Cerebral Cortex November 2009;19:2595--2604

doi:10.1093/cercor/bhp011

Advance Access publication February 24, 2009

Relationships between Brain Activationand Brain Structure in NormallyDeveloping Children

Lisa H. Lu1,2, Mirella Dapretto3,4, Elizabeth D. O’Hare1,4,

Eric Kan1, Sarah T. McCourt1, Paul M. Thompson1,4, Arthur

W. Toga1, Susan Y. Bookheimer3,4 and Elizabeth R. Sowell1,4

1UCLA Laboratory of Neuro Imaging, Department of Neurology,

Los Angeles, CA 90095, USA, 2Roosevelt University, Department

of Psychology, Chicago, IL 60605, USA, 3UCLA Department of

Psychiatry and Biobehavioral Sciences, Los Angeles, CA 90095,

USA and 4UCLA Interdepartmental Ph.D. Program for

Neuroscience, Los Angeles, CA 90095, USA

Dynamic changes in brain structure, activation, and cognitiveabilities co-occur during development, but little is known abouthow changes in brain structure relate to changes in cognitivefunction or brain activity. By using cortical pattern matchingtechniques to correlate cortical gray matter thickness and functionalbrain activity over the entire brain surface in 24 typically developingchildren, we integrated structural and functional magnetic resonanceimaging data with cognitive test scores to identify correlates ofmature performance during orthographic processing. Fast-namingindividuals activated the right fronto-parietal attention network inresponse to novel fonts more than slow-naming individuals, andincreased activation of this network was correlated with moremature brain morphology in the same fronto-parietal region. Theserelationships remained even after effects of age or general cognitiveability were statistically controlled. These results localized corticalregions where mature morphology corresponds to mature patterns ofactivation, and may suggest a role for experience in mediating brainstructure--activation relationships.

Keywords: attention, fMRI, imaging, language, morphometry

Introduction

Structural brain imaging studies have revealed dynamic spatial

and temporal patterns of brain development, but how such

morphological changes relate to cognitive skills remain largely

unexplored. Specifically, the dorsal cortices of the frontal and

parietal lobes show dramatic and regionally variable trajecto-

ries of change during childhood and adolescence (Sowell,

Thompson, Holmes, Jernigan, et al. 1999; Sowell et al. 2001,

2003; Gogtay et al. 2004; Sowell, Thompson, Leonard, et al.

2004; Shaw et al. 2006), with parietal cortices thinning

most rapidly between childhood and adolescence (Sowell,

Thompson, Holmes, Batth, et al. 1999), and frontal lobes

thinning most rapidly between adolescence and young adult-

hood (Sowell, Thompson, Holmes, Jernigan, et al. 1999). This

is important as executive functions attributable to frontal

structures (Fuster 2002) continue to develop between

adolescence and adulthood. However, considerable variability

in cortical structure has been observed cross-sectionally

(Sowell et al. 2003), even between individuals of the same

age, and the growth trajectory is associated with gray matter

thickening in some brain regions and thinning in other

regions (Sowell et al. 2003; Sowell, Thompson and Toga 2004;

Lu et al. 2007). Such variability challenges efforts to un-

derstand the relationship between morphological maturation

and cognitive skill maturation.

Significant variability in functional brain activity has also

been observed during childhood and adolescence (Turkeltaub

et al. 2003; for example). Developmental differences in acti-

vation patterns have manifested as a greater extent of activation

in children than in adults (Thomas et al. 1999; Gaillard et al.

2000), increase in the degree of hemispheric lateralization with

age (Holland et al. 2001), greater intensity of activation in

adults than in children (Thomas et al. 1999; Gaillard et al. 2003;

Booth et al. 2004), or increased activation with age in key

regions related to the task accompanied by decreased ac-

tivation with age in regions less centrally related to the task

(Rivera et al. 2005). It has been speculated that through

experience and maturation, there is a shift from more diffuse

activation to more focal activation as plasticity declines and

efficiency improves (Durston and Casey 2006; Durston et al.

2006), though developmental changes progressing from fewer

to greater number of connections have also been proposed and

supported (Brown et al. 2005).

It is tempting to speculate that changes in functional

activation observed during the childhood and adolescent

period of rapid cognitive development are related to changes

in brain structure, but little evidence to support this has been

reported in the literature. In this study, we used cortical

pattern matching techniques to anchor anatomical land-

marks across individuals (Sowell, Thompson, Leonard, et al.

2004; Thompson et al. 2004) and integrated structural

morphology with corresponding functional activation. Figure

1 depicts the model that guided our study. We first defined

‘‘mature’’ activation by identifying brain regions where

activation intensity corresponded with skill level improvement.

Then we looked for regions where ‘‘mature’’ activation

corresponded with morphological growth trajectories associ-

ated with maturation in previous reports (i.e., thinning in dorsal

fronto-parietal regions; thickening in perisylvian regions)

(Sowell, Thompson, Holmes, Jernigan, et al. 1999; Sowell et al.

