One major methodological shortcoming of OI is spatial resolution. Typically in the range of few centimeters, topographical assignment of the activations is hampered by a sparse spatial sampling over the scalp. In the most commonly used next-neighbor approach, the individual measurement channels are defined by a pair of light-emitting and light-detecting probe separated by an inter-optode distance (IOD) of 2–4cm. These discrete measurement points are then interpolated into a topographic map of functional activation, which is superimposed onto the brain area covered by the probe array. To enhance the spatial resolution a multi-distance approach using several sources and detectors has been proposed. Since light attenuation of multiple neighbors with different IODs is assessed, this approach yields measurements of overlapping photon paths with different sampled target depths (Barbour et al., 2001; Boas et al., 2004). The differential sensitivity of the volumes interrogated by light, leads to a rough depth discrimination and facilitates the separation of signals from superficial and deeper layers. Thus sensitivity to the cerebro-cortical signal is augmented. Also such an approach enhances lateral resolution. A pioneering study (Zeff et al., 2007) recently demonstrated the feasibility of retinotopic mapping of the primary visual cortex, based on eccentricity and angular stimulation (see also White and Culver, 2010). The study proves that state-of-the-art technology and advanced analytical tools enhance spatial resolution dramatically in non-invasive optical imaging, opening an entirely new perspective for the application in basic and clinical research.

Here, we extend the use of high-density optical imaging to another sensory system and test whether this technique is sensitive enough to functionally map the somatotopic organization of the somatosensory cortex. The somatosensory system consists of two major cortical representations, the primary (SI) and the secondary somatosensory cortex (SII), both of which have been shown to be somatotopically organized. While SII is located in the deeper part of the inferior parietal cortex, which is not accessible to non-invasive optical techniques, the SI hand area can be regarded an ideal candidate for optical imaging access, due to its superficial location at the posterior wall of the central sulcus. The idea of somatotopy has long been established. Based on the observation that motor convulsion or dysesthesia ‘march’ along typical routes over the body in some simple focal epileptic seizures and that these ‘routes’ are highly reproducible within and between patients, Jackson postulated a somatotopic organization of the somatosensory and motor cortex already in the 19th century (Jackson, 1863). In 1931, cortical somatotopic organization was first experimentally proven in humans by applying electrical stimulation to precentral and postcentral cortices (Foerster, 1931). Finally, the term “somatotopy” was introduced by Penfield (Penfield and Boldrey, 1937). With respect to somatosensation, he applied electrical stimulation to the surface of the postcentral gyrus in patients undergoing brain surgery. Based on the patients’ reports of thereby elicited sensations, the idea of the sensory homunculus was established (Penfield and Boldrey, 1937; Penfield and Rasmussen, 1950). The gross anatomy (head, arm, trunk, and leg) is clearly differentially represented by areas of contralateral sensory (SI) and primary motor (M1) cortices. The representations follow a latero-medial sequence along the precentral and postcentral gyri, respectively. The representational organization within the hand sub-division is more complex and still partly controversial. In SI a basic somatotopic arrangement was found using MEG (Baumgartner et al., 1991; Hari et al., 1993; Nakamura et al., 1998), EEG (Buchner et al., 1995) and fMRI studies differing in type of stimulation, number of fingers stimulated and spatial resolution (Sakai et al., 1995; Wolf et al., 1996; Gelnar et al., 1998; Kurth et al., 1998; Stippich et al., 1999; Francis et al., 2000; Nelson and Chen, 2008; Schweizer et al., 2008; Weibull et al., 2008). The few studies covering the entire SI finger area at a high spatial resolution found a map similar to that described by Penfield. The thumb is represented in the most lateral, anterior, and inferior position, followed by the other fingers ordered along the central sulcus in a superior to medial direction (Maldjian et al., 1999; Kurth et al., 2000; van Westen et al., 2004; Schweizer et al., 2008).

