Elsevier

NeuroImage

Volume 23, Issue 3, November 2004, Pages 1192-1202
NeuroImage

Automatized clustering and functional geometry of human parietofrontal networks for language, space, and number

https://doi.org/10.1016/j.neuroimage.2004.09.023Get rights and content

Abstract

Human functional MRI studies frequently reveal the joint activation of parietal and of lateral and mesial frontal areas during various cognitive tasks. To analyze the geometrical organization of those networks, we used an automatized clustering algorithm that parcels out sets of areas based on their similar profile of task-related activations or deactivations. This algorithm allowed us to reanalyze published fMRI data (Simon, O., Mangin, J.F., Cohen, L., Le Bihan, D., Dehaene, S., 2002. Topographical layout of hand, eye, calculation, and language-related areas in the human parietal lobe. Neuron 33, 475–487) and to reproduce the previously observed geometrical organization of activations for saccades, attention, grasping, pointing, calculation, and language processing in the parietal lobe. Further, we show that this organization extends to lateral and mesial prefrontal regions. Relative to the parietal lobe, the prefrontal functional geometry is characterized by a partially symmetrical anteroposterior ordering of activations, a decreased representation of effector-specific tasks, and a greater emphasis on higher cognitive functions of attention, higher-order spatial representation, calculation, and language. Anatomically, our results in humans are closely homologous to the known connectivity of parietal and frontal regions in the macaque monkey.

Introduction

Functional magnetic resonance imaging has become an invaluable tool to analyze the anatomical organization of functional areas in the human brain. The organization of human visual areas, in particular, has seen considerable progress thanks both to the existence of previous work in animals (Van Essen et al., 2001) and to the presence of retinotopy as a major organization factor (Hasson et al., 2003). The major principles of geometrical organization of the parietal and frontal lobes, however, have proven more elusive (see, however, Astafiev et al., 2003, Buccino et al., 2001, Culham and Kanwisher, 2001, Koechlin et al., 2003). The parietal cortex currently appears as a mosaic of distinct specialized areas involved in a variety of visuospatial tasks including finger pointing, grasping, and eye or attention orienting (Simon et al., 2002). Likewise, the human frontal lobes are implied in a variety of cognitive processes such as motor programming, working memory, memory retrieval, executive control process, attentional selection, conflict resolution, and decision making, whose neural substrates are only beginning to be delineated. Furthermore, parietal and frontal activations often overlap in a variety of tasks, either because those regions are organized in a much less rigid fashion than posterior sensory-motor areas (Dehaene et al., 1998, Duncan and Owen, 2000), or because the tasks often share abstract components such as attention orienting or working memory. Interestingly, frontoparietal networks have also been found in resting state conditions where the level of attention load is measured with EEG (Laufs et al., 2003).

As a first step toward clarifying the anatomical organization of the parietal lobe, in a previous fMRI study, we scanned the same subjects during the performance of four visuospatial tasks (grasping, pointing, saccades, and visuospatial attention) and two cognitive tasks (calculation and phonemic detection) (Simon et al., 2002). Those tasks were selected because they were known to involve the parietal lobe. We observed a systematic anterior-to-posterior organization of activations in the parietal lobe, with subregions associated with grasping only, grasping and pointing, all visuomotor tasks, attention and saccades, and saccades only. We also demonstrated that the higher cognitive tasks of calculation and language lead to distinct activations of the inferior parietal lobule, within the intraparietal sulcus, that occupied reproducible geometrical locations relative to the above sensorimotor map. This overall organization was comparable to the known anatomy of macaque monkey areas and suggested putative human homologs of the monkey areas AIP, MIP, V6A, and LIP.

The aim of the present study is to extend those results by further characterizing, using fMRI during the same six tasks, the functional organization of the human frontal lobe and the geometry of parietofrontal networks. In our previous study, we acquired scans covering the whole of the parietal and frontal regions (though not the lower temporal and occipital lobes). However, our analysis was voluntarily limited to a mask covering the parietal lobe. This was done in part for practical reasons, as we had to manually explore the 31 possible intersections of active areas in each of our six tasks (26 − 1). The restriction to the parietal lobe also allowed us to limit the risk of false positives related to the large number of possible task intersections that we analyzed.

To extend this initial analysis to both parietal and frontal networks, we used a clustering method, adapted to functional images by three of us (F.K., G.F., and J.B.P.), that permits the automatized detection of sets of areas with a common profile of functional activation. The main idea is that regions or voxels can be described most adequately by its response profile across a range of different task. Our method thus uses the similarity of functional response profiles measured in distant brain regions to provide an automated classification of voxels into “classes” forming putative brain-scale networks. The K-means clustering algorithm that we use is one of the simplest and most popular classification techniques, and its description can be found in many textbooks. Thus, our main objective is not to describe this method, but rather to demonstrate how it allows extension of our previous results to the frontal lobe, thus yielding a better description of functional parietofrontal networks. Related ideas have been pursued in a few recent papers (Bokde et al., 2001, Passingham et al., 2002, Shinomoto et al., 2002), but practical examples are still rare.