2001, 2003; Gogtay et al. 2004; Sowell, Thompson, Leonard,

et al. 2004; Shaw et al. 2006). If there are regions of overlap,

then morphological maturation is associated with maturation of

functional activation. It is important to note that both genetic

influences that unfold with age and experience contribute to

skill level. We statistically parsed out variance associated with

age to illuminate structure--activation relationships indepen-

dent of age effects. We hypothesized that mature cortical

thickness patterns would be associated with mature activation

patterns. Specifically, we expected thinning in frontal-parietal

regions to be associated with more mature activation patterns

during attentional aspects of an orthographic processing

functional magnetic resonance imaging (fMRI) task and

thickening in perisylvian language regions to be associated

with more mature activation patterns on reading aspects of

the task.

� The Author 2009. Published by Oxford University Press. All rights reserved.

For permissions, please e-mail: journals.permissions@oxfordjournals.org

Materials and Methods

SubjectsTwenty-eight normally developing children without history of neuro-

logical, psychiatric, or developmental disorders were recruited from

the community near the University of California, Los Angeles, CA. All

learned English as their native language. All subjects and their parents

gave written assent/consent according to procedures approved by the

UCLA Institutional Review Board. Three subjects whose structural or

functional scan had inadequate image quality were excluded. One

subject was excluded due to excessive movement. Table 1 provides

demographic descriptions of the remaining 24 subjects (age ranged

from 6 to 15).

Functional TaskOrthographic processing is a skill which develops rapidly during

childhood (Sprenger-Charolles et al. 2003) and involves detection of

visual features within printed stimuli. The orthographic processing task

described by Turkeltaub et al. (2003) required subjects to indicate the

presence of an ascending character in visually presented stimuli of either

real words or false font strings (e.g., k, l for the real word condition, and

similar nonletter tall characters in the false font string condition) by

button press with their right index finger and to indicate absence with

their right middle finger (Fig. 2). This task (Turkeltaub et al. 2003) was

ideal for our purposes because it activates both brain regions whose

maturational trajectory is characterized by gray matter thickening (i.e.,

primary language cortices in the perisylvian region) as well as regions

characterized by thinning (i.e., dorsal frontal and parietal regions) (Sowell,

Thompson, Leonard, et al. 2004). The false font stringminusword contrast

was thought to reflect attentional response to the salience of novel print,

as all task demands for the false font string and word conditions were

exactly the same except that false font strings were novel, word-like

stimuli never seen before. The word minus false font string contrast was

thought to reflect implicit reading. Stimuli and task parameters were

exactly as described by Turkeltaub et al. (2003) except interstimulus

fixation lasted 2.8 s and intercondition rest period lasted 16 s.

Behavioral TasksBecause our functional activation task was an orthographic processing

task, we chose a cognitive measure whose maturation level was

characterized by fluency with orthography: naming speed. The speed

with which letters and digits are named becomes faster with age

(Wagner et al. 1999), and naming speed in the preschool age range

predicts later reading achievement (Wagner et al. 1997; Wolf and

Bowers 1999). Naming speed was therefore selected as the operation-

alized index of cognitive development, and correlations between brain

activation and naming speed (i.e., time in seconds) allowed us to

identify ‘‘mature’’ patterns of activation with the orthographic process-

ing task. Activation (i.e., increased or decreased) associated with faster

naming was defined to be more ‘‘mature.’’

Structural Image AcquisitionHigh-resolution T1-weighted sagittal volumes were collected from a 1.5

Tesla (T) Siemens Sonata scanner (repetition time [TR], 1900 ms; echo

time [TE], 4.38 ms; flip angle, 15�; matrix size, 256 3 256 3 160; field of

view [FOV], 256 mm; voxel size, 1 3 1 3 1 mm; acquisition time, 8 min,

8 s). We chose to acquire the structural images on the 1.5T magnet

because it yields less susceptibility artifact than the 3T magnet, thus

allowing more consistent tissue segmentation throughout the volume.

Two to 4 acquisitions were acquired for each subject. Raters blind to

subject age and sex evaluated image quality, and data from at least 2

acquisitions were averaged to enhance signal-to-noise ratio. There was

no significant relationship between the number of image acquisitions

used and age (Pearson’s r = 0.21, P = 0.32).

Figure 1. Model that guided the present study. Regions where there is overlapbetween activation and skill improvement reflect regions where activation intensitychanges with better skill (striped). Activation of such regions is therefore defined as‘‘mature.’’ Activation is then correlated with morphology. We looked for regions where‘‘mature’’ activation corresponded with morphological growth trajectory (i.e., thinningin dorsal fronto-parietal regions; thickening in perisylvian regions). If there are regionsof overlap (checker board), then morphological maturation is associated withmaturation of functional activation.

