The Developmental Trajectory of Brain-Scalp Distance from Birth through Childhood: Implications for Functional Neuroimaging

Measurements of human brain function in children are of increasing interest in cognitive neuroscience. Many techniques for brain mapping used in children, including functional near-infrared spectroscopy (fNIRS), electroencephalography (EEG), magnetoencephalography (MEG) and transcranial magnetic stimulation (TMS), use probes placed on or near the scalp. The distance between the scalp and the brain is a key variable for these techniques because optical, electrical and magnetic signals are attenuated by distance. However, little is known about how scalp-brain distance differs between different cortical regions in children or how it changes with development. We investigated scalp-brain distance in 71 children, from newborn to age 12 years, using structural T1-weighted MRI scans of the whole head. Three-dimensional reconstructions were created from the scalp surface to allow for accurate calculation of brain-scalp distance. Nine brain landmarks in different cortical regions were manually selected in each subject based on the published fNIRS literature. Significant effects were found for age, cortical region and hemisphere. Brain-scalp distances were lowest in young children, and increased with age to up to double the newborn distance. There were also dramatic differences between brain regions, with up to 50% differences between landmarks. In frontal and temporal regions, scalp-brain distances were significantly greater in the right hemisphere than in the left hemisphere. The largest contributors to developmental changes in brain-scalp distance were increases in the corticospinal fluid (CSF) and inner table of the cranium. These results have important implications for functional imaging studies of children: age and brain-region related differences in fNIRS signals could be due to the confounding factor of brain-scalp distance and not true differences in brain activity.


Introduction
Human brain and behavior both undergo remarkable changes during development, and there is intense interest in using functional neuroimaging techniques to better understand neurodevelopment. Many of these techniques, including fNIRS, EEG, MEG and TMS measure (or evoke) brain function using probes placed on or near the scalp. The physical distance between the scalp and brain is therefore a critical parameter for these techniques, especially for fNIRS. In fNIRS, and in related techniques such as event-related optical signaling [1], low-power near-infrared light is directed through the scalp and intervening tissues into the surface of the brain [2,3]. Due to the differential absorption of specific wavelengths of near-infrared light by oxygenated and deoxygenated hemoglobin, concentration changes can be determined by measuring changes in the amount of nearinfrared light sensed by detectors located on the scalp some distance from the near-infrared transmitter. Hence, fNIRS measures the same hemodynamic signal as measured with blood-oxygen level dependent functional magnetic resonance imaging (BOLD fMRI), the most popular method for examining human brain function [4,5]. However, unlike fMRI, fNIRS depends on the transmission of light through the scalp, skull, meninges and CSF. The spatial sensitivity profile of fNIRS can be characterized as having a ''banana'' shape, with one end of the banana at the emitter, one end at the detector, and the body of the banana dipping down to sample the cortex [6]. The optimal placement of the detector and emitter therefore depends on the depth that light must penetrate: if the emitter and detector are close, more light will travel from the emitter to the detector, but none will travel through the brain; if the emitter and detector are distant, little light will reach the detector, resulting in poor signalto-noise ratio. Our study was spurred by our experience in recording responses from auditory cortex in children with fNIRS [7]. In order to determine the optimal emitter-detector distance, we wanted to establish the distance between the brain and scalp for the different aged children in our study population. While there are published studies of brain-scalp distance in adults [8,9,10] and children [11,12], we could find little information on how brainscalp distance changes during development. To fill this gap, we examined MRI scans from 71 healthy children ranging in age from 1 day to 12 years old. We hypothesized that there would be significant differences between ages, with younger children having reduced brain-scalp distances; and significant differences between brain areas, with greater brain-scalp distances in some regions relative to others.

Methods
Experiments were conducted in accordance with the Institutional Review Board of the University of Texas Health Science Center at Houston. Written informed consent was obtained from the guardian of each subject, and assent from the child subject if appropriate, prior to data collection. Information about scalpbrain distance was extracted from T1-weighted anatomical MRI images collected from each subject. The total subject population (n = 71) was assembled from 3 datasets.