In a nutshell, our method starts from the six three-dimensional images that characterize activations in each of the six tasks of interest. Each cerebral voxel is therefore characterized by a profile of six statistical values (Student's t tests), each of which defines the significance of activation in a given task relative to its control. Limiting the analysis only to voxels active in at least one task, the method groups together voxels that exhibit a similar profile of activation across the six tasks. The grouping is done regardless of whether the voxels are close or distant in cortical space. In the end, the voxels end up being grouped into functionally homogeneous sets of area (technically called “classes” in this paper), each of which is characterized by a maximally homogenous response to different tasks. Crucially, the analysis proceeds without ever specifying which profiles of activation are expected. Thus, the method is able to discover areas specifically activated by a single task, just as easily as areas activated by the intersection of several tasks.

Application of this method to our previous data set allowed us to isolate eight different functional networks, most of which have a parietal, a lateral frontal, and a mesial component. We demonstrate that the parietal components tightly reproduce our earlier findings with a manual method. The availability of the frontal components of these networks allows us to better understand the pathways linking the frontal and parietal lobes, their global geometry, and their possible relation to macaque monkey circuits.

Section snippets

Stimuli and tasks

Examples of the stimuli and tasks are presented in Fig. 1. Details have been published elsewhere (Simon et al., 2002). Briefly, cerebral activation was studied in six conditions: saccades, pointing, visuospatial attention, grasping, phoneme detection, and subtraction.

  • In the saccade task, subjects moved their eyes toward a filled white square that appeared at random locations on a peripheral circle; the control task was central fixation.

  • The pointing task was similar except that the subject

Results

Fig. 2 shows three-dimensional views of each of the eight classes as well as their typical profile of response to the six tasks. In Fig. 3, the eight classes are presented within a single brain volume; axial slices and three-dimensional views provide information about the geometrical relations between the activated voxels. Table 1 summarizes the peak locations and coordinates of the clusters within each class.

The response profile of each class could be easily summarized, in almost all cases, by

Discussion

We have described a simple method for the automatized analysis of intersections between functional brain activation patterns. The main advantage of this method is that, contrary to our previous work (Simon et al., 2002), it does not require experimenter intervention and multiple pairwise comparisons between all the tasks under study. The only manual choices lie in the selection of the statistical threshold and of the number of clusters. The algorithm then automatically detects, across the whole

Conclusions

We have reported a global topographical geometry of functional parietofrontal networks in humans. An important suggestion of the present work is that insertion of human cognitive functions for calculation and language does not deviate from the functional topography of sensorimotor activations, but follows similar rules of topographic parietofrontal projection and symmetry around the central sulcus.

References (45)

  • R.I. Schubotz et al.

    Functional–anatomical concepts of human premotor cortex: evidence from fMRI and PET studies

    NeuroImage

    (2003)
  • S. Shinomoto et al.

    New classification scheme of cortical sites with the neuronal spiking characteristics

    Neural Netw.

    (2002)
  • O. Simon et al.

    Topographical layout of hand, eye, calculation, and language-related areas in the human parietal lobe

    Neuron

    (2002)
  • D.C. Van Essen et al.

    Mapping visual cortex in monkeys and humans using surface-based atlases

    Vision Res.

    (2001)
  • E. Wojciulik et al.

    The generality of parietal involvement in visual attention

    Neuron

    (1999)
  • S.V. Astafiev et al.

    Functional organization of human intraparietal and frontal cortex for attending, looking, and pointing

    J. Neurosci.

    (2003)
  • F. Binkofski et al.

    A fronto-parietal circuit for object manipulation in man: evidence from an fMRI-study

    Eur. J. Neurosci.

    (1999)
  • G. Buccino et al.

    Action observation activates premotor and parietal areas in a somatotopic manner: an fMRI study

    Eur. J. Neurosci.

    (2001)
  • J. Cohen

    Statistical Power Analysis for the Behavioral Sciences

    (1988)
  • S. Dehaene et al.

    Towards an anatomical and functional model of number processing

    Math. Cogn.

    (1995)
  • S. Dehaene et al.

    A neuronal model of a global workspace in effortful cognitive tasks

    Proc. Natl. Acad. Sci. U. S. A.

    (1998)
  • S. Dehaene et al.

    Sources of mathematical thinking: behavioral and brain-imaging evidence

    Science

    (1999)
  • Cited by (107)

    • Processing symbolic and non-symbolic proportions: Domain-specific numerical and domain-general processes in intraparietal cortex

      2019, Brain Research
      Citation Excerpt :

      Furthermore, shared activation for grasping and pointing was observed in left IPS extending into superior parietal lobule (SPL) and postcentral sulcus (Simon et al., 2002). These motor tasks (i.e., grasping and pointing) also showed joint activation with visuospatial tasks such as saccades and attention orienting in bilateral SPL (Simon et al., 2004, 2002). Interestingly, overlapping activation was also observed with mental calculation, language, and saccades in the left posterior segment of the IPS beneath the angular gyrus (Simon et al., 2002).

    View all citing articles on Scopus
    View full text