www.elsevier.com/locate/ynimg NeuroImage 34 (2007) 1506 – 1518 Virtual spatial registration of stand-alone fNIRS data to MNI space Daisuke Tsuzuki, a,1 Valer Jurcak, a,1 Archana K. Singh, a,c Masako Okamoto, a Eiju Watanabe, b and Ippeita Dan a,⁎ a Sensory and Cognitive Food Science Laboratory, National Food Research Institute, 2-1-12 Kannondai, Tsukuba 305-8642, Japan Jichi Medical University, Japan c University of Tsukuba, Japan b Received 3 August 2006; revised 24 October 2006; accepted 26 October 2006 Available online 3 January 2007 The registration of functional brain data to common stereotaxic brain space facilitates data sharing and integration across different subjects, studies, and even imaging modalities. Thus, we previously described a method for the probabilistic registration of functional near-infrared spectroscopy (fNIRS) data onto Montreal Neurological Institute (MNI) coordinate space that can be used even when magnetic resonance images of the subjects are not available. This method, however, requires the careful measurement of scalp landmarks and fNIRS optode positions using a 3D-digitizer. Here we present a novel registration method, based on simulations in place of physical measurements for optode positioning. First, we constructed a holder deformation algorithm and examined its validity by comparing virtual and actual deformation of holders on spherical phantoms and real head surfaces. The discrepancies were negligible. Next, we registered virtual holders on synthetic heads and brains that represent size and shape variations among the population. The registered positions were normalized to MNI space. By repeating this process across synthetic heads and brains, we statistically estimated the most probable MNI coordinate values, and clarified errors, which were in the order of several millimeters across the scalp, associated with this estimation. In essence, the current method allowed the spatial registration of completely stand-alone fNIRS data onto MNI space without the use of supplementary measurements. This method will not only provide a practical solution to the spatial registration issues in fNIRS studies, but will also enhance cross-modal communications within the neuroimaging community. © 2006 Elsevier Inc. All rights reserved. Keywords: Optical topography; Diffused optical imaging; Transcranial magnetic stimulation; Probabilistic brain atlas; Talairach coordinate system; Spatial normalization; Human brain mapping; Clinical application ⁎ Corresponding author. Fax: +81 29 838 8122. E-mail address: dan@affrc.go.jp (I. Dan). 1 The two authors contributed equally to this work. Available online on ScienceDirect (www.sciencedirect.com). 1053-8119/$ - see front matter © 2006 Elsevier Inc. All rights reserved. doi:10.1016/j.neuroimage.2006.10.043 Introduction Functional near-infrared spectroscopy (fNIRS) is gaining popularity as a non-invasive tool for monitoring brain activity. fNIRS utilizes the tight coupling between neural activity and regional cerebral blood flow and monitors relative regional changes of hemoglobin concentration (reviewed in Hoshi, 2003; Koizumi et al., 2003; Obrig and Villringer, 2003; Strangman et al., 2002). Since fNIRS is a compact experimental system, is less expensive, easily portable, and relatively tolerant of body movements, it provides researchers with a means to use a wide variety of flexible experimental setups for clinical diagnosis and psychological experiments. fNIRS was only used to monitor a single or a few channels until the invention of differential illumination technology, which prevents cross-talk among closely situated illuminators, enabling up to a few dozen channels to be simultaneously monitored (Maki et al., 1995). Currently, even whole-head monitoring systems with more than a hundred channels are commercially available (Koizumi et al., 2003). As the number of channels increases, however, it becomes more tedious to set the optodes. This is one trade off for the convenience of fNIRS. Ultimately, the region of interest (ROI) tends to be confined to a smaller number of channels. Along with this, determining fNIRS channel locations is gaining importance: to realize reproducible fNIRS measurements across subjects and studies, the channel or optode locations should be statistically defined. Nevertheless, channel locations are only vaguely described in most fNIRS studies. Considering that an increasing number of studies are being performed in stand-alone settings, there should be a standardized way of describing the fNIRS channel locations, preferably in the common language of neuroimaging. Meanwhile, there is a strong trend in the neuroimaging community to represent different brain activation data in a common anatomical platform, which allows group analysis over multiple subjects and further comparison across different studies. In pursuit of a common arena for cross-modal assessment, there appeared a movement called “probabilistic atlas” for expressing all functional brain data as entries in a brain atlas that expands into space and time (Mazziotta et al., 2000, 2001a,b; Toga and Thompson, 2001). D. Tsuzuki et al. / NeuroImage 34 (2007) 1506–1518 1507 developed a virtual 10–20 measurement method for already acquired MR images to extend the reference head–brain database, and we further developed this method to be applicable to 10–10 and 10–5 systems (Jurcak et al., 2005, 2007). Fourth, we established a registration method based on the reference head– brain database, which enables fNIRS data registration on the standard MNI brain without MRIs of the subject (probabilistic registration method; Singh et al., 2005). We also described the error factor associated with probabilistic registration for both single-subject and group analysis (Singh et al., 2005). However, careful measurement of 10–20 landmarks and optode positions on a subject’s head using a 3D-digitizer is necessary to reproduce the optode placements in the MR images of the reference head–brain Fig. 1. Alignment of optodes and channels on each 3 × 3 and 3 × 5 holder. The white and gray circles indicate positions of optodes. When the white circles are illuminators, the gray circles are detectors, and vice versa. The x's indicate the positions of channels defined as a midway between neighboring optode pairs. Although the term itself is not usually addressed, its concept has already become widespread as a means of spatial data presentation in stereotaxic standard coordinate systems such as Talairach or MNI coordinates (Collins et al., 1994, Talairach and Tournoux, 1988). For expression of functional imaging data in these stereotaxic systems, structural brain imaging data are “normalized” or fit to the standard template brain by linear and nonlinear transformation processes (reviewed in Brett et al., 2002). The Talairach coordinate system is based on a single brain specimen