Dataset 1
The first dataset consisted of fourteen healthy full-term newborns (mean age 1.9 d; mean birth weight 3183 g, mean birth length 50.1 cm). The newborns were scanned using the 3 tesla whole-body MRI scanner (Phillips Medical Systems, Bothell, WA) at the University of Texas Medical School at Houston using an 8-channel head coil. All children were scanned after feeding during natural sleep, without sedation. Silicone earplugs were used to reduce ambient scanner noise. Images were collected using a magnetization-prepared 180 degree radio-frequency pulses and rapid gradient-echo (MP-RAGE) sequence optimized for graywhite matter contrast. One hundred sagittal slices were collected (slice thickness 0.94 mm) with in-plane resolution of 0.703 mm (2566256 matrix).

Dataset 2
The second dataset consisted of seventeen healthy children aged 6 to 12 years that were scanned using the 3 tesla scanner at the University of Texas Medical School at Houston using an 8channel phased array head coil. Two MP-RAGE acquisitions were averaged to improve signal-to-noise ratio. One hundred and sixty sagittal slices were collected (slice thickness 1.0 mm) with in-plane resolution of 0.938 mm (2566256 matrix). External ear defendertype headphones were used to reduce ambient scanner noise.

Dataset 3
The third dataset consisted of forty subjects aged 7 months to 6 years obtained from release 4 of the Pediatric MRI Data Repository created by the NIH MRI Study of Normal Brain Development [13]. This is a multi-site, longitudinal study of typically developing children, from ages newborn through young adulthood, conducted by the Brain Development Cooperative Group. A listing of the participating sites and a complete listing of the study investigators can be found at http://www.bic.mni.mcgill. ca/nihpd/info/participating_centers.html. T1-weighted images were acquired from each subject with slightly varying parameters, depending on the acquisition site. Axial images were acquired (slice thickness of 3.0 mm) with in-plane voxel sizes between 0.938 and 1.0 mm. Additional information about the protocols are available from http://www.NIH-PediatricMRI.org.

Software Used for Data Analysis
All image analysis was conducted using AFNI, SUMA and other components of a freely available, open-source suite of programs widely used for performing MRI and fMRI analysis (http://afni. nimh.nih.gov) [14,15,16]. 3dSkullStrip (a program in the AFNI and SUMA suite) extracts the brain from MRI images using a variation on a surface growing approach [17] along with edge detection and size heuristics. 3dSkullStrip first corrects for gross spatial image non-uniformities and then expands a spherical surface iteratively until it envelops the brain [18]. Once the brain surface is found, it is progressively pushed outward and smoothed until it encounters the largest external gradient representing the scalp/air interface. 3dSkullStrip was found to be robust in a recent comparison of brain extraction methods [18]. IsoSurface is a program that performs isosurface extraction from a volume dataset. In this case, the volume is a binary mask of the brain or skull, and the extracted surface represents the outer boundary of that mask using the marching cubes algorithm [19]. SurfToSurf projects data from nodes on one surface to nodes on another. Each node from surface 1 is mapped to a node on the mesh of surface 2 that is closest to the intersection of the surface normal at node 1 with the mesh of surface 2.

Creation of Scalp Surface Models
It is important to note that all distance calculations were performed in three dimensions. While MRI data is commonly visualized as two-dimensional slices (e.g. Figure 1A), electromagnetic radiation, such as the infrared light used for fNIRS, propagates in three dimensions. Calculation of distance from slice views would lead to overestimation of the brain-scalp distance, because a closer distance could be found out-of-plane, unless the slice plane is precisely aligned with the geodesic. To allow accurate 3-D distance calculations, measuring the minimum brain-scalp distance that would actually be travelled by infrared light, a surface model of the scalp was created for each subject from the T1-weighted MRI using 3dSkullStrip and visualized using SUMA. The volumes were manually edited as needed using the Draw Dataset plug-in of AFNI and the IsoSurface program to extract the scalp boundary (yellow contour in Figure 1A).