Table 1Demographic description of 24 normally developing subjects and their mean performance on

behavioral measures

Mean SD

Age (range 6--15) 10.54 2.81Number of males 11 —Handedness quotienta 73.88 23.12WISC-4 FSIQb 113.30 15.82CTOPP rapid naming standard scorec 102.62 12.71Ortho word % correct 93.8 7.1Ortho false font string % correct 95.3 5.6

aDextrality quotient from the modified Edinburgh Handedness Inventory (Oldfield 1971).bWechsler Intelligence Scale for Children (WISC), 4th ed, prorated FSIQ (Wechsler 2003).cComprehensive Test of Phonological Processing (CTOPP; Wagner et al. 1999). WISC and CTOPP

scores have a mean of 100 and SD of 15.

Figure 2. Orthographic processing task. Subjects were instructed to press a buttonwith their right index finger if an ascender was present, and a button with their rightmiddle finger if there was no ascender. In the examples given here, the correctresponse for ‘‘alarm’’ and its corresponding false font string was ‘‘yes,’’ and thecorrect response for ‘‘sauce’’ and its corresponding false font string was ‘‘no.’’ Falsefont strings matched real words for length, size, and location of ascenders anddescenders. Words containing the letters i and j were excluded to avoid confusion.Scan repetition time was 4 s. Ten volumes were acquired from each experimentalblock and 4 from each rest period. The initial 2 volumes corresponding to theinstruction (‘‘GET READY!’’) were discarded to exclude measurements preceding T1equilibrium. The experiment lasted 7 min, 52 s and yielded 116 usable volumes.

2596 Brain Activation--Structure Relationships in Children d Lu et al.

Functional Image AcquisitionFunctional imaging data were obtained from a 3T Siemens Allegra head-

only scanner. First, we acquired a high-resolution structural T2-weighted

echo-planar image (EPI) volume in the anterior commissure--posterior

commissure plane (TR, 5000 ms; TE, 33 ms; matrix size, 128 3 128; FOV,

20 cm; 36 slices; voxel size, 1.6 3 1.6 3 3; 3 mm thick; 1-mm gap)

coplanar with the functional scan to allow for spatial registration of each

subject’s data into a standard coordinate system. During the orthographic

processing task, one functional scan lasting 7 min and 52 s was acquired

covering the whole brain volume (116 images; EPI gradient echo

sequence; TR, 4000 ms; TE, 25 ms; flip angle, 90�; matrix size, 64 3 64;

FOV, 20 cm; 36 slices; voxel size, 3.1 3 3.1 3 3; 3 mm thick; 1 mm gap).

The EPI acquisition available on the 3T magnet allows us to collect more

image volumes, with better signal-to-noise ratio and shorter scanning

time than would be possible on the 1.5T magnet. Thus, functional and

structural data were acquired on 2 different magnets, and spatially

registered in image processing.

Subjects viewed visual stimuli from magnet-compatible goggles

containing 2 miniature television screens with full 512 3 512 resolution

(Resonance Technology, Northridge, CA). Stimuli were presented using

MacStim 3.2 psychological experimentation software (WhiteAnt

Occasional Publishing, West Melbourne, Australia).

Structural Imaging Data AnalysisPreprocessing of high-resolution structural imaging data from the 1.5T

scanner were identical to those described previously (Sowell et al.

2002; Sowell, Thompson, Leonard, et al. 2004). Briefly, the MR images

were preprocessed with a series of manual and automated procedures

executed by analysts blind to subject age and sex: 1) transform brain

volumes into a standardized 3D coordinate space (Mazziotta et al.

1995) using a 12 parameter, linear, automated image registration

algorithm (Woods et al. 1993); 2) semiautomated tissue segmentation

was conducted for each volume data set to classify voxels based on

signal intensity as most representative of gray matter, white matter, or

cerebral spinal fluid (Shattuck et al. 2001); 3) remove nonbrain tissue

(i.e., scalp, orbits) and the cerebellum, and exclude the left hemisphere

from the right; 4) automatically extract the cortical surface of each

hemisphere, which was represented as a high-resolution mesh of 131

072 triangulated elements spanning 65 536 surface points in each

hemisphere (MacDonald et al. 1994); 5) trace 35 sulcal and gyral

landmarks on the lateral and medial surfaces of each hemisphere using

detailed criteria that we have developed (Sowell et al. 2002) for

delineating the starting and stopping points for each sulcus using brain

surface atlases as references (Ono et al. 1990; Duvernoy et al. 1991) (17

on the lateral surface of each hemisphere: Sylvian fissure, central,

precentral, postcentral, superior temporal sulcus (STS) main body, STS

ascending branch, STS posterior branch, inferior temporal, superior

frontal, inferior frontal, intraparietal, primary intermediate sulcus,

secondary intermediate sulcus, transverse occipital, olfactory, occipito-

temporal, and collateral sulci. Twelve on each interhemispheric

surface: callosal sulcus, inferior callosal outline, superior rostral sulcus,

inferior rostral sulcus, paracentral sulcus, anterior and posterior

segments of the cingulate sulcus, outer segment double parallel

cinglate sulcus when present, parieto-occipital sulcus, anterior and

posterior segments of the calcarine sulcus, and the subparietal sulcus.