One subject revealed no focal motor activation during finger-tapping; therefore we do not report on the results of this subject.

Figure ​Figure22 shows the activation maps for the remaining four subjects. T-value maps based on HbR-changes were superimposed on individual anatomical images. The four subjects show a maximum hemodynamic response to finger tapping located on the precentral gyrus. In contrast, the hemodynamic response to somatosensory stimulation is located on the postcentral gyrus. Besides this anterior-posterior differentiation, the localization of the activation to first versus fifth finger stimulation differed with respect to their maximal amplitude.

Across all four subjects the activation elicited by stimulation of the fifth finger was located more superior along the central sulcus compared to first finger stimulation. The temporal evolution of the hemodynamic response is illustrated in Figure ​Figure3.3. Mean changes of HbR and HbO across all runs are shown for the 5 s stimulation period and the ensuing 8s of rest. Data for first and fifth finger of subject 2 are shown. Clearly both conditions led to an increase in HbO (red) and a decrease in HbR (blue) with the typical vascular response latency peaking 5–7s after stimulus onset (0s on the time axis). After thresholding of the results at their absolute maxima, maps revealed different locations of the activation patterns of first and fifth finger stimulation. Vibrotactile stimulation of the first finger led to a more inferior and anterior activation pattern compared to the pattern of the fifth finger in a lateral 2D-view of the brain. This differential localization held for both HbO and HbR. Although HbO and HbR showed slight differences with respect to size and temporal dynamics of the activation patterns, both parameters followed typical dynamics of the vascular response with an increased focal oxygenation resulting from a rise of blood volume and flow velocity. The latter overcompensates the oxygen uptake and thus yields an increased washout of HbR. The anatomical structures subjacent to the fiber array confirmed that the hemodynamic changes to first and fifth finger stimulation originated from the postcentral gyrus.

Electrophysiological methods provided the first successful demonstration that the homuncular organization is accessible to non-invasive techniques in healthy adults. MEG (Baumgartner et al., 1991; Hari et al., 1993; Nakamura et al., 1998) and EEG (Buchner et al., 1995) located differential sources in the pericentral region notably also providing first evidence that this representation can be altered by plastic changes in disease (Elbert et al., 1998; Braun et al., 2003). While these latter findings show the high sensitivity of MEG and EEG to subtle changes in source amplitude and orientation, the same sensitivity constitutes a fundamental limitation of electrophysiological imaging related to the inverse problem of all localization approaches. With the advent of blood oxygen level dependence-(BOLD) fMRI, a number of groups used the inherent advantage of MRI supplying a high-resolution structural image on which the functional activation can be individually mapped (Sakai et al., 1995; Lin et al., 1996; Gelnar et al., 1998; Kurth et al., 1998; Stippich et al., 1999; Francis et al., 2000; Nelson and Chen, 2008; Schweizer et al., 2008; Weibull et al., 2008). Some of these studies were able to produce functional maps highly similar to Penfield's invasively assessed maps. In these images stimulation of the thumb results in the most lateral, anterior and inferior position, followed by the other fingers orderly along the central sulcus in a superior to medial direction (Maldjian et al., 1999; Kurth et al., 2000; van Westen et al., 2004; Schweizer et al., 2008).