Brain-Scalp Distance at all Brain Locations
For two subjects (shown in Figure 1) the brain-scalp distance was calculated at all brain locations. A brain hull surface was created to represent the envelope model of the brain using 3dSkullStrip and IsoSurface (green contour in Figure 1A). This 2manifold convex hull surface is akin to a shrink wrap of the brain: the wrapping surface closely hugs the brain where it is convex and is taut where the brain is concave. For each node on the brain hull, we searched for the intersection along the direction of the surface normal at that node and the scalp surface using SurfToSurf. The distance from the node on the convex hull to the intersection point on the scalp surface is the brain to scalp distance. The brain-scalp distance is poorly defined in places where the cortex is very far from the scalp, such as on the ventral surface of the brain. The distance is also meaningless in locations with scalp discontinuities, such as the ear pinna and ear canal, so brain-scalp distances in these locations were not analyzed ( Figure 1).

Selection of Manual Landmarks
As calculation of brain-scalp distance at all brain locations was labor intensive, for the remaining subjects we calculated brain-scalp distance for a limited number of manually-selected brain locations. In order to analyze brain landmarks of interest to developmental neuroimagers, especially those using fNIRS, we searched the published experimental fNIRS literature in March 2011, using keyword searches of 11 databases. Based on the results of the literature search, we selected nine cortical landmarks that spanned most of cortex and could be reliably identified on the T1-weighted MRI scan ( Figure 2). The first landmark was the occipital pole. The occipital pole was selected as the most posterior location in occipital lobe, which is of interest for fNIRS studies of visual function [20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37]. The left and right parieto-occipital sulcus (POS) landmarks were defined as the posterior and superior edge of the left and right POS, the most posterior location on the medial surface of the precuneus of the parietal lobe [38]. The vertex landmark was defined as the most superior location in the cerebral cortex, and commonly falls in the precentral gyrus. This is the location of primary motor cortex, and is important for neuroimaging studies of motor function [39,40,41,42]. The left and right Heschl's gyrus landmarks were chosen as the lateral-posterior edge of left and right Heschl's gyrus (HG), where it intersects the posterior lateral surface of the temporal lobe. HG contains core and belt areas of auditory cortex, and is especially important for developmental neuroimaging studies of auditory function and language [25,28,38,43,44,45,46,4 7,48,49,50,51,52,53,54,55,56,57,58,59,60]. The left and right inferior frontal gyrus (IFG) landmarks consisted of the most lateral portion of the pars triangularis. The left IFG landmark is near the cortical location of Broca's area, a critical node in the language network. The frontal pole landmark consisted of the most anterior location on the cerebral cortex, and was typically located near the intersection of superior frontal gyrus and the orbital gyri, relevant for frontal and prefrontal fNIRS studies of cognition [22,23,26,29,32,40,56,57,58,61,62,63,64,65,66,67,68,69,70,71,72,73,74,75,76,77,78,79,80].