with detailed descriptions of anatomical features including Brodmann estimates (Talairach and Tournoux, 1988). The MNI system is an extension of the Talairach system: the standard template of the MNI system was generated by fitting the brains of multiple subjects to the Talairach template and subsequently averaging them (Collins et al., 1994). Functional data are also registered for the normalized brain and hence to the MNI or Talairach coordinate systems. Accordingly, presenting functional mapping data on standard coordinate space has become a common practice for tomographic functional brain mapping methods, such as functional magnetic resonance imaging (fMRI) and positron emission tomography (PET). Despite the increasing importance of standard stereotaxic coordinate systems in neuroimaging studies, they have only recently been introduced to fNIRS studies (Okamoto et al., 2006a,b). Our group has presented a series of papers discussing the probabilistic estimation of fNIRS channels in standard stereotaxic coordinate systems. First we created an initial reference head–brain database presenting a probabilistic correspondence between the 10–20 standard positions in the real world and the MNI standard coordinate spaces (Okamoto et al., 2004). Second, we presented an algorithm to automatically project any given head surface position onto the underlying brain surface using MRIs of the subject, thereby presenting a theoretical framework to transform fNIRS data obtained on the head surface to the cortical surface (Okamoto and Dan, 2005). Third, we Fig. 2. Various fNIRS optode holders. (A) 3 × 3 rigid holder, top view. (B) 3 × 3 rigid holder placed on a glass head phantom. (C) A rigid holder cannot fit the head surface perfectly. As indicated by a pencil tip, there is a gap between the optode holder and the surface of the glass head phantom. (D) 3 × 5 elastic rubber holder, top view. (E) 3 × 5 elastic rubber holder placed on a glass head phantom. The holder undergoes elastic global deformation to fit the head surface. (F) 3 × 5 flexible holder, top view. The holder is made up of several strips. T-shaped reference strip is highlighted (black arrow). Examples of flexible strips are also highlighted (gray arrows). (G) 3 × 5 flexible holder placed on a glass head phantom. The holder fits the head surface as the angle between neighboring strips changes slightly (e.g., see skew in the upper left corner indicated by a black arrow head). 1508 D. Tsuzuki et al. / NeuroImage 34 (2007) 1506–1518 database. This imposes a certain burden on subjects and thus limits the application of the probabilistic registration method. Therefore, in this study, we propose a novel registration method to realize a 3D-digitizer-free registration of fNIRS data (virtual registration method). Essentially, the method simulates the placement of an optode holder on the scalp, taking into consideration its deformation and the registration of the optodes and channels onto subjects’ brains. We first describe an algorithm used to virtually place optode holders onto the head surface. Second, we examine the validity of the algorithm by comparing the predicted optode positions and the actual deformations of the optode holders on spherical phantoms and on real head surfaces. Third, we demonstrate virtual registration of an fNIRS optode holder onto MNI space with a simulated group dataset comprised of 1000 virtual subjects. We include a description of the associated error factor. Taken together, we present a virtual registration method for completely stand-alone fNIRS optodes and channels onto MNI space, based on the guidance of the 10–20 system and on prior knowledge of optode holder locations and deformations. Method and theory Subjects and source MRI datasets Six healthy adult volunteers (3 males and 3 females, aged 25 to 44 years) participated in the validation study of the deformation algorithm. Written informed consent was obtained after a complete explanation of the study. The study was approved by the institutional ethics committee of the National Food Research Institute of Japan. To simulate an fNIRS subject population, we used MRI datasets acquired from our previous study (Okamoto et al., 2004), which consist of the whole-head MRI images of 17 healthy volunteers (Mongoloid; 9 males, 8 females; aged 22 to 51 years). The details of the methods for image processing and virtual head surface measurements are described in our previous study (Jurcak et al., 2005). General description of fNIRS channel settings We will begin with a general description of optode holders and channel setting. We will take 3 × 5 and 3 × 3 holders as examples. Eight illuminators and seven detectors (or vice versa) in the 3 × 5 holder, or five illuminators and four detectors (or vice versa) in the 3 × 3 holder, are alternately placed in a lattice pattern (Fig. 1). The regions between optodes represent channels where cortical activation is measured. The actual signal source is estimated on the cortex as a photon density profile in a space or plane (Okada and Delpy, 2003; Kawaguchi et al., 2003). However, reconstruction of the signal sensitivity profile is beyond the scope of this paper. For simplicity, we will use the common definition of a channel: the midpoint between optodes (or its cortical projection point), where maximum photon density is realized. Specification of optode holders Before going into the details of virtual optode placement, it may be worthwhile to describe current trends in optode holders because the specifications of a holder influence the exact locations of optodes. In the early days, multichannel fNIRS studies used a fixedshape optode holder, where all optodes were placed on a hard plastic holder molded in a head shape (Figs. 2A, B). Obviously, there is a technical limitation to using a rigid holder: it does not fit heads with different sizes well (Fig. 2C). To circumvent this problem, the optode holder has undergone drastic changes, and now easy optode placement and better fixation are possible by using an elastic rubber holder (Figs. 2D, E; Hitachi Medical Corporation, Kashiwa, Japan) or a flexible holder (Figs. 2F, G; Shimadzu Corporation, Kyoto, Japan). Below, we will explain our method of virtual optode placement using these two types of holders as examples. We will treat them separately as each type requires a different strategy for reproducible placement and deformation simulation. Fig. 3. Procedure for reproducible setting of the elastic holder on a head surface. (A) In this example, we select T7 as the reference point on the scalp (green dot) and place a 3 × 5 holder so that the primary reference optode (the center optode on the lowest row, black dot) aligns with T7. (B) After the alignment, we rotate the holder around the reference scalp point so that the secondary reference optodes (red dots) are aligned with the reference