Our results confirm that sensitivity of optical imaging is sufficient to detect similarly subtle differences in the lateral distribution of activation foci. Sensitivity to a functional activation paradigm across studies is somewhat arbitrary depending on the exact stimulation protocol, individual anatomy, and a number of issues of the analytical tools and assumptions. A key question is how such an approach can be useful for future research. Here, one apparent concern is the rather high exclusion rate of half of our subjects. Three subjects were excluded due to a wrong localization of the pad with respect to cortical anatomy. This is somewhat at odds with the comparatively small inter-individual variance of the central sulcus with reference to the 10–20 system. Using magnetic resonance imaging, Steinmetz et al. (1989) reported a 1.2cm rostral to caudal variation of the central sulcus (16 subjects) when relating cortical anatomy to the grid referenced by the external bony landmarks. This point, determined in sagittal slices 4cm from midline and with respect to the line defined by the vertex point (Cz) and the pre-auricular point, referred to the location of the middle finger representation of the motor and sensory homunculi. Another MR-study reported inter-individual variances of the same order, when cortical anatomy was related to the 10–10 system (Koessler et al., 2009). Across 16 subjects the standard deviations of the C4 electrode position, close to the central sulcus, was below 10mm (SD for C4: x =4.6mm, y=8.3mm, z =6.4mm). It may be possible that our projection of the optical probes and the cortical surface has added to the variance of the pad's location in the excluded subjects. Additionally, the central sulcus was located by a rather simple expert-rating approach using preprocessed representations of individuals’ brain surface. Since the focus in this paper is on the approach rather than an anatomical question, we used this quite conservative inclusion criterion. In clinical applications structural imaging data will be available in most patients, thus information on the individual anatomy and more importantly individual lesions will usually allow for an informed positioning of the probe array. Furthermore powerful novel projection algorithms have been proposed (Okamoto et al., 2004; Custo et al., 2010) for low-density imaging and are currently being developed for high-density tomographic data.

Another subject was excluded since motor performance did not elicit reliable activation. There are two reasons for using a reliable response to finger tapping as an inclusion criterion. Firstly, motor-related activation has been shown to be much more robust not only in optical (Obrig et al., 1996) but also in BOLD-contrast imaging, partially due to the fact that somatosensory feedback is inherent to natural motor performance. Additionally the hemodynamic response to finger tapping was expected to project to the precentral gyrus (M1 and premotor areas). Hence to judge the potential of high-resolution optical imaging we considered it mandatory to differentiate in both an anterior-posterior and a medio-lateraldimension.

What are the limitations of the approach? Clearly parts of the brain located in deep sulcal structures with some distance to the surface cannot be reached by optical approaches. The resolution of sub-divisions of SI, as it has been shown in primates (Merzenich et al., 1978; Kaas et al., 1979; Nelson et al., 1980; Sur et al., 1982; Pons et al., 1985), may be beyond the lateral resolution even of high-density optical approaches. The somatotopic organization is most pronounced in area 3b, and in area 3a located on the posterior wall of the central sulcus. These factors may contribute to the issue of precision of our approach. For high-resolution MRI differentiations of the sub-divisions of SI have been reported (Eickhoff et al., 2005; Schweizer et al., 2008). Refining co-registration with structural MRI and re-localizing the center of the image pad to optimally cover the area of interest may enhance the accuracy of the here presented approach. However, considering some fundamental limitations of diffuse-optical tomography we think that such sub-differentiation will be beyond even more sophisticated methodological approaches. The limitations inherent to non-invasive optical imaging can be divided into three fields. (i) The underlying physical principles of photon transport in strongly scattering media restrict the precision of the reconstruction of the sampled tissue. This is due to the fact that strong assumptions concerning the background optical properties are mandatory. Deviations from the assumed optical properties in the individual clearly limit precision. (ii) Anatomically, the variability of gyral structure and the thickness of the extra-cerebral tissue reduce the spatial accuracy of the derived activation foci. Since optical imaging does not supply anatomical information, the measured activation pattern can only be projected onto a gyral map when relying on external landmarks and cranio–cerebral correlations with the 10–10/10–20 electrode placement system (Okamoto et al., 2004) or when a structural MRI is performed in the individual subject. (iii) Physiologically, systemic changes in hemodynamics must be considered. This notorious issue in optical imaging may lead to a contamination of the vascular response with respect to amplitude but also timing (Boden et al., 2007). In high-resolution approaches this may be considered a lesser issue, since the high focality of the response is an intrinsic argument for the cortical origin of the detected changes. However, when comparing amplitude and temporal dynamics between groups or in their development over time, this may introduce another source of noise.