Landmark Analysis
Scalp surface models were created as detailed above. For each manually-selected landmark in each subject, the SurfaceMetrics program was used to calculate the Euclidean distance between the landmark and the nearest location on the scalp surface in three dimensions. The location of the closest point on the scalp surface was manually verified for each landmark. For some landmarks in some subjects, the scalp surface could not be reconstructed accurately, a failure that was easily detectable during the manual verification step. As landmark-to-scalp distance could not be calculated without an accurate scalp reconstruction, these landmarks were excluded from the remainder of the analysis. In particular, the raw brain volumes from the NIH dataset were treated with a de-identifying procedure that removed potentially identifying facial features [81]. The deidentified datasets were missing a substantial amount of scalp and adjacent soft tissue from the MRI image near the front of the head, preventing calculation of the brain-scalp distance for the frontal pole landmark for these datasets.
For analysis of variance (ANOVA), the Matlab routine anovan was used. The dependent variable was the distance between the landmark and the scalp surface and the factors were age, handedness and landmark location. To analyze age effects, subjects were grouped into 3 age groups: 0-18 months inclusive; 19 months-5 years inclusive; 6 years-12 years inclusive. The subjects in dataset 2 were right-handed according to the Edinburgh handedness inventory [82]. For the subjects in dataset 3, handedness was assessed as ''the hand most commonly used to Figure 1. Whole-brain calculation of brain-scalp distance. A. Mid-sagittal slice through a full-term newborn brain (FT2009, age 2 days). The green line shows the reconstruction of the brain hull (bounding box of the cerebral cortex). The yellow line shows the reconstruction of the scalp surface. Both surfaces are truncated on the ventral surface of the brain (distance estimates were not computed for inferior brain regions where brainscalp distance is ill-defined.) The white boxes show the areas enlarged in (B) and (C). B. Enlargement of anterior portion of mid-sagittal slice of newborn brain. The white line shows the distance between the reconstruction of the brain hull (green line) and the scalp surface (yellow line) at one location in medial prefrontal cortex. Note that while the brain-scalp distance is shown on a two-dimensional slice for illustration, all distances were calculated using the minimum distance in three dimensions. C. Enlargement of posterior portion of mid-sagittal slice of newborn brain. D. Lateral view of newborn brain hull surface model. Colors indicate distance between brain and scalp at each brain location (color scale shown below). Lack of color indicates regions with ill-defined brain-scalp distance in inferior regions of the brain. E. Superior view of newborn brain hull surface model. F. Mid-sagittal slice through a child brain (subject CBB, age 7 years). G. Enlargement of anterior portion of mid-sagittal slice of child brain. Same scale as (B). H. Enlargement of posterior portion of mid-sagittal slice of child brain. I. Lateral view of child brain hull surface model. The brain is shown to the same scale as the newborn brain in (D). J. Superior view of brain hull surface model. Red colors indicate greatest distance between brain and scalp. doi:10.1371/journal.pone.0024981.g001 hold a pencil.'' The ANOVA with handedness as a factor was performed only on children for whom handedness data was available. Degrees of freedom are reported for each statistical test, along with statistical significance.
To provide a description of the relationship between age and scalp-brain distance, a linear function was fit to the data D = a * t+b where D is the scalp-brain distance, t is the age in months, and a and b are free parameters (Figure 3).

Intervals within the brain-scalp distance
The scalp-brain distance contains multiple intervals, including the CSF, the meninges, the cranium (including the compact bone and the cranial bone marrow) and the skin and subcutaneous fat of the scalp (cutis). As shown in Figure 4, only some of these intervals can be distinguished on T1-weighted MRI. The cranial bone marrow in the diploic space is hyperintense on T1-weighted MRI due to its high fat content and is visible in both infants and children. The cutis is also visible as a single hyperintense band. The other compartments are not distinguishable on T1-weighted MRI. We selected four points within the scalp-brain distance adjacent to each manually-selected landmark, L (   . Growth charts for brain-scalp distance. A. Mean growth chart averaged across all brain landmarks. Black line indicates best-fit growth function D = a * t+b where D is brain-scalp distance in mm (y -axis) and t is age in months (x -axis). Each blue square shows average brain-scalp distance in one subject. Manual selection of these four points was performed for 20 subjects, 10 subjects from Dataset 1 (newborns) and 10 subjects from Dataset 2 (children). All datasets were interpolated to 0.1 mm resolution before selection of these intervals to permit precise marking. All points for a given landmark were selected in a single plane: axial for the occipital and frontal poles; coronal for the vertex; and sagittal for the parietal landmarks. The IFG and HG landmarks were not processed because the temporalis muscle interfered with identification of the cranial bone marrow and cutis. As we were unable to create a surface model of the inner and outer boundaries of the cranial bone marrow, interval measurements were done only in two-dimensions, not the more accurate threedimensions used for total brain-scalp distance. This means that the interval measurements are an over-estimate of the true distance, but provide a rough estimate of developmental changes in the composition of the brain-scalp distance.