curve (Fpz–T7–Oz shown in green) on the scalp in a balanced manner as in panel C. (D) A different strategy uses the secondary reference point (yellow dot), placed on a reference curve (Fpz–T7–Oz). D. Tsuzuki et al. / NeuroImage 34 (2007) 1506–1518 Reproducible placement of elastic holder An elastic holder is usually made by molding the rubber into a spherical surface so that a universal holder can fit adult subjects with common head sizes. In order to achieve reproducible optode orientation, the elastic probe holder should be placed according to the following guidelines. First, a given reference optode is set on a reference point on the scalp (Fig. 3A). Second, two or more secondary reference optodes are aligned with the reference curve in a balanced manner so that they are located equidistantly from the reference curve (Figs. 3B, C). Alternatively, a secondary reference optode is placed on a reference curve on the scalp (Fig. 3D). If this procedure is followed, elastic 1509 probe holders can be fixed to scalps across subjects in a reproducible manner. Virtual deformation algorithm for elastic holder When an elastic rubber holder is set on a scalp, it is deformed in various ways according to the shape of the head. In an attempt to understand the deformation pattern, we applied the holder to various surfaces. We found two tendencies. First, deformation occurs in such a way as to retain the global shape of the holder. Second, the final fitting adjustment is made by skewing the four corners of the holder. We developed a deformation algorithm that mimics these tendencies. We generated a virtual optode holder so that optodes are located Fig. 4. Virtual deformation algorithm for the elastic holder. An example of the 3 × 5 holder is shown. (A) On the head surface, we arbitrarily set a center optode (black dot) from which we generate a virtual holder. (B) We draw two perpendicular arcs (dotted blue curves) crossing at the center optode. (C) We set secondary reference optodes (blue dots; primary reference optode shown as a red dot) on the arcs at 30 mm intervals. (D) We set edge points (purple dots) neighboring secondary reference optodes so that distances from the edge optodes to two neighboring optodes along arcs are 30 mm. (E) We extend the procedure to the four corner optodes (purple dots). (F) At this point, the virtual holder has deviated from the final destination, so we shift the center optode position (black dot) to yield a new center optode (black open circle) where we begin a new holder generation as in A. We repeat the above processes until we finally place the holder in the desired manner as in panel G. The primary reference optode (red dot) is aligned with T7, and the secondary reference optodes (purple arrow heads) are aligned with the reference curve (Fpz–T7–Oz shown in green) in a balanced manner. 1510 D. Tsuzuki et al. / NeuroImage 34 (2007) 1506–1518 Fig. 5. Procedure for reproducible setting of a flexible holder on a head surface. (A) In this example, we select T7 as the reference point on the scalp (green dot) and place a 3 × 5 flexible holder with the reference T-shaped strip (red curves) perpendicularly intersecting at the primary reference optode (black dot). We place the holder so that the primary reference optode aligns with T7, and the optodes on the reference row (red dots) are well aligned with the reference curve on the scalp (Fpz–T7–Oz; green curve). (B) We set the holder arbitrarily and adjust its orientation so as to fulfil the rule of setting as in panel C. (D) Another possible setting is to place the holder so that the secondary reference optode (yellow dot) is on the reference curve (Fpz–T7–Oz) on the scalp. at equal distances from neighboring optodes and that virtual optodes are always propagated from the center of the holder. We will provide a more detailed explanation of how our algorithm accomplishes virtual holder placement by taking a 3 × 5 optode holder with 30 mm inter-optode distances placed on a given head surface (Fig. 4). First, we set a temporary point on the head surface near the expected center of the holder placement (Fig. 4A) and draw two perpendicular arcs which intersect at that point (Fig. 4B). Second, we set optode positions for primary and secondary reference points on the arcs at 30 mm intervals (Fig. 4C). Third, we set edge points so that the distance from an edge point to two neighboring optodes along the arcs is 30 mm (Fig. 4D). Fourth, we extend the procedure to the four corner points (Fig. 4E). However, in order to set the virtual holder to a given target on a scalp, an additional step was necessary. Virtual holder generation always starts at the center of the holder, but any optode or channel can be selected as the primary reference optode or channel (Fig. 4F). Therefore, we can roughly locate the virtual holder so that the primary reference optode or channel can be close to the reference point on a scalp (e.g., T7 in Fig. 4F). Then, we shift the primary reference optode slightly (Fig. 4F) and re-generate the virtual holder. If the resulting orientation is not satisfactory, we repeat this Fig. 6. Virtual deformation algorithm for a flexible holder. An example for a 3 × 5 holder is shown. (A) We set the primary reference optode (black dot) on the reference point on the scalp, which is T7 in this case (green circle). We draw two perpendicular arcs (red curves) from the scalp reference point, so that the horizontal arc aligns with the reference curve (Fpz–T7–Oz shown in green) on the scalp. (B) We place secondary reference optodes (blue dots) on the arcs at 30 mm intervals. (C) We set tertiary reference optodes (purple dots) so that distances from two neighboring secondary reference optodes were 30 mm. (D) Similarly, we set other optodes so that distances from horizontally or perpendicularly neighboring optodes were 30 mm. D. Tsuzuki et al. / NeuroImage 34 (2007) 1506–1518 1511 procedure until the location and orientation of the virtual holder satisfy the given rule of placement (Fig. 4G). arrange the holder so that the secondary optode or channel can be on the reference curve on the scalp (Fig. 5D). Reproducible placement of flexible holder Virtual deformation algorithm for flexible holder With a flexible holder, optode holes are connected with short strips of fixed lengths (Fig. 2F, gray arrows) and a large T-shaped strip (Fig. 2F, black arrow). The short strips freely rotate, resulting in three-dimensional deformation, which allows the holder to fit to a head surface with constant inter-optode distances. Meanwhile, the T-shaped strip maintains the global shape of the holder by reducing the freedom of rotations to some extent. In order to achieve reproducible optode orientation, the probe holder should be arranged according to the following guidelines. First, a primary reference optode is set on an optode holder. Second, a