Six midline landmark curves bordering the longitudinal fissure were

delineated to establish hemispheric gyral limits); 6) transform the

image volumes back into their own native image acquisition space by

mathematically inverting the transformation which took them into

standard space; 7) spatially register all segmented images and brain

surfaces for each individual to a standard orientation by using the first

and last points on 20 of the 35 manually defined anatomical landmarks

matched to a standard atlas (Sowell et al. 2003); and (8) measure

cortical thickness in millimeters at each anatomically matched cortical

surface point.

The thickness of gray matter was calculated using the Eikonal Fire

Equation (Sapiro 2001; Thompson et al. 2004). Although the brain

images acquired for this study had voxel dimensions of approximately 1

3 1 3 1 mm, we supersampled the imaging data to create voxel

dimensions of 0.33 mm3 using trilinear interpolation (Ratnanather et al.

2004; Aganj et al. forthcoming). The 3D Eikonal equation was applied

only to voxels that segmented as gray matter, and a smoothing kernel

was used to average gray matter thickness within a 15-mm sphere at

each point on the cortical surface. The cortical surface area within each

sphere likely varied depending on its location within the 3-dimensional

thickness volume for each subject. Nonetheless, these methods allowed

us to calculate cortical thickness for each subject at an effective

resolution much finer than that of the original voxel size in the image,

given that the error associated with localizing anatomy on the inner and

outer cortical surfaces is averaged with the unbiased error of all other

voxels within the smoothing kernel. Once preprocessing was

completed, points on the cortical surfaces surrounding and between

the sulcal contours drawn on each individual’s brain surface were

calculated using the averaged sulcal contours as anchors to drive 3D

cortical surface mesh models from each subject into correspondence

(Thompson et al. 2004). This cortical pattern matching technique, also

known as high-dimensional continuum mechanical image warping

(Thompson et al. 2001, 2003), allows the creation of average surface

models while accounting for cortical variability across subjects. All

analyses of the thickness maps were conducted in each subject’s native

(unscaled space).

To map gray matter thickness onto the surface rendering of each

child’s brain, the coordinate of each point on the cortical surface for

each child (anatomically matched across individuals) was mapped to

the same anatomical location in their ‘‘thickness’’ volume, and the

average maximum thickness of gray matter within a 15-mm sphere was

calculated. In a previous report, we helped to establish the validity of

these methods by showing close regional correspondence between

maps of cortical thickness created for normally developing children in

vivo (Sowell, Thompson, Leonard, et al. 2004) and for the post mortem

data of Von Economo (von Economo 1929). In our earlier report

(Sowell, Thompson, Leonard, et al. 2004), we also assessed the test-

retest reliability of measures of cortical thickness in individuals scanned

twice at short time intervals, demonstrating maximum error estimates

of 0.15 mm.

Functional Imaging Data AnalysisPreprocessing and statistical analysis of functional imaging data were

carried out using FSL (Oxford Centre for Functional Magnetic Resonance

Imaging of the Brain [FMRIB]’s Software Library, Oxford University,

Oxford, UK; www.fmrib.ox.ac.uk/fsl). We corrected for motion by using

MCFLIRT (Jenkinson et al. 2002) and for slice-timing by using Fourier-

space time-series phase-shifting. Nonbrain tissues were removed using

BET (Smith 2002). Spatial smoothing was applied with a Gaussian kernel

of full width half maximum of 6 mm. Mean-based intensity normalization

of all volumes by the same factor was applied, as well as highpass

temporal filtering (Gaussian-weighted least-squares straight line fitting,

with sigma = 15 s).

After preprocessing, statistical analyses were performed at the single-

subject level by using the general linear model within FSL (FEAT [FMRI

Expert Analysis Tool] version 5.63). Each experimental condition was

modeled using a boxcar function convolved with a canonical hemo-

dynamic response function. Time-series statistical analysis was carried

out using FILM (FMRIB’s Improved Linear Model) with local autocor-

relation correction (Woolrich et al. 2001). Volumes for which greater

than 2 mm of correction of motion was required were modeled as

a covariate of no interest. Results were rendered on Z statistic images

thresholded using clusters determined by Z > 1.7 and a (corrected)

cluster significance threshold of P = 0.05 (Worsley et al. 1992). Each

subject’s functional data were first registered to corresponding

structural volumes using 6-parameter rigid-body transformation, then

spatially normalized to the Montreal Neurological Institute-152

template using 12-parameter affine registration via FLIRT (Jenkinson

and Smith 2001; Jenkinson et al. 2002) for group analyses.

Mixed-effects group analyses were carried out using FLAME (FMRIB’s

Local Analysis of Mixed Effects) (Beckmann et al. 2003; Woolrich et al.

2004). Variances and parameters resulting from single-subject fixed

effects analyses were carried into higher-level mixed-effects analysis to

allow for inferences to be drawn at the population level. Higher-level

statistical maps were thresholded by using clusters determined by Z >

1.7 and a (corrected) cluster significance threshold of P = 0.05.

