on semiautomated tissue class segmentation. That is, for each tissue class (e.g., gray matter, white matter, cerebral spinal fluid), each voxel is assigned a probability of belonging to that tissue class. These probabilities are proportional to the volume of, for example, gray matter, taking into account the voxel size. If the voxel size is 2 mm x 2 mm x 2mm, then one voxel is 8 mm^3. If that voxel is entirely composed of gray matter, it’s probability is 1, and this value would indicate 8 mm^3 of gray matter. Values less than 1 indicate a smaller volume. For example, if a voxel contained equal parts gray matter and white matter, it’s intensity would be slightly brighter (reflecting that in a T1-weighted image, white matter gives a stronger signal), and might have a probability of 0.5 for gray matter and 0.5 for white matter. In this example, that would reflect 4 mm^3 of gray matter. Popular software packages for performing VBM analyses include SPM, Christian Gaser’s VBM toolbox for SPM (, and FSL.

Another analysis approach is to use a tissue class segmentation routine to define a boundary between tissue classes, essentially drawing a line between, say, gray matter and white matter. This enables the construction of cortical surface models and analysis of surface-based cortical thickness ([Dale1999], [Fischl1999a], [Fischl1999b]). Because they produce explicit models of macroanatomical landmarks, surface-based analyses may result in more accurate registration across subjects, assuming the surface models are accurate. Subcortical structures are generally not included in surface models, although they can be analyzed separately. FreeSurfer performs surface-based cortical thickness analyses.

Finally, it is also possible to generate a voxel-based cortical thickness measure ([Hutton2008], [Das2009]), which may be thought of as a hybrid of the above two approaches. These approaches generally rely on a traditional volume-based alignment across subjects, but provide thickness (rather than volume) values for each voxel.

These approaches rely on different assumptions regarding neuroanatomy and image registration, and may be more sensitive to different sorts of changes in brain anatomy.

Basic steps in a VBM analysis

  1. Perform an initial check of the data to look for visual artifacts or outliers (see Structural MRI quality assessment).
  2. Reorient the structural image to be in generally close alignment with template space (origin aligned and oriented similarly).
  3. Segment the image into different tissue classes (e.g., gray matter, white matter, CSF).
  4. Inspect the segmented images for errors.
  5. Warp the different tissue class images into a common space (i.e., registration or spatial normalization).
  6. Smooth the images.
  7. Perform statistical analyses on the smoothed, normalized gray matter images.