Paul Thompson's Research Publications

Quantifying and Correcting for Variable Cortical Morphology in Functional Imaging using a Deformable Probabilistic Brain Atlas

Proceedings of the 3rd International Conference on Functional Mapping of the Human Brain, Copenhagen, Denmark (May 19-23, 1997), NeuroImage 5(4), May 1997

Paul Thompson, David MacDonald, Michael S. Mega, Colin J. Holmes, Alan C. Evans, Arthur W. Toga

Laboratory of Neuro Imaging, Department of Neurology, Division of Brain Mapping, UCLA School of Medicine, Los Angeles, California 90095


Montreal Neurological Institute, McGill University, Montreal, Canada



Analysis of functional data localized at the cortex presents unique challenges and problems for cross-subject and group comparisons, due to the extreme degree of variation in gyral morphology and patterning in human populations. To address these difficulties, we describe the construction of a deformable probabilistic atlas, whose goals are to (1) quantify, analyze and encode variations in gyral and sulcal topography; (2) detect and localize deviations from normal cortical anatomy; (3) compare and integrate 3D functional data from many subjects, in a way which accounts for the large differences in superficial and deep sulcal patterns from one individual to another.

Nested Brain Surfaces after Registration into Standardized Talairach Stereotaxic Space. 3D frontal views are shown of two normal subjects' brain surfaces, after digital transformation into Talairach stereotaxic space. Our basis for comparison among cortical shapes is a deformation map, which represents the local transformation required to deform a high-resolution surface mesh, representing the cortical surface in one subject, onto a target surface from another individual. Methods are introduced in this paper that allow this displacement map to be constructed in a complex, biologically meaningful way. In particular, a network of lobar, sulcal, and cytoarchitectural landmarks are displaced into structural correspondence with their counterparts in a target brain. The resulting displacement map encodes regional differences in sulcal and gyral topography from one subject to another. This procedure is extremely important for quantitative measurements of cortical differences as it overcomes problems caused by wide variations in the sulcal pattern within normal populations. Indicated here (in color) is the magnitude of the local displacement required to deform one cortical surface into correspondence with a target surface from another subject (blue). Differences in regional shape are, therefore, assessed by computing the local deformation that transforms one cortical surface into another, according to strict biological criteria.

Atlas Construction: Quantification of 3D Cortical Variability

3D (384x384x256 resolution) T1-weighted fast SPGR (spoiled GRASS) MRI volumes, acquired from 10 normal subjects (mean age: 72 yrs.) and 10 age-matched subjects with clinically-determined Alzheimer's Disease (AD), were digitally transformed into Talairach stereotaxic space. Residual variability of cortical features, even after stereotaxic transformation, was considerable. A high-resolution surface representation of the cerebral cortex was automatically extracted from each scan [1], and connected systems of parametric meshes [2,3] were used to model critical lobar, functional and cytoarchitectural boundaries in 3 dimensions. These included: the parieto-occipital, calcarine, central, cingulate, callosal, superior frontal and rhinal sulci and Sylvian fissure in both brain hemispheres. Differences in cortical topography were recorded in the form of complex 3D fluid transformations in stereotaxic space, which bring both the cortical surface and sulcal patterns of different subjects into register [4]. Anatomical information on the sulcal and gyral pattern was exploited, in order to accurately transform systems of connected tissue interfaces into structural correspondence with their counterparts in a target brain. These surface-to-surface maps were extended to a full 3D volumetric transformation, using a high-dimensional warping algorithm [5]. The resulting family of volumetric warps was used to derive statistical measures of local anatomical variation across the cortical surface. Statistical analysis of the deformation fields led to the calculation of a high-resolution Gaussian random vector field, encoding probabilistic information on confidence regions for cortical surface points [6]. Regional patterns of abnormal cortical structure were mapped out and highlighted in new subjects [4]. Sulcal variability maps revealed striking directional trends in the patterns of local anatomic variability. Confidence limits on surface variation increased dramatically towards the posterior sylvian and anterior cingulate cortex, and were systematically greater in the group of subjects with Alzheimer's Disease than in the group of age-matched control subjects.

High-Dimensional Registration of 3D Functional Data

Spatially-detailed transformations, which reconfigure one subject's cortical anatomy into the shape of another, remove subject-specific shape differences. They can therefore transfer co-registered functional data from many different subjects onto a single neuroanatomic template or atlas. They can also correct for post mortem anatomic change, in studies where precise cortical registration is essential. In one application [7], whole brain cryomacrotome sections were derived post mortem from a subject with AD, and were stained to produce 3D histologic maps of synapse density, neurofibrillary tangles and beta-amyloid distribution. Images of these neurochemical maps were elastically warped (using networks of cortical features to drive the high-dimensional transformation) into the anatomical configuration of a pre-mortem 3D MRI and co-registered FDG-PET volume, for correlation and analysis. Cortical constraints on the transformation enabled precise registration of neurochemical maps with cortical tissue producing the 3D PET signal in vivo. Applications of the high-dimensional transformation and probabilistic algorithms presented here include the accurate transfer of multi-subject 3D functional, vascular and histologic maps onto a single anatomic template for subsequent comparison and integration [7], analysis of cortical variation in human populations [2], and detection and mapping of structural abnormalities in the cerebral cortex [4].


[1]. MacDonald D et al. (1994) Proc SPIE 2359, 160-169; [2]-[6]: Thompson PM et al.: [2]. Journal of Neuroscience (1996) 16(13):4261-4274; [3]. NeuroImage (1996) 3(1):19-34; [4]. J. Comp. Assist. Tomography, [in press]; [5] IEEE Transactions on Medical Imaging (1996) 15(4):1-16; [6]. Proc. Visualization in Biomedical Computing (1996) 4:383-392; [7]. Mega MS et al., NeuroImage, [in press]. Grant Support. (to P.T.): United States Information Agency Grant No. G-1-00001, Washington DC; Howard Hughes Medical Institute, Bethesda, MD; and U.S.-U.K. Fulbright Commission, London. (to A.W.T.): NLM (LM/MH05639), NSF (BIR93-22434), NCRR (RR05956), NIMH/NIDA (P20 MH/DA52176).

Related Publications

Contact Information

  • Mail:

    Paul Thompson
    73-360 Brain Research Institute
    UCLA Medical Center
    10833 Le Conte Avenue
    Westwood, Los Angeles CA 90095-1761, USA.

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