Cerebral Cortex November 2009, V 19 N 11 2597

Coregistering Functional (3T) and Structural (1.5T) DataBefore correlation between activation and gray matter thickness could

be conducted, data acquired from 2 different scanners were coreg-

istered into the same space. Structural T1- and T2-weighted images

(from 1.5T and 3T scanners, respectively) were first registered into the

standardized International Consortium for Brain Mapping’s 305 3D

coordinate space (Mazziotta et al. 1995) using a 12 parameter, linear,

automated image registration algorithm (Woods et al. 1993) (trans-

formation 1), then registration was further refined by rigid-body

transformation of T2-weighted volume into the T1-weighted space

(transformation 2). These 2 transformation files were combined by

multiplying together transformation matrices and the result was

applied to the activation data (the unthresholded t-map for each

subject representing the false font string minus word contrast) to bring

the functional activation map into correspondence with the high-

resolution structural data collected on the 1.5T magnet. Then the high-

dimensional continuum mechanical image warping transformation

created from the 1.5T T1-weighted structural data was applied to this

3T T2-weighted structural data (which was registered to the functional

activation data also collected on the 3T magnet) to bring both

functional activation and cortical thickness into anatomical correspon-

dence. An average activation map (using the unthresholded t-map for

each individual) for each individual was created using a spatial

smoothing kernel of a 15 mm (radius) sphere, identical to the one

used for the cortical thickness maps.

Statistical AnalysesCorrelations between the 2 contrasts (t-scores representing the

magnitude of false font string minus word and word minus false font

string) and gray matter thickness were examined by calculating

Pearson’s r correlation coefficients on a point-by-point basis for each

cortical surface point (approximately 60 000 per hemisphere).

Correlations between the activation at each surface point and naming

speed for each individual subject were also calculated. Note that the

word minus false font string contrast is simply the inverse of the false

font string minus word contrast, and thus statistical maps in Figure 5

were only shown for the false font string minus word contrast. To

control for multiple comparisons, we conducted permutation analyses

(Nichols and Holmes 2004) by randomly permuting activation intensity

values and associated correlates (i.e., gray matter thickness or naming

speed) in 1000 new analyses. The number of significant correlations

(Pearson’s r with P < 0.05) in these 1000 new analyses fell along

a normal curve, against which we compared our observed number of

significant correlations with original, nonpermutated data. The observed

number of significant correlations was deemed to pass correction for

multiple comparisons if the probability of it occurring by chance was less

than 0.05. Permutation analyses were conducted within 10 regions of

interest (ROIs) in each hemisphere, 6 on the lateral surface and 4 on the

medial surface of the brain. Coarse ROIs (Fig. 3) were created for each

individual from a probabilistic atlas for the frontal, parietal, temporal, and

occipital lobes (Evans et al. 1994) with 2 modifications. First, the frontal

lobe was divided into dorsal and ventral ROIs by an axial plane passing

through the intersection of the inferior frontal sulcus and the precentral

sulcus. Second, we created a perisylvian ROI based on our previous

longitudinal study of children whose age range overlapped with the age

range of subjects in the present study (Sowell, Thompson, Leonard, et al.

2004). This perisylvian ROI corresponded to the only brain region where

we observed gray matter thickening with age.

Multiple regression was used to evaluate the relationship between

activation and naming speed with age regressed out. The relationship

between activation and naming speed independent of general cognitive

development was also evaluated with multiple regression with Full

Scale Intelligence Quotient (FSIQ) as covariate.

Results

Functional Task Performance

All 24 subjects performed the orthographic processing task

(i.e., correctly identifying words or false font strings with tall

characters) with at least 78% accuracy and overall significantly

better than chance (one-sample t-test compared to chance

performance of 50%, t (23) = 38, P < 0.01), and accuracy was

equivalent for word and false font string conditions (see Table 1;

paired sample t (23) = –1.25, P = 0.22). Response time (RT) to

words and false font strings were similar (paired sample t(23) =0.60, P = 0.56). Both accuracy rate and response time improved

with age (Word accuracy, r = 0.50, P = 0.01; False font string

accuracy, r = 0.65, P < 0.01; Word RT, r = –0.56, P < 0.01; False

font string RT, r = –0.58, P < 0.01). Because the contrast of

interest was the difference between false font string and word

conditions, whether accuracy and response time differences

between false font string and word conditions were related

to age was most pertinent for the purposes of this study, and

they were not (Word-false font string accuracy and age, r = –0.02,

P = 0.94; Word-false font string RT and age, r = –0.17, P = 0.43).