reference row of optodes that crosses the primary reference optode is set. Third, a column of reference optodes that crosses the primary reference optode and that keeps a fixed angle (usually 90°) to the reference row of optodes is set (Figs. 2F, 5A). To set the holder on a scalp, we need a reference point and a reference curve that runs through the reference point. First we set the primary reference optode or channel on the reference point on the scalp (Fig. 5A). Second, we align the reference column or row with the reference curve (Figs. 5B, C). Alternatively, we can define a secondary optode or channel on the reference column or row and The probe orientation of a flexible holder can be defined using a relatively simple algorithm. First, a reference optode is set on the reference point on the head surface (Fig. 6A). Second, the first reference curve that crosses the reference optode is set on the head surface, and the optode positions are determined at 30 mm intervals along the curve (Fig. 6B). Third, we draw a second reference curve that is perpendicular to the first reference curve and set optode positions at 30 mm intervals along it (Fig. 6B). Fourth, we set an additional optode position at 30 mm from the two neighboring optode positions along the head surface (Fig. 6C), and we repeat this procedure until all optode positions are determined (Fig. 6D). Examination of deformation algorithms using spherical phantoms and real heads We examined whether the above algorithms allow us to simulate actual deformation by placing the elastic and flexible holders on spherical phantoms of various sizes (Fig. 7). To determine the sizes of phantoms, we examined the curvature of various head regions using synthetic heads (see Virtual registration for details of synthetic Fig. 7. Comparison of virtual and actual holder deformation using a spherical phantom. (A) Fixation of a spherical phantom and a transmitter for the 3D-digitizer. (B) Placement of holder and plotting of the optode position with a color marker. (C) Measurement of the optode positions using the 3D-digitizer. (D) The ideal sphere in virtual space with a spherical phantom of the same size. (E) An elastic holder (black dots) is generated on the ideal sphere. (F) The actual (black dots) and virtual (white crosses) optode positions are aligned for comparison. 1512 D. Tsuzuki et al. / NeuroImage 34 (2007) 1506–1518 heads). We isolated a head surface area that is approximately the same size as the area covered by a 3 × 3 or 3 × 5 holder. Then we fit a sphere to the extracted head region using the least square method and obtained the radius of the best-fit sphere. We randomly selected a region and orientation from the synthetic heads for 5000 repetitions to obtain the distribution of the radii of the best-fit spheres. We used spheres with sizes that cover the 95% confidence interval of the distribution. We placed 3 × 3 and 3 × 5 elastic and flexible optode holders on the spherical phantoms (Fig. 7B). The optode positions were marked on the spheres, and their positions in the real world coordinate system were measured with a magnetic 3D-digitizer (Polhimus) (Fig. 7C). Meanwhile, we generated virtual spheres of the same sizes as the spherical phantoms (Fig. 7D). On the surfaces of the virtual spheres, we generated virtual holders (Fig. 7E). We aligned the actual and virtual optode positions on the surface of the virtual spheres and compared them (Fig. 7F). We examined the validity of the algorithms by comparing the deformations of the holders on actual head surfaces with the virtual deformations (Fig. 8). In order to make the surface of a subject’s head as smooth as possible, we covered the head tightly with a swimming cap and vinyl tape (Fig. 8A). We placed 3 × 5 elastic and flexible optode holders on the five different regions of the head: frontopolar, right and left temporal, and the parietal and occipital regions (Table 3). After marking the optode positions (Fig. 8B), we measured their positions (Fig. 8C). We included head surface regions within an additional margin of 15 mm from the regions covered by the holders and ultimately measured approximately 900 regional head surface positions for reconstruction on the computer (Figs. 8C, D). For head surface reconstruction, we smoothed the head surface by removing protruding points (artifacts during measurement), wrapped the remaining points with mesh grids, and interpolated the regions more dense with points at 0.25 mm intervals. Consequently, approximately 170,000 points comprising a flat surface that matches a real subject’s head were generated. On the surface of the reconstructed head surface, we generated virtual holders (Fig. 8E). We aligned the actual and virtual optode positions on the surface of the reconstructed head surfaces and compared their positions (Fig. 8F). Virtual registration Head shape and size vary among subjects, while the area covered by an optode holder is relatively invariant. Therefore, virtual registration of fNIRS optodes and channels onto MNI space is necessarily accompanied by spatial errors. Error estimation is an integral issue for virtual registration in order to clearly express the validity and the limitations of spatial estimations of channel locations. In our previous study, we described errors associated with 3D-digitizer-assisted probabilistic registration (Singh et al., Fig. 8. Comparison of virtual and actual holder deformation using a real head surface. (A) We smooth the head surface using a swimming cap and fix a transmitter for a 3D-digitizer. (B) We place an elastic optode holder and plotted the optode positions with a color marker. (C) We measure the locations of optodes and neighboring head surfaces. (D) The head surface, including optodes (black dots), is realized in virtual space. (E) From the center optode (black dot), we generate the virtual holder. Virtual optodes are shown in white crosses. (F) A virtual holder (white crosses) thus generated is best aligned to the actual holder positions (black dots) and subjected to comparison. D. Tsuzuki et al. / NeuroImage 34 (2007) 1506–1518 2005). We demonstrated a method for a typical group study, and instead of MR images of subjects, we used our own reference database comprised of normalized MR images of 17 individuals (who had not participated in the study). We use a random effects ANOVA model to elucidate random errors arising from variability among reference brains and among subjects. In the current study, we used simulated datasets to assess the variability among subjects (Fig. 9). Ideally, we would have preferred real datasets, but we needed a large group for a closer estimation of the null distribution. Therefore, we simulated our group data from un-normalized MRI datasets of the 17 individuals (Fig. 9A) using two parameters: head size (Fig. 9B) and head shape (Fig. 9C). From the seventeen