A myelo-architectonic method for the structural classification of cortical areas
Introduction
The extensive application of MR imaging techniques with sub-millimetric spatial resolution has revived interest in a topographically organized description of the human cortex that transcends images of superficial lobar or sulcal patterns. There are encouraging indications that microscopic cortical anatomy could be acquired through MR noninvasive imaging of the brain (Clark, 1992; Barbier et al., 2002). However, at present, the internal structure of the cortex is examined chiefly through histological analysis of post-mortem specimens (Annese and Toga, 2002). In this respect there has been recent reprisal of classical architectonic investigations (e.g., Annese and Toga, 2002, Roland et al., 1997, Zilles et al., 2002).
Early work demonstrated that the cortical mantle is not a uniform layer of gray matter, but that it is striped by a distinct internal lamination Baillarger, 1840, Gennari, 1782, Vicq d'Azyr, 1786. Moreover, such stratified structural design was shown to vary locally on the surface of the hemispheres Bevan-Lewis, 1878, Meynert, 1872. In fact, systematic histological surveys of the cortical surface culminated in very complex parcellation schemes Brodmann, 1909, Campbell, 1905, Elliot Smith, 1907, Vogt and Vogt, 1919, von Economo and Koskinas, 1925. Two principal histological features, complementary to a certain extent, have shaped classical maps of the human cortex: the size and distribution of neuronal cell bodies and the density and arrangement of myelinated fibers across the depth of the cortex. These patterns are studied histologically at low power magnification in cross sections of the cortex. It is these features that, following specific staining techniques, make the cortex appear unquestionably and ubiquitously laminated (see Figs. 1A and B).
The density of staining in each layer of the cortex can be recorded by algorithms that measure pixel intensity values along cross-sectional lines (traverses) drawn manually (Zilles and Schleicher, 1993) or semiautomatically (Schleicher et al., 1999) from the pial surface to the white matter border. This approach derives from the original photometric studies of cortical myelination by Hopf (1965) in which each cortical region was characterized by the shape of locally sampled intensity profiles.
A crucial issue in the implementation of an observer-independent and automatic architectonic analysis is the outline of the traverses across the cortex, that is, the paths along which pixel intensities will be measured. The cortical ribbon is not homogeneous internally but presents a complex radial and tangential internal structural framework. The effect of curvature on the internal structure of the cortex was predicted by Bok (1959) on theoretical grounds and studied empirically by Smart and McSherry, 1986a, Smart and McSherry, 1986b in the developing brain of the ferret (Mustelidae). Bok (1959) subdivided the cortical volume into a regular grid of constant dimensions, Smart and McSherry (1986b) drew their cortical grid by tracing the intersection of horizontal laminae with radial lines that followed glial fibers and cellular columns. Both studies indicated that the cortical layers are subject to considerable geometrical distortion during gyrogenesis as indicated by the change in the curvature and direction of radial lines. These studies suggest that a model of the internal structure of the cortex is necessary to generate correct intensity profiles and hence to obtain a meaningful characterization of architectonic areas. In fact, if the transverses are not consistently perpendicular to the cortical layers, or if they intersect, the resulting intensity profiles will be distorted and will contain high levels of noise.
To model the curvature of intracortical layers, we adopted the heat conduction differential equation that Jones et al. (Jones et al., 2000) applied to MRI brain volumes to measure cortical thickness. In our case, working with 2-D histological images, the solution of the Laplace equation generates a series of nested equipotential contours (n intermediate cortical layers) between the pial surface and the gray and white matter border. This computational framework produces correct field lines along which the density of staining can be measured. These lines satisfy two important conditions: they do not intersect and they are orthogonal to each inner cortical layer.
Cortical architectonic parcellation requires the definition a priori of the structural elements that will provide anatomical information and that will gauge the comparison between different areas. Our classification is based on architectonic features of myelination: those structural patterns produced by intracortical fiber systems. In this respect, myelin staining reveals evident local structural variations in the cortex produced by changes in both the laminar and radial disposition of fibers. Horizontal and vertical fiber bundles can be arranged in several different ways across the width of the cortex. In fact, one limitation of classical and customary studies of cortical architecture is that the discrimination between these structural patterns has been subjective or based on very complex and arbitrary schemes of interpretation (Vogt, 1910).