Functional Task Activation

Cortical pattern matching was used to create group average

t-maps (uncorrected) representing attentional response to

novel print (false font string minus word contrast) as shown in

Figure 4a. Also shown are axial slices with activation that sur-

vived correction for multiple comparisons in the traditional

functional image analysis (cluster threshold of P = 0.05 as

evaluated with FSL). Activation of the occipital region is highly

significant reflecting greater activity to false font strings than

real words, as would be expected given that false font strings

are more novel in visual form than real words and thus may

elicit more activation of processing streams important for

distinguishing visual forms. Fronto-parietal activation is also

observed, which survived correction for multiple comparisons

Figure 3. Regions of interest used in the permutation analyses. Lateral regions are color coded as follows: ventral frontal, yellow; dorsal frontal, pink; temporal, dark blue;occipital, green; parietal, light blue; perisylvian, brick red (created from a statistical map published previously; Sowell, Thompson, Leonard, et al. 2004). Medial regions are colorcoded as follows: dorsal frontal, purple; ventral frontal, olive green; parietal, dark blue; occipital, red; callosal subcortical area (not tested in permutations), white.

2598 Brain Activation--Structure Relationships in Children d Lu et al.

on the right but not on the left. These fronto-parietal regions

are part of an attentional network that generates top-down

influence on the visual cortex (Kastner and Ungerleider 2000),

and thus has been posited to underlie orienting of attention to

visual tasks (Fan et al. 2005; Raz and Buhle 2006). Significant

activation during implicit reading (word minus false contrast)

is observed in the left inferior frontal gyrus and survived cor-

rection for multiple comparisons in the traditional functional

image analysis as seen on horizontal slices in Figure 4b.

Activation in left inferior parietal and posterior temporal regions

is subthreshold and observed only with uncorrected t-maps on

the cortical surface renderings (Fig. 4b). These results are

generally consistent with Turkeltaub et al.’s findings using this

same orthographic processing task, where both left inferior

Figure 4. Functional activation of response to novel print and implicit reading. (a) An uncorrected statistical map of response to novel print (false font-word) is rendered on thecortical surface, and activation that passed correction for multiple comparisons (cluster threshold set at Z[ 1.7, P 5 0.05) is rendered on axial slices. Regions responding tonovel print more than real words included right parietal (BA 7, 19), right frontal (BA 9, 10), right inferior temporal (BA 37), and bilateral occipital (BA 18) regions. (b) Anuncorrected statistical map of implicit reading (word-false font) is rendered on the cortical surface, and activation that passed correction for multiple comparisons (clusterthreshold set at Z[ 1.7, P5 0.05) is rendered on axial slices. Regions responding more to real words than novel print included the left inferior frontal gyrus (BA 44, 45, 47). Leftposterior temporal and inferior parietal activation (BA 22, 40) evident on the surface rendering but not on the axial slices indicated subthreshold activation in those regions.

Cerebral Cortex November 2009, V 19 N 11 2599

Figure 5. Relationships between activation, gray matter thickness, and naming speed. For all surface maps (a, c, d, and e), cortical surface points with statistically significantPearson’s r correlation coefficients are differentiated from points with nonsignificant values (gray) by color coding, and 3 different levels of statistical significance (P# 0.05, 0.01,and 0.005) are rendered. Relationship between activation and gray matter thickness. (a) Negative correlations (i.e., red, orange, yellow) indicate that those with thinner cortexactivated the fronto-parietal network more in response to novel print. (b) Scatter plots of 3 right parietal surface points (averaged) show that response to novel print (false

2600 Brain Activation--Structure Relationships in Children d Lu et al.

frontal and posterior temporal activation were found in adults

but activation in only one of these regions passed threshold in

children (Turkeltaub et al. 2003).

Activation and Gray Matter Thickness

T-statistic maps representing attentional response to novel

print for each subject were correlated with gray matter

thickness values for each subject at each cortical surface point

(anatomically matched across subjects and between activation

and thickness maps within subjects). As shown in Figure 5a,

greater attentional response to novel print than real words is

associated with thinner cortex in bilateral fronto-parietal

networks. Permutation testing was conducted within regions

of interest to correct for multiple comparisons, and significant

negative relationships in the right dorsal frontal and parietal

regions were unlikely to be due to chance (Table 2). The

normal maturational trajectory of cortex within dorsal parietal

and frontal regions is gray matter thinning during the age range

that we studied (Sowell et al. 2003; Gogtay et al. 2004; Sowell,

Thompson, Leonard, et al. 2004), and the present sample of

subjects conforms to this pattern (Fig. 6). Therefore, more

‘‘mature’’ individuals (i.e., subjects with thinner cortex in these

regions) activate the fronto-parietal attentional network more

than less ‘‘mature’’ individuals (i.e., subjects with thicker cortex

in these regions) when processing novel print compared to real

words. In contrast, there was no significant positive relation-

ship between activation and gray matter thickness in the

perisylvian region (Fig. 5a). Lack of gray matter thickening in

the perisylvian region of the present sample of subjects (Fig. 6)

may have contributed to decreased ability to detect activation--

morphological relationships in this region.

