Finger representations in human primary somatosensory cortex as revealed by high-resolution functional MRI of tactile stimulation

Abstract

Fine-scale functional organization of the finger areas in the human primary somatosensory cortex was investigated by high-resolution BOLD MRI at 3 T using a multi-echo FLASH sequence with a voxel size of 2 mm³. In six subjects independent tactile stimulation of the distal phalanx of the fingers of the right hand resulted in small circumscribed and barely overlapping activations precisely located along the posterior wall of the central sulcus. Three out of six subjects showed a complete succession of activation sites for all five fingers. The maps also allowed for the identification of individual variations in finger somatotopy. When registered onto the individual high-resolution MRI anatomy and compared with cytoarchitectonical maps, the finger representations were confirmed to lie within Brodmann area 3b as the main input region of the primary somatosensory cortex.

Introduction

Mapping the functional topography of primary sensory areas is essential to unravel the organization of the sensory input at the level of the cerebral cortex. In humans, the first and still valid map of the primary somatosensory cortex (S1) was acquired by Penfield (Penfield and Bouldrey, 1937, Penfield and Rasmussen, 1950). The sensory homunculus, a schematic drawing of the neuronal representations of the body surface, was based on verbal reports of sensations elicited by electrical stimulation of the surface of the postcentral gyrus of patients undergoing brain surgery. Such maps not only illustrated the topographical layout of specific body parts, but also unraveled pronounced disproportions for some neuronal representations that reflect differences in peripheral innervation density. Prominent examples are the large areas of the face and hand, the latter having discrete representations for each of the five fingers.

The advantage of a relatively good spatial resolution of cortical electrical stimulation is clearly opposed by its invasiveness and limited applicability. The development of noninvasive imaging techniques such as magnetoencephalography offered spatial resolution in the range of several millimeters, so that a differentiation of individual finger representations seemed possible (Baumgartner et al., 1991, Nakamura et al., 1998). Ultimately, however, the advent of functional magnetic resonance imaging (fMRI) promised a sufficiently high spatial resolution for a reliable fine-scale somatotopic mapping of neuronal finger representations in S1.

So far, a number of studies demonstrated blood oxygen level dependent (BOLD) MRI responses in human S1 that are elicited by a sensory stimulation of fingers, for example see Sakai et al. (1995), Lin et al. (1996), Gelnar et al. (1998), Kurth et al. (1998), Hansson and Brismar (1999), Stippich et al. (1999), Francis et al. (2000). The studies are diverse concerning the type of stimulation, the number of fingers stimulated, and the applied spatial resolution. Only two reports mapped the entire finger area in S1 (Maldjian et al., 1999, Kurth et al., 2000) and suggested that the layout of BOLD MRI activations matches the Penfield homunculus — with the little finger at the most medial position followed by the other fingers along the central sulcus in inferior lateral direction. While these results demonstrated BOLD MRI to be a valid tool for mapping the somatosensory system, they also opened the question about the degree of representational detail achievable by using dedicated high-resolution techniques. In fact, Maldjian et al. (1999) and Kurth et al. (2000) applied gradient-echo echo planar imaging (GE-EPI) sequences with voxel sizes of 10.5 mm³ (activations smoothed using a 4 mm × 4 mm × 6 mm Gaussian Kernel) and 13.5–16.1 mm³ (activations smoothed using a minimum cluster size of two contiguous voxels), respectively.

The purpose of this study was to map finger representations in S1 of individual subjects at even higher resolution, that is with a considerably smaller voxel size and without smoothing, and to co-register and compare respective activation maps to probabilistic cytoarchitectonical maps (www.bic.mni.mcgill.ca/cytoarchitectonics) of S1 as reported by the Jülich group (Geyer et al., 1999, Geyer et al., 2000, Eickhoff et al., 2005). Here, an in-plane resolution of 1 mm × 1 mm and section thickness of 2 mm, that is a voxel size of only 2 mm³, could be realized with a recently developed multi-echo FLASH sequence (Voit and Frahm 2005), specifically designed for high to ultra-high-resolution fMRI.

Functional mapping at high resolution seems to be particularly important for S1 because it is only 2 mm wide — half the width of, for example, the primary motor cortex (Fischl and Dale, 2000, Meyer et al., 1996). Thus, to minimize partial volume effects and achieve a reliable localization of activations in cortical gray matter, the in-plane dimensions of an image voxel should be clearly below 2 mm. This is equally important in view of the immediate vicinity of S1 to the primary motor cortex (M1), located just across the central sulcus. In addition, the low sensitivity of the high-resolution multi-echo FLASH sequence to susceptibility differences yields almost distortion-free T2^⁎-weighted functional images that may be directly superimposed onto the corresponding high-resolution anatomical scans without the need for registration. Again, a reliable attribution of activations to either of the two cortices seems possible only at suitable resolution and without possible registration inaccuracies.

On the other hand, high-resolution fMRI poses considerable challenges such as a limited volume coverage (per unit measuring time) and a reduction of the signal-to-noise ratio (SNR). For an axial orientation, the spatial ordering of finger representations along the central sulcus can only be covered in several sections spanning a 15–25 mm volume (Maldjian et al., 1999, Kurth et al., 2000). Therefore, the present high-resolution recordings were guided by low-resolution functional maps in conjunction with a targeted double-oblique slice orientation (Kleinschmidt et al., 1997, Dechent and Frahm, 2003). The approach focuses on the hand area of the left central sulcus and attempts to cover the BOLD MRI activations of all five fingers within a maximum of three sections of 2 mm thickness.

The SNR decline in high-resolution fMRI is a linear function of the reduced voxel size (Edelstein et al., 1986). To compensate for the concomitantly reduced detectability of weak activations, the number of stimulation cycles was increased and finger stimulation was accomplished with a specially designed piezo-electric Braille module which emerges as a novel technique for efficient tactile stimulation with considerably higher selectivity than electrical stimulation resulting in more focal pattern of activation (Zhang et al., 2007). Moreover, the choice of a spatial resolution at half the width of S1 helped to reduce partial volume effects with non-activated voxels and therefore compensated for at least part of the SNR reduction in terms of fMRI contrast-to-noise. And finally, the application of a two-step thresholding technique for data analysis (Baudewig et al., 2003) improved the identification of activations with both high specificity and sensitivity and without the need for spatial smoothing which was generally avoided to not sacrifice the acquired resolution.