Whole-brain results
For two subjects (one full-term newborn and one seven years old) brain-scalp distance was calculated at all brain locations. There were large differences in brain-scalp distance between the two subjects ( Figure 1), with the older child showing larger brainscalp distances throughout the brain. The mean brain-scalp distance across the entire brain surface was 5.961.5 mm (mean 6 SD) for the newborn and 10.161.9 mm for the 7-year old. There were also large differences in brain-scalp distance between different brain regions, with areas near the vertex displaying the greatest distance. In the newborn, the mean distance was 7.660.4 mm for regions near the vertex compared with 4.960.5 mm for regions near the occipital pole. In the 7-year old, the distance was 12.960.7 mm for the vertex compared with 9.461.0 mm for the occipital pole.
As shown by the summary data (Table 1), the main effect of age was driven by larger scalp-brain distances in older subjects for most landmarks. The main effect of landmark was driven by significantly greater distances in parietal and occipital regions. To better understand how brain-scalp distance changed with age, we constructed growth charts similar to those used for other anthropometric variables for the mean scalp-brain distance across landmarks for each subject and for each individual landmark ( Figure 3). Growth curves were fit separately for each landmark, allowing the age and brain area of interest to be used to calculate a rough estimate of brain-scalp distance in future studies of subjects for whom an MRI is not available (all fit parameters provided in Figure 3 legend). The interaction between age and landmark is visible in the growth curves. Each landmark showed significant increases with age, with the steepest growth curve observed in parietal and occipital lobes (r 2 = 0.48, F 1,86 = 80, p = 10 213 ). The frontal pole brain-scalp distance showed the least change with age.
At all ages, there were significant laterality effects in some landmarks. The greatest laterality effect was observed in Heschl's gyrus, with a significantly greater distance in right compared with left Heschl's gyrus beginning at birth and continuing through development (averaged across ages, 9.3 mm vs. 7.7 mm, p = 10 25 ). A greater brain-scalp distance was also observed in right compared with left inferior frontal gyrus (10.9 mm vs. 9.5 mm, p = 0.002) but not between right and left parietal landmarks.

Components of brain-scalp distance
We estimated the thickness of four intervals within the brainscalp distance: the CSF and inner table (L to A in Figure 4), the cranial bone marrow, the outer table of the cranium, and the cutis (subcutaneous fat and skin) for newborns and children aged 5-12 years from Datasets 1 and 2 ( Figure 4). An ANOVA was performed with two factors, age (grouped into 2 age ranges: newborns and children aged 5 years-12 years inclusive) and interval. Significant effects were found for age (F 1,392 = 76, p = 10 216 ) and interval (F 3,392 = 158, p = 10 266 ), with a significant interaction (F 3,392 = 25, p = 10 214 ).
All four intervals showed a significant increase with age. The largest increase was in the CSF and inner table, with an increase

Discussion
Our examination of brain-scalp distance was motivated by the desire to use fNIRS to examine auditory and language function in patients that spanned a wide age range [7]. As fNIRS depends on the transmission of infrared photons through the skull and brain tissues, changes in brain-scalp distance present an additional confounding variable that is little understood. We found significant differences between brain regions, with the greatest brain-scalp distance over parietal regions, and the smallest difference over more inferior regions of the occipital and temporal lobes. There was a significant effect of laterality in some brain areas, with greater brain-scalp distance in right compared with left temporal and frontal regions. Across landmarks, brain-scalp distance increased with age, with the exception of the frontal pole, where it stayed relatively constant.
Our findings reflect differences in brain-scalp distances between parietal and temporal regions that are consistent with previous studies of adults [8,9,10] and children [11,12]. In particular, the hemispheric asymmetry that we observed in temporal regions is interesting given the importance of this region for language. In newborns, the left hemisphere is larger than the right hemisphere, especially in the temporal lobes, likely related to left-hemisphere dominance in speech and language [83,84,85]. Therefore, one possible explanation for the observed asymmetry in brain-scalp distance is that the skull is symmetric, but that left hemisphere brain growth results in a reduced distance between brain and the inner margin of the skull. Once established, asymmetries could remain due to the push-pull relationship between brain growth and skull expansion. For instance, hydrocephalus leads to increased intracranial pressure, causing an abnormally large skull (macrocephaly) [86]. While head circumference and brain volume are related [87] endocranium shape changes are not driven exclusively by brain growth [88]. Therefore, scalp-brain distance changes with age are likely due to a complex interplay of neural development and bone growth [89,90,91,92].