Because attentional network activation in our experiment is

based on activation during one condition (false font strings)

relative to another (real words), the relative contribution of

each condition must be evaluated for interpretation. Scatter

plots for activation and thickness values for the average of 3

cortical surface points chosen from right parietal regions

where correlations between thickness and activation were

significant (from Fig. 5a) are plotted in Figure 5b. These plots

show that the significant correlation between thickness and

attentional network activation (false font minus word) is driven

more by false font strings (false font minus rest, r = –0.36) than

by real words (word minus rest, r = 0.15).

Table 2Permutation test results showing the probability that the number of significant correlations (at P # 0.05) within each ROI occurred by chance

Lateral ROI Response to novelprint and gray matter thickness

Response to novel printand naming speed

Response to novel printand naming speed with age regressed

Response to novel print andnaming speed with IQ regressed

Left Right Left Right Left Right Left Right

Dorsal frontal ns 0.0246 0.0022 0.0031 0.0326 ns 0.0108 0.0111Ventral frontal ns ns 0.0121 0.0055 ns ns 0.0235 0.0249Parietal ns 0.0091 0.0075 0.0081 ns ns 0.0257 0.0177Occipital ns ns ns ns ns ns ns nsTemporal ns ns ns ns ns ns ns nsPerisylvian ns ns 0.0198 0.0014 ns 0.0278 0.0403 0.0043

Note: ns 5 not significant.

Figure 6. Correlation between age and gray matter thickness. Cortical surface points with statistically significant Pearson’s r correlation coefficients are differentiated from pointswith nonsignificant values (gray) by color coding, and 3 different levels of statistical significance (P # 0.05, 0.01, and 0.005) are rendered. Results conform to documentedpattern of cortical thinning in the dorsal fronto-parietal region among larger subject groups with more extended age ranges than in the present study; however, the present studysample did not show cortical thickening in the perisylvian region as others have reported (Sowell et al. 2003; Gogtay et al. 2004; Sowell, Thompson, Leonard, et al. 2004).

font-word) is mainly driven by the false font string condition (false-rest) and not by the word condition (word-rest). Relationship between activation and naming speed. (c)Negative correlations indicate that individuals with faster-naming speed activated diverse bilateral regions in response to novel print, including the fronto-parietal attentionnetwork. After effects of age (d) and general cognitive development (e) were statistically removed, naming speed still correlated significantly with the fronto-parietal network’sresponse to novel print. Naming speed predicted activation of the fronto-parietal attention network independently of age or IQ.

Cerebral Cortex November 2009, V 19 N 11 2601

Activation and Naming Speed

Activation representing attentional response to novel print (false

font minus word) was correlated with naming speed to identify

the activation pattern associated with more ‘‘mature’’ skill (i.e.,

faster-naming speed). Resulting statistical maps (Fig. 5c) bear

spatial correspondence to the activation--thickness map in

fronto-parietal regions (Fig. 5a). Naming speed was measured

by time, and larger values represented slower, or worse,

performance. Negative relationships between activation and

naming speed indicate that those with more ‘‘mature’’ skill

(faster) activated the fronto-parietal attentional network more in

response to novel print. Similar relationships are also found in

bilateral perisylvian and inferior frontal regions, where more

‘‘mature’’ skill presumably corresponds to other functional

demands of processing false font strings compared to real words.

Parceling Skill Level from Age and General CognitiveDevelopment

Both activation--thickness (Fig. 5a) and activation--performance

(Fig. 5c) maps indicate greater activation of the fronto-parietal

network in response to novel print among individuals with

more ‘‘mature’’ skills. Because older subjects were faster at

naming (r = –0.71, P < 0.001), it was possible that relationships

between activation and performance were spuriously corre-

lated through shared variance with age. Thus, we re-evaluated

the relationship between naming speed and activation using

age as a covariate. Resulting statistical maps (Fig. 5d) show

regions where naming speed predicted activation in response

to novel print independent of age effects. Skill in naming letters

is accounting for some of the variance in activation of the

fronto-parietal network in response to novel print and the

shared age variance does not completely account for the ob-

served relationships between activation and performance.

To examine whether the correlation between activation and

namingspeed is specific tonamingskill level rather thana reflection

of general cognitive development, we used FSIQ as a covariate in

a separate analysis. Resulting statistical maps (Fig. 5e) look

remarkably similar to Figure 5c, indicating that general intellectual

functioning does not completelymediate the relationship between

naming speed and activation in response to novel print.

Discussion

By integrating functional activity, structural morphology, and

cognitive skill performance, we were able to identify brain

regions where functional activation during orthographic

processing was related to cortical thickness. The correlation

between activation and naming speed (shown in Fig. 5c)

allowed us to determine that greater difference in activation

between false font strings and words reflects a more ‘‘mature’’

pattern of activation. This ‘‘mature’’ pattern of activation was

associated with thinner cortex in the right fronto-parietal

region (Fig. 5a), where cortex thins during childhood and

adolescence (Sowell et al. 2003; Gogtay et al. 2004). These data

are consistent with the notion that more ‘‘mature’’ cortical

thickness patterns (i.e., thinner) are associated with more

‘‘mature’’ activation patterns. This relationship was preserved in

the fronto-parietal attention network even after controlling for

age or IQ.