nasion (Nz) to inion (Iz) observations, we estimated the distribution of head sizes (i.e., generated the normal distribution of head size; 185.6 ± 7.5 mm) and randomly assigned a head size for each simulation according to the normal distribution. Then, we measured a head shape represented by Nz–Iz, AL–AR (left and right preauricular points), and m–Cz distances, where m stands for the gravity center of Nz, Iz, AL, and AR. There is likely to be a certain degree of coherence between these distances and head size: a larger Nz–Iz distance would mean larger AL–AR and m–Cz distances. To preserve this coherence, we used this ratio (Iz–Nz: AL–AR: m–Cz) to represent the head shape. We generated synthetic head and brain sets by randomly combining head sizes and shapes, along with one of the MRI datasets (Fig. 9D). For each synthetic head and brain, we applied the virtual holder deformation algorithm to estimate the head surface points representing fNIRS optodes and channels, projected the head surface points to the brain (as described in Okamoto and 1513 Dan, 2005), and obtained the corresponding cortical surface points (Fig. 9E). In order to register the positional data to MNI space, we first transformed the data obtained from the synthetic head and brain back to the original MRI dataset. We then transformed the positional data to MNI space using our in-house program, as previously described (Okamoto and Dan, 2005). We repeated the above procedure 1000 times. The common estimate and estimation error for each optode or channel location are calculated in MNI space as mean and standard deviation for the 1000 simulated datasets (Fig. 9F). Results Examination of deformation algorithms using spherical phantoms We examined whether holder deformation algorithms can simulate actual deformation by placing elastic and flexible holders on spherical phantoms. Prior to the physical measurements, we used virtual heads to explore the range of spheres that fit the curvature of various head regions. Fig. 10 shows the distribution of the radii of the best-fit spheres. For 3 × 5 holders, the minimum and maximum radii were 60.2 and 183.6 mm, respectively; the 95% confidence interval was 68.6 to 127.1 mm. For 3 × 3 holders, the minimum and maximum radii were 55.6 and 245.9 mm, respectively; the 95% confidence interval was 64.2 to 136.9 mm. Accordingly, we used spheres with radii of 62.5 to 223 mm to investigate deformation algorithms. For 3 × 5 elastic holders, we set the primary reference point for holder placement at the center, and its horizontally neighboring point as the secondary reference point. For 3 × 5 flexible holders, Fig. 9. Flow chart for the virtual registration method. First, we randomly choose one of the MRI datasets (A) and combine two parameters: a head size parameter (Nz–Iz distance) from the normalized head size distribution (B), and a head–shape ratio (Nz–Iz, AL–AR, and m–Cz ratio) (C), to generate a synthetic head and brain set (D). We then apply virtual holder deformation to estimate optode and channel locations on head (blue dots) and brain (red dots) surfaces (E) and transfer the spatial information to the MNI space. We repeat this procedure (A–E) 1000 times. From the 1000 datasets, we calculate common estimates and estimation errors for each optode or channel location in MNI space (F). 1514 D. Tsuzuki et al. / NeuroImage 34 (2007) 1506–1518 were negligible (less than 2 mm) for all the cases examined. Therefore, we concluded that the algorithm can closely mimic the deformation of elastic and flexible holders. Demonstration of the virtual registration method Now that the virtual holder deformation algorithms have been validated, we will demonstrate how to implement the virtual registration method in actual situations. We prepared the placement of the holder on five different scalp regions, as summarized in Table 3. We used elastic and flexible holders for this demonstration, but it should be noted that we used slightly different rules for placing the holders to enhance their deformation characteristics, but not to evaluate which is better. The results of virtual registration for optode positions on the scalp and their cortical projections are shown in Fig. 11. In all cases, the centers of the circles represent the most likely estimate and their radii represent the standard deviation of displacement in MNI space. In terms of lower error rates, spatial estimates were most stable in the frontopolar region, followed by temporal regions, and the parietal region, and were the least stable in the occipital region. This regional tendency is the same as the correlation of 10–20 positions and MNI space as we used them as guidance for optode holder placement. However, estimation errors remain within 13 mm. This is a reasonable compromise for virtual registration without MR images and digitizer measurements. Discussion Fig. 10. Distribution of the regional curvatures for the scalp regions covered by (A) 3 × 5 holders and (B) 3 × 3 holders. The distributions are based on random placement simulations of the virtual holders for 5000 repetitions. we set the center point as the primary reference point, and the reference curves to cross perpendicularly at the center of the holder. We also tested the deformations of 3 × 3 elastic and flexible holders. The results are summarized in Table 1. The discrepancy between the simulation using the virtual holder deformation algorithm and the actual holder deformation was negligible (less than 2 mm) for most cases. The only exception was fitting a 3 × 5 elastic holder on a 125 mm sphere, where the maximum deviation was 2.9 mm. However, as demonstrated in the regional best-fit sphere simulation, such an extreme case is unlikely in actual situations. Therefore, we concluded that the algorithm for the virtual holder registration is valid for spheres that correspond to most adult scalp surfaces. Examination of deformation algorithms using real heads Although spherical phantoms can be used to accurately simulate extreme deformations in optode holders that may occur with both small and large heads, rather complex deformations occurring on actual head surfaces with irregular shapes may be beyond their scope. Hence, we examined whether the above algorithms can be used to simulate actual deformations on a head surface by placing the holders onto four different regions of the head surfaces of six subjects. The results are summarized in Table 2. Discrepancies between simulations using the virtual holder deformation algorithm and actual holder deformation on real heads As presented above, our method makes the virtual spatial registration of stand-alone fNIRS optodes and channels onto the MNI space possible. Here we would like to discuss how this method can facilitate fNIRS research, especially in a more global context surrounding the whole neuroimaging community. Neuroimaging