An independent algorithm for the unsupervised classification of myelo-architectonic types would need to be instructed on what to look for along the cortical ribbon. A basic target could be the six-layered isocortical scheme, with two main horizontal bands (Vogt, 1910; see Fig. 1C); the profile arrays produced from each sample would then be classified according to their similarity to the target. In this case, the architectonic classification would be driven by a strong initial hypothesis on cortical structure. However, since cortical myelo-architecture shows extreme—and not well documented—variations, we did not gauge our classification with respect to a single basic architecture. Instead, we propose an automated scheme to assess profile similarity based on significant components derived from wavelet analysis.
Mapping the topography of cortical areas and thus understanding the relationship between structure and function in the cerebral cortex are both a classification and a localization problem. The definition of unambiguous architectonic templates is the prerequisite for topographic histological surveys and hence for comparative studies and for the generalization of architectonic maps to a population. Therefore, we compute averages of profiles that belong to the same class in order to construct synthetic structural models that are easily recognizable and that can be localized topographically on the cortical mantle.
Section snippets
Preparation of histological material
The material presented in this publication has been sampled from whole serial sections (n = 20) belonging to four different human brain specimens. Specimens were removed from the skull and fixed by immersion in 4% phosphate-buffered paraformaldehyde. The tissue was embedded, but not infiltrated in 7% gelatin creating a coating thick enough (0.5–1 cm) to minimize distortion of the gyri and to hold in place parts of the section that would otherwise naturally dissociate.
Specimens were cut frozen
Results
Intensity profiles presented high-frequency fluctuations due to the texture of myelination in the original images. By smoothing individual profiles in the wavelet domain, we obtained a compact representation of the data where the signal was represented by only few significant coefficients that were well localized in the time-frequency domain. For example, in Fig. 3, only 5% of the coefficients were sufficient to reconstruct the original profile with good accuracy.
The border between the primary
Cortical topology
We presented an automatic method to describe the structural topography of the cortex based on the density and disposition of myelinated fibers. It was clear from the start that the high degree of curvature of the mantle was a major obstacle to the characterization of local cortical architecture. To avoid this potential problem, there have been efforts to chart the entire cortical surface by orthogonal slices producing hundreds of blocks sampled ad hoc in regions where the surface of the gyri
Acknowledgements
This work was supported by research grants from the National Library of Medicine (5 R01 LM05639), the NIH–National Center for Research Resources (P41 RR13642) and the NIH–National Institute of Mental Health (P20 MH65166 and P01 MH52176). J.A. wishes to thank Dr. Michael S. Gazzaniga (Center for Cognitive Neuroscience, Dartmouth College) and Dr. William F. Hickey (Department of Pathology, Dartmouth Medical School) for their scientific and academic support. A.P. was also supported by the INRIA
References (57)
- et al.
Postmortem anatomy
Representation in the cerebral cortex and its areal lamination patterns
Observer-independent method for microstructural parcellation of cerebral cortex: a quantitative approach to cytoarchitectonics
NeuroImage
(1999)A stereological approach to human cortical architecture: identification and delineation of cortical areas
J. Chem. Neuroanat.
(2000)- et al.
A robust transcortical profile scanner for generating 2-d traverses in histological sections of richly curved cortical courses
NeuroImage
(2002) - et al.
Cyto- and myeloarchitecture of human visual cortex and the periodical GABAA receptor distribution
Computer-Assisted Reasoning in Cluster Analysis
(1995)Recherces sur la structure de la couche corticale des circonvolutions du cerveau
Mém. Acad. R. Méd.
(1840)- et al.
Imaging cortical anatomy by high-resolution MR at 3.0T: detection of the stripe of Gennari in visual area 17
Magn. Reson. Med.
(2002)