Implications for fNIRS Studies
A recent comparison of fMRI and fNIRS based on simultaneous measurements in both imaging modalities [8] found that the correlation between fMRI and fNIRS is negatively correlated with the scalp brain distance. At increasing scalp brain distances, there was much larger variance in the fNIRS signal and correspondingly weaker correlation of fNIRS with fMRI (used as a reference because it does not depend on transmission through the scalp and skull). These findings have important implications for interpreting and performing fNIRS studies of children. The optimal emitterprobe distance in fNIRS is proportional to the depth of the tissue from which the experimenter wishes to record. Therefore, our findings suggest that wider emitter-detector distances should be used in studies of parietal lobe and sensorimotor cortex because the brain-scalp distance is greater there, since the signal-tonoise of the fNIRS signal depends on brain-scalp distance [57,93,94,95,96]. Many infant fNIRS studies draw conclusions based on the change in concentration volume compared with baseline. For instance, more significant fNIRS responses in the left compared with right hemisphere are interpreted as a leftlateralized response to language [44,52]. However, our results show that the brain-scalp distance is significantly shorter in the left hemisphere than in the right hemisphere, a potential confounding factor. Even weak responses in right hemisphere may represent a robust response [46,47,50]. Our results also suggest caution in interpreting changes with age: decreases in responses to a stimulus in older children could indicate brain plasticity or simply an increase in brain-scalp distance. There was a high degree of variability in brain-scalp distance across landmarks within individual subjects, and for the same landmark across subjects. This variability will be additive with inter-subject differences in brain function, making group comparisons more difficult.
Most of the increase in brain-scalp distance with age was attributable to increases in the CSF and inner table distance. This is relevant because CSF has a scattering coefficient that is an order of magnitude less than other tissues within the brain-scalp interval. Therefore, CSF can act as a waveguide for the infrared light used in fNIRS, blurring the signal and reducing spatial resolution [97]. To better understand these effects, finite or boundary element models can be constructed [98,99]. Using these models together with the known fNIRS detector-sensory geometry, it is possible to more accurately infer hemoglobin concentration changes and hence brain activity differences [11,12,97].
Changes in distance between brain and scalp are also important for other human brain mapping techniques. In transcranial magnetic stimulation (TMS), electromagnetic radiation travels through the skull from a coil held outside the head in order to excite cortical neurons. This technique has received FDA approval for treatment of depression [100] and may be important for pediatric populations [101]. However, the electrical field induced by TMS falls off as the square of the distance from the coil. Therefore, in order to judge the optimal TMS current, it is important to have information about the brain-scalp distance: if the distance is small, low current will be sufficient to induce activity, while at greater distances, the same TMS current may not evoke any activity. One study estimated that for each additional mm between the scalp and brain, an additional 3% of TMS stimulator output is required to induce an equivalent motor response [9]. In the present manuscript, we selected brain landmarks of interest (such as Heschl's gyrus) and calculated the distance to the nearest point on the scalp. It is also possible to reverse this process and pick a point on the scalp, such as the inion, and calculate the distance to the nearest location on the brain, or to a particular landmark (such as the posterior margin of the calcarine fissure). This approach could be useful to understand how developmental changes in the scalp and brain influence the 10-20 system used for electrode placement used in EEG or optode placement in NIRS [10,102,103,104].

Conclusions
Our study serves as a reminder of the importance to neuroscientists of non-brain head tissue. While usually ignored, these tissues can have a considerable influence on human brain mapping studies. In PET studies of anxiety, increases in temporalis muscle blood flow due to teeth clenching were wrongly attributed to increased neural activity in the temporal lobe [105,106,107]. In a more recent example, gamma-band electroencephalography (EEG) changes were thought to reveal information about oscillatory neuronal activity related to higher cognitive function [108,109,110]. However, it is now believed that the gamma-band signal originates not from the brain but from eye-muscle potentials [111,112]. Neuroimaging of extra-brain tissue compartments can also produce useful information. For instance, the MRI signal from the orbit can be used to assess various eye movement parameters, including the saccade frequency [113] and fixation location [114].