techniques serve not only neuroimaging research, but also as important tools for other scientific disciplines. In this perspective, fNIRS has great potential. Its affordability, compactness, and less restrictiveness allow flexible experimental settings and thus can contribute to the expansion of the frontiers of functional neuroimaging research. For this purpose, data sharing within the neuroimaging community is essential. By describing fNIRS data on common stereotactic coordinate systems, cross-reference between fNIRS data and functional and anatomical data obtained by other neuroimaging modalities is made possible. The current study presents a practical and fair solution to this data sharing issue. Table 1 Examination of deformation algorithms using spherical phantoms Diameter (mm) 125 148 197 248 296 446 Elastic holder 3 × 5 Flexible holder 3 × 5 Elastic holder 3 × 3 Flexible holder 3 × 3 Max Avg Max Avg Max Avg Max Avg 2.9 1.2 1.0 0.6 0.7 1.3 1.1 0.6 0.5 0.3 0.3 0.6 1.3 1.5 1.2 0.8 0.9 0.6 0.7 0.7 0.6 0.3 0.4 0.3 1.7 0.8 1.0 0.5 0.4 0.8 0.7 0.4 0.5 0.3 0.3 0.4 0.8 0.7 1.4 1.0 0.7 0.5 0.5 0.3 0.6 0.3 0.4 0.3 Max represents the maximum deviation between actually and virtually deformed holders. Avg represents the average deviation for all the optode positions on the holder tested. Both Max and Avg are presented in mm. 1515 D. Tsuzuki et al. / NeuroImage 34 (2007) 1506–1518 Table 2 Examination of deformation algorithms using real heads Head region Subject 1 Subject 2 Subject 3 Subject 4 Subject 5 Subject 6 Max Avg Max Avg Max Avg Max Avg Max Avg Max Avg 1.4 1.5 1.4 1.0 0.7 0.8 0.6 0.6 1.6 1.7 1.8 1.0 0.7 0.8 0.7 0.5 1.8 1.6 1.3 1.2 0.7 0.7 0.6 0.7 1.3 1.0 1.8 1.5 0.8 0.7 0.7 0.6 1.2 1.3 1.6 1.6 0.5 0.7 0.8 0.6 1.5 1.1 1.3 1.0 0.6 0.5 0.8 0.5 Flexible holder (3 × 5) Frontopolar 1.4 Occipital 1.9 Left temporal 1.8 Right temporal 1.4 0.6 0.9 0.9 0.8 1.9 1.2 1.5 1.7 0.9 0.6 0.7 0.6 1.5 0.9 1.2 1.2 0.8 0.6 0.6 0.6 1.6 1.4 1.2 1.7 0.8 0.9 0.6 0.8 1.3 1.1 1.6 1.3 0.7 0.7 1.0 0.6 1.4 1.6 1.4 1.3 0.7 0.8 0.8 0.7 Elastic holder (3 × 5) Frontopolar Occipital Left temporal Right temporal Max represents the maximum deviation between actually and virtually deformed holders. Avg represents the average deviation for all the optode positions on the holder tested for each participant. Both Max and Avg are presented in mm. In previous studies, we described a 3D-digitizer-mediated method that enabled probabilistic registration of fNIRS optodes and channels onto the MNI standard coordinate system (probabilistic registration method; Singh et al., 2005). However, the probabilistic registration method requires some additional measuring time, and this may not be desirable, especially in emerging clinical situations such as pre-surgery language dominance determination (Watanabe et al., 1998), preoperative planning for tumor removal (Fujiwara et al., 2004), focus diagnosis of epilepsy (Watanabe et al., 2000, 2002), rehabilitation monitoring (Miyai et al., 2001, 2003), psychiatric diagnosis (Ehlis et al., 2005; Fallgatter and Strik, 1997, 1998, Suto et al., 2004; Kameyama et al., 2006), and postoperative cerebral oxygenation monitoring (Murata et al., 2003; Hoshino et al., 2006). The virtual registration method demonstrated in the current study provides a promising alternative solution to spatial registration issues in fNIRS studies, especially for clinical applications. The virtual registration method has two prerequisites. First, probe holders for fNIRS are set in a reproducible manner, preferably with the guidance of a stable scalp landmark setting such as the international 10–20 systems or its derivatives, or the 10–10 or 10–5 systems (Jasper, 1958; Chatrian et al., 1985; Nuwer et al., 1998; Klem et al., 1999; Oostenveld and Praamstra, 2001). Second, the deformation of a probe holder should, to a certain degree, follow predictable patterns. As long as these prerequisites are satisfied, the virtual registration method presented in the current study enables fNIRS optode and channel positions to be registered on standard stereotaxic brain coordinate systems, including the MNI system, without using any additional devices. In this study, we used two probe holders for demonstration only because they were locally available. Since virtual registration algorithms can be developed for any probe holders that deform in predictable ways, we are planning to extend virtual registration algorithms to other types of probe holders. The accuracy of the estimations of the virtual registration method is approximately the same as that of the 3D-digitizermediated probabilistic registration method (Singh et al., 2005). For most of the lateral cortical regions, the errors of estimation, expressed as standard deviations, are approximately 1 cm. This range of spatial error may be sufficient for most functional studies, provided that the widths of major gyri are approximately 1 cm. However, spatial errors were not identically distributed among brain regions, e.g., we observed larger errors in the virtual registration in the occipital region. This is primarily because virtual registration is affected by errors of nearby reference positions (i.e., 10–20 positions) used for transformation to MNI standard coordinate system (Okamoto et al., 2004; Singh et al., 2005; Jurcak et al., 2007). In the occipital region, reference positions are relatively large mainly due to unambiguity in inion location thereby causing larger errors. Such large errors cannot be avoided as long as we use 10–20, 10–10, or 10–5 systems for placing probe holders. In search of improved virtual registration, we are looking for plausible alternative to conventional 10–20, 10–10, or 10–5 systems. One major obstacle of the current study is how users can apply this method in practical situations. The procedures used in the current study are not very straightforward, requiring the user to provide many parameters and small adjustments for each virtual holder registration. We are planning to modify the program to store the results of virtual holder registrations as an empirical database, so that the users need not spend time inputting the same parameters more than once. Since many previously published Table 3 Specifications for holder placement used to demonstrate virtual registration method Scalp region Reference optode Reference row Reference point on the scalp Reference curve Frontopolar Occipital Left temporal Right temporal Parietal [3,3] 3 Fpz T7–Fpz–T8 [3,3] 3 Oz T7–Oz–T8 [3,3] 3 T7 Fpz–T7–Oz [3,3] 3 T8 Fpz–T8–Oz [3,3] 3 Cpz T7–Cpz–T8 For example, the frontopolar column is read as the reference optode on the third row of the third column of the holder, which is placed on Fpz so that the third optode row aligns to the T7–Fpz–T8 reference curve. 