Paul Thompson's Research Publications

Automated Analysis of Structural MRI Data

1Paul Thompson PhD, 2Judith L. Rapoport MD, 3Tyrone D. Cannon PhD, 1Arthur W. Toga PhD

1Laboratory of Neuro Imaging, Dept. Neurology, UCLA School of Medicine
2Child Psychiatry Branch, National Institute of Mental Health, NIH
3Depts. Psychology, Psychiatry and Human Genetics, UCLA

Chapter 5 in: Brain Imaging in Schizophrenia [.pdf 251KB; Without Figures]

Editors: Stephen Lawrie, Eve C. Johnstone, Daniel Weinberger,
Oxford University Press, 2003

[Click on the Image for a Larger Version]

Methods for Analyzing MRI Data. This schematic illustrates six major types of analysis of structural images, showing some of the main types of data used in each case. Voxel based morphometry (VBM; top left panels) compares anatomy voxel by voxel to find voxels (shown here in blue) where the tissue classification (gray, white matter, CSF) depends on diagnosis or other factors. Results are typically plotted in stereotaxic space (lower panel), and their significance is assessed using random field or permutation methods (see text for details). Deformation based morphometry (DBM) can be used to analyze shape differences in the cortex, or brain asymmetries (colored sulci: red colors show regions of greatest asymmetry). Tensor based morphometry (TBM) uses 3D warping fields with millions of degrees of freedom (top) to recover and study local shape differences in anatomy across subjects or over time (red colors indicate growth rates in the corpus callosum of a young child). Other methods focus on structures such as the cerebral cortex, which can be flattened to assist the analysis (bottom left), or the lateral ventricles (‘Shape Modeling’). If anatomic structures are represented as parametric surface meshes (Thompson et al., 2000), their shapes can be compared, their variability can be visualized, and they can be used to show where gray matter is lost (e.g., in Alzheimer’s disease: bottom left panel, red colors denote greatest gray matter loss in the limbic and entorhinal areas). Fine scale anatomical parcellation (lower right) can be used to compare structure volumes across groups, or to create hand labeled templates that can be automatically warped onto new MRI brain datasets. This can create regions of interest in which subsequent analyses are performed. [VBM data courtesy of Elizabeth Sowell, Ph.D. (adapted from Sowell et al., 2000), and parcellation data courtesy of Jacopo Annese, Ph.D., UCLA Laboratory of Neuro Imaging].


In this chapter, we review computational methods to detect structural differences in the brain. We describe the mathematical concepts underlying several common approaches, including voxel-based morphometry, deformation morphometry, and tensor-based morphometry, as well as shape modeling, and traditional volumetrics (see Figure). We also cover hybrid approaches (e.g., cortical pattern matching) which analyze shape and tissue distribution in the same analysis. Each technique is optimized to detect specific features, and has its own strengths and limitations. We highlight illustrative clinical findings from studies of development, dementia and schizophrenia. We also describe newer techniques that map patterns of brain change over time (e.g., to study disease progression or medication effects). Finally we describe how imaging statistics can be expanded to assess genetic effects on brain structure, in family, twin or allele-based designs.

Related Publications

  • (back to main list)

    Contact Information

  • Mail:

    Paul Thompson, Ph.D.
    Assistant Professor of Neurology
    UCLA Lab of Neuro-Imaging and Brain Mapping Division
    Dept. Neurology and Brain Research Institute
    4238 Reed Neurology, UCLA Medical Center
    710 Westwood Plaza
    Westwood, Los Angeles CA 90095-1769, USA.

  • E-mail:
  • Tel: (310)206-2101
  • Fax: (310)206-5518