1516 D. Tsuzuki et al. / NeuroImage 34 (2007) 1506–1518 Fig. 11. Demonstration of virtual registrations. Cases for elastic holders are shown. Virtual holders were placed on five different regions on the head: (A) frontopolar, (B) occipital, (C) left and (D) right temporal, and (E) parietal regions. They were projected onto the cortical surface (F–J). Rules for placement are summarized in Table 3. Red circles indicate the optode positions for elastic holders. Centers of circles represent the most likely estimates and their radii represent the standard deviation of displacement. Axes are correspondent to those in MNI space. D. Tsuzuki et al. / NeuroImage 34 (2007) 1506–1518 fNIRS studies satisfy the prerequisites mentioned above, we are planning to re-examine them spatially. It is important to note that the virtual registration method is sensitive to the type of holder used, and how it is set. For example, what seems to be a detailed description (e.g., “we placed the center of the lower optode rows of 3 × 5 optode holders with an interoptode distance of 30 mm at Fpz, and the remaining four lower optodes were aligned with the Fp1–Fpz–Fp2 line”) is not sufficient to correctly place a virtual holder. The type of holder used must also be included. There is another important pitfall. We have to make sure that the inter-optode distance remains exactly the same. For example, a commercial, flexible holder uses a patch between holder and head to ensure a comfortable fit, but, depending on the head size, this patch results in a change in inter-optode distance. We have already incorporated this parameter into our simulation. In general, any careful reports on the methods used are preferred. Although the current method is optimized for fNIRS studies, it could be easily modified for transcranial magnetic stimulation (TMS) mapping data (Wassermann et al., 1992; Brasil-Neto et al., 1992), which are typically represented by regional mesh grids developed on a scalp and are topographically similar to an fNIRS optode holder. However, the current algorithm is optimized to simulate probe positioning based on two commercially available fNIRS optode holders and needs further modification to be applied for TMS studies. In conclusion, this study serves as an important practical extension to our series of studies. The virtual registration method enables the registration of multi-subject or single-subject fNIRS data into the MNI coordinate system not only without structural MRIs of the subjects, but also without a 3D-digitizer. Reporting fNIRS data in MNI space is only recently becoming a common practice. However, with the methods presented in this study, this practice will become readily accessible. The use of common stereotaxic platforms will make it possible for fNIRS researchers to have access to a wealth of neuroimaging data expressed in MNI and Talairach spaces and also to convenient tools available in these spaces, such as a maximum probability atlas (Hammers et al., 2003) and computer-assisted anatomical labeling (Lancaster et al., 2000; Le Goualher et al., 1999; Nowinski and Belov, 2003; Tzourio-Mazoyer et al., 2002). On the other hand, the presentation of fNIRS data in common stereotactic space will help all neuroimaging researchers understand and assess fNIRS data. Thus, we believe that the current study will enhance multi-modal communication within the neuroimaging community. Conflict of interest statement All authors hereby declare that they have no financial and personal relationships with other people or organization that could inappropriately influence our work. Acknowledgments We thank the subjects who participated in this study. We are grateful to Dr. Haruka Dan for her helpful advice. We thank Ms Akiko Oishi for preparation of the manuscript and data, and Ms Melissa Nuytten for examination of the manuscript. This work is partly supported by the Industrial Technology Research Grant Program in 03A47022 from the New Energy and Industrial Technology Development Organization (NEDO) of Japan, and 1517 Grant-in-Aid for Scientific Research 18390404 from the Japan Society for the Promotion of Science. References Brasil-Neto, J.P., McShane, L.M., Fuhr, P., Hallett, M., Cohen, L.G., 1992. Topographic mapping of the human motor cortex with magnetic stimulation: factors affecting accuracy and reproducibility. Electroencephalogr. Clin. Neurophysiol. 85, 9–16. Brett, M., Johnsrude, I.S., Owen, A.M., 2002. The problem of functional localization in the human brain. Nat. Rev., Neurosci. 3, 243–249. Chatrian, G.E., Lettich, E., Nelson, P.L., 1985. Ten percent electrode system for topographic studies of spontaneous and evoked EEG activity. Am. J. EEG Technol. 25, 83–92. Collins, D.L., Neelin, P., Peters, T.M., Evans, A.C., 1994. Automatic 3D intersubject registration of MR volumetric data in standardized Talairach space. J. Comput. Assist. Tomogr. 18, 192–205. Ehlis, A.C., Herrmann, M.J., Wagener, A., Fallgatter, A.J., 2005. Multichannel near-infrared spectroscopy detects specific inferior-frontal activation during incongruent Stroop trials. Biol. Psychol. 69, 315–331. Fallgatter, A.J., Strik, W.K., 1997. Right frontal activation during the continuous performance test assessed with near-infrared spectroscopy in healthy subjects. Neurosci. Lett. 223, 89–92. Fallgatter, A.J., Strik, W.K., 1998. Frontal brain activation during the Wisconsin Card Sorting Test assessed with two-channel near-infrared spectroscopy. Eur. Arch. Psychiatry Clin. Neurosci. 248, 245–249. Fujiwara, N., Sakatani, K., Katayama, Y., Murata, Y., Hoshino, T., Fukaya, C., Yamamoto, T., 2004. Evoked-cerebral blood oxygenation changes in false-negative activations in BOLD contrast functional MRI of patients with brain tumors. NeuroImage 21, 1464–1471. Hammers, A., Allom, R., Koepp, M.J., Free, S.L., Myers, R., Lemieux, L., Mitchell, T.N., Brooks, D.J., Duncan, J.S., 2003. Three-dimensional maximum probability atlas of the human brain, with particular reference to the temporal lobe. Hum. Brain Mapp. 19, 224–247. Hoshi, Y., 2003. Functional near-infrared optical imaging: utility and limitations in human brain mapping. Psychophysiology 40, 511–520. Hoshino, T., Sakatani, K., Kano, T., Murata, Y., Katayama, Y., 2006. Cerebral blood oxygenation changes induced by bypass blood flow in moyamoya disease and non-moyamoya cerebral ischaemic disease. Acta Neurochir. (Wien) 148 (5), 551–557. Jasper, H.H., 1958. The ten–twenty electrode system of the International Federation. Electroencephalogr. Clin. Neurophysiol. 10, 367–380. Jurcak, V., Okamoto, M., Singh, A., Dan, I., 2005. Virtual 10–20 measurement on MR images for inter-modal linking of transcranial and tomographic neuroimaging methods. NeuroImage 26, 1184–1192. Jurcak, V., Tsuzuki, D., Dan, I., 2007. 10/20, 10/10, and 10/5 systems revisited: their validity as relative head-surface-based positioning systems. NeuroImage 34, 1600–1611. Kameyama, M., Fukuda, M., Yamagishi, Y., Sato, T., Uehara, T., Ito, M., Suto, T., Mikuni, M., 2006. Frontal lobe function in bipolar disorder: a multichannel near-infrared spectroscopy study. NeuroImage 29, 172–184. Kawaguchi, H., Hayashi, T., Kato, T., Okada, E., 2003. Evaluation of spatial resolution of near-infrared topography using spatial sensitivity profile. Photon migration and diffuse-light imaging. SPIE 5138, 249–257. Klem, G.H., Luders, H.O., Jasper, H.H., Elger, C., 1999. The ten–twenty electrode system of the International Federation. The International Federation of Clinical Neurophysiology. Electroencephalogr. Clin. Neurophysiol., Suppl. 52, 3–6. Koizumi, H., Yamamoto, T., Maki, A., Yamashita, Y., Sato, H., Kawaguchi, H., Ichikawa, N., 2003. Optical topography: practical problems and new applications. Appl. Opt. 42, 3054–3062. Lancaster, J.L., Woldorff, M.G., Parsons, L.M., Liotti, M., Freitas, C.S., Rainey, L., Kochunov, P.V., Nickerson, D., Mikiten, S.A., Fox, P.T., 2000. Automated Talairach atlas labels for functional brain mapping. Hum. Brain Mapp. 10, 120–131. 1518 D. Tsuzuki et al. / NeuroImage 34 (2007) 1506–1518 Le Goualher, G., Procyk, E., Collins, D.L., Venugopal, R., Barillot, C., Evans, A.C., 1999. Automated extraction and variability analysis of sulcal neuroanatomy. IEEE Trans. Med. Imaging 18, 206–217. Maki, A., Yamashita, Y., Ito, Y., Watanabe, E., Mayanagi, Y., Koizumi, H., 1995. Spatial and temporal analysis of human motor activity using noninvasive NIR topography. Med. Phys. 22, 1997–2005. Mazziotta, J., Toga, A., Evans, A., Fox, P., Lancaster, J., Woods, R., 2000. A probabilistic approach for mapping the human brain. In: Toga, A.W., Mazzoiotta, J.C. (Eds.), Brain Mapping: The Systems. Academic Press, San Diego, pp. 141–156. Mazziotta, J., Toga, A., Evans, A., Fox, P., Lancaster, J., Zilles, K., Woods, R., Paus, T., Simpson, G., Pike, B., Holmes, C., Collins, L., Thompson, P., MacDonald, D., Iacoboni, M., Schormann, T., Amunts, K., PalomeroGallagher, N., Geyer, S., Parsons, L., Narr, K., Kabani, N., Le Goualher, G., Boomsma, D., Cannon, T., Kawashima, R., Mazoyer, B., 2001a. A probabilistic atlas and reference system for the human brain: International Consortium for Brain Mapping (ICBM). Philos. Trans. R. Soc. London, Ser. B Biol. Sci. 356, 1293–1322. Mazziotta, J., Toga, A., Evans, A., Fox, P., Lancaster, J., Zilles, K., Woods, R., Paus, T., Simpson, G., Pike, B., Holmes, C., Collins, L., Thompson, P., MacDonald, D., Iacoboni, M., Schormann, T., Amunts, K., PalomeroGallagher, N., Geyer, S., Parsons, L., Narr, K., Kabani, N., Le Goualher, G., Feidler, J., Smith, K., Boomsma, D., Hulshoff Pol, H., Cannon, T., Kawashima, R., Mazoyer, B., 2001b. A four-dimensional probabilistic atlas of the human brain. J. Am. Med. Inform. Assoc. 8, 401–430. Miyai, I., Tanabe, H.C., Sase, I., Eda, H., Oda, I., Konishi, I., Tsunazawa, Y., Suzuki, T., Yanagida, T., Kubota, K., 2001. Cortical mapping of gait in humans: a near-infrared spectroscopic topography study. NeuroImage 14, 1186–1192. Miyai, I., Yagura, H., Hatakenaka, M., Oda, I., Konishi, I., Kubota, K., 2003. Longitudinal optical imaging study for locomotor recovery after stroke. Stroke 34, 2866–2870. Murata, Y., Katayama, Y., Sakatani, K., Fukaya, C., Kano, T., 2003. Evaluation of extracranial–intracranial arterial bypass function by using near-infrared spectroscopy. J. Neurosurg. 99, 304–310. Nowinski, W.L., Belov, D., 2003. The Cerefy Neuroradiology Atlas: a Talairach–Tournoux atlas-based tool for analysis of neuroimages available over the Internet. NeuroImage 20, 50–57. Nuwer, M.R., Comi, G., Emerson, R., Fuglsang-Frederiksen, A., Guerit, J.M., Hinrichs, H., Ikeda, A., Luccas, F.J., Rappelsburger, P., 1998. IFCN standards for digital recording of clinical EEG. International Federation of Clinical Neurophysiology. Electroencephalogr. Clin. Neurophysiol. 106, 259–261. Obrig, H., Villringer, A., 2003. Beyond the visible-imaging the human brain with light. J. Cereb. Blood Flow Metab. 23, 1–18. Okada, E., Delpy, D.T., 2003. Near-infrared light propagation in an adult head model. II. Effect of superficial tissue thickness on the sensitivity of the near-infrared spectroscopy signal. Appl. Opt. 16, 2915–2922. Okamoto, M., Dan, I., 2005. Automated cortical projection of head-surface locations for transcranial functional brain mapping. NeuroImage 26, 18–28. Okamoto, M., Dan, H., Sakamoto, K., Takeo, K., Shimizu, K., Kohno, S., Oda, I., Isobe, S., Suzuki, T., Kohyama, K., Dan, I., 2004. Threedimensional probabilistic anatomical cranio-cerebral correlation via the international 10–20 system oriented for transcranial functional brain mapping. NeuroImage 21, 99–111. Okamoto, M., Matsunami, M., Dan, H., Kohata, T., Kohyama, K., Dan, I., 2006a. Prefrontal activity during taste encoding: an fNIRS study. NeuroImage 31, 796–806. Okamoto, M., Dan, H., Singh, A.K., Hayakawa, F., Jurcak, V., Suzuki, T., Kohyama, K., Dan, I., 2006b. Prefrontal activity during flavor difference test: application of functional near-infrared spectroscopy to sensory evaluation studies. Appetite 47, 220–232. Oostenveld, R., Praamstra, P., 2001. The five percent electrode system for high-resolution EEG and ERP measurements. Clin. Neurophysiol. 112, 713–719. Singh, A.K., Okamoto, M., Dan, H., Jurcak, V., Dan, I., 2005. Spatial registration of multichannel multi-subject fNIRS data to MNI space without MRI. NeuroImage 27, 842–851. Strangman, G., Boas, D.A., Sutton, J.P., 2002. Non-invasive neuroimaging using near-infrared light. Biol. Psychiatry 52, 679–693. Suto, T., Fukuda, M., Ito, M., Uehara, T., Mikuni, M., 2004. Multichannel near-infrared spectroscopy in depression and schizophrenia: cognitive brain activation study. Biol. Psychiatry 55, 501–511. Talairach, J., Tournoux, P., 1988. Co-Planar Stereotaxic Atlas of the Human Brain. Thieme, New York. Tzourio-Mazoyer, N., Landeau, B., Papathanassiou, D., Crivello, F., Etard, O., Delcroix, N., Mazoyer, B., Joliot, M., 2002. Automated anatomical labeling of activations in SPM using a macroscopic anatomical parcellation of the MNI MRI single-subject brain. NeuroImage 15, 273–289. Toga, A.W., Thompson, P.M., 2001. Maps of the brain. Anat. Rec. 265, 37–53. Wassermann, E.M., McShane, L.M., Hallett, M., Cohen, L.G., 1992. Noninvasive mapping of muscle representations in human motor cortex. Electroencephalogr. Clin. Neurophysiol. 85, 1–8. Watanabe, E., Maki, A., Kawaguchi, F., Takashiro, K., Yamashita, Y., Koizumi, H., Mayanagi, Y., 1998. Non-invasive assessment of language dominance with near-infrared spectroscopic mapping. Neurosci. Lett. 30, 49–52. Watanabe, E., Maki, A., Kawaguchi, F., Yamashita, Y., Koizumi, H., Mayanagi, Y., 2000. Noninvasive cerebral blood volume measurement during seizures using multichannel near infrared spectroscopic topography. J. Biomed. Opt. 5, 287–290. Watanabe, E., Nagahori, Y., Mayanagi, Y., 2002. Focus diagnosis of epilepsy using near-infrared spectroscopy. 43, Suppl. 9, 50–5.