One assumption of the present study is that the ortho-

graphic processing task used is facilitated by naming skills. The

faster children are at naming letters, the easier it should be to

decide if a tall letter is in the stimulus. Naming speed is

a learned skill but it also correlates with age. Coupling of age

and skill level confounds activation due to genetic influences

that unfold with age and activation associated with higher skill

level. We attempted to dissociate the 2 by using multiple

regression to examine effects of skill level while statistically

controlling for age effects. Results showed that activation in the

fronto-parietal attention network was still associated with

naming speed once the effects of age were covaried out. That

is, chronological age does not fully explain the relationship

between thinning fronto-parietal cortex and faster-naming

speed. As both age and experience with print contribute to

naming speed, it is tempting to speculate that the relationship

between the fronto-parietal attention network and naming

speed reflects experience effects. However, the present study

design does not allow for such inference because ‘‘experience’’

was neither quantified nor measured. We are not the only

group to have parsed out age effects from activation--skill level

relationships. Our findings are consistent with Schlaggar and

colleagues’ reports of functional activation that varies

independently of age and is performance related (Schlaggar

et al. 2002; Brown et al. 2005).

In a separate analysis, we controlled for IQ while examining

the relationship between activation and skill level. It is possible

that individuals with higher intelligence have faster-naming

speed and that activation--naming speed relationship found in

fronto-parietal regions is driven by effects of intelligence.

Activation of the fronto-parietal attention network was

associated with naming speed independent of IQ. This finding

is important because it suggests that general cognitive ability

does not fully explain activation in the fronto-parietal attention

network during orthographic processing. Faster-naming indi-

viduals activate this fronto-parietal attention network more

regardless of effects of general cognitive ability.

The fronto-parietal attention network is involved in both overt

and covert orienting of attention to spatial location (Fan et al.

2005; Raz and Buhle 2006). As reviewed by Kastner and

Ungerleider (2000), top-down bias from the attention system to

visual processing includes enhancement of neural responses to

an attended stimulus and increasing stimulus salience. Function-

al connectivity of the superior parietal lobule and frontal eye

fields put them in a position to serve as sources of top-down

biasing signals to visual processing streams (Kastner and

Ungerleider 2000). The strongest determinant of neural respon-

siveness of the lateral intraparietal area turns out tobe the salience

of the stimulus (Colby and Goldberg 1999). The most parsimo-

nious explanation of enhanced activation of this fronto-parietal

attention network among individuals with more advanced skills,

after controlling for age or general cognitive ability, is that false

font strings appear more salient than real words to those with

more advanced naming skill. Naming speed can be enhanced by

experience, so sensitivity of this parietal--frontal attention

network to false font strings may be a specific consequence of

experience. In other words, false font strings and real words may

be equally novel to slow-naming individuals (i.e., less experience

with print, after accounting for their age or IQ) resulting in less

difference in activation between the 2 conditions. But among

fast-naming individuals (i.e., more experience with print, after

controlling for age or IQ), word stimuli were likely quite familiar

and only the false font strings were novel and salient, resulting

in greater activation of fronto-parietal attention networks to the

false font strings relative to real words.

2602 Brain Activation--Structure Relationships in Children d Lu et al.

The functional task used here was initially designed to elicit

implicit reading (Turkeltaub et al. 2003). We did not observe

statistically significant relationships between cortical thick-

ness and implicit reading activation in classical brain language

regions. Cortical thickness in these regions have been shown

to increase with age into young adulthood, but variability is

much higher and effects much smaller than the thinning that

occurs in frontal and parietal cortices during this age range

(Sowell et al. 2003). Cortical thickening was not observed in

these regions in the present sample, perhaps due to limited

cross-sectional sample size or a relatively restricted age range.

Activation related to the implicit reading aspects of the task

may be less robust than activation related to attentional

aspects of the task, as suggested by the observation that

activation in posterior perisylvian language regions did not

survive correction for multiple comparisons. That reading

requires attentional resources is not a novel finding, as

Turkeltaub et al. (2003) also reported activation in fronto-

parietal attention areas for words and false font strings relative

to rest using the same task. Our study highlights that the

difference in activation pattern of children from that of adults

reported in other studies may not reflect differences in the

targeted cognitive system under study, but may in fact reflect

supportive cognitive systems required to complete the task.

‘‘Mature’’ activation patterns for a given task may involve

supportive systems as much as the target system, consistent

with other developmental fMRI reports (reviewed in Durston

and Casey 2006).

The current results are consistent with the notion that the

protracted course of functional activation development in the

human brain is associated with skill level, at least in fronto-

parietal attention networks, and activation corresponds to more

‘‘mature’’ morphology in these regions. From this, we posit that

structural brain development may also be related to skill level

and not merely to genetic influences that unfold with age. There

may be relationships between mature patterns of activation and

morphology in other regions of the brain that we lacked

sufficient power to detect. The current study is a correlational

study and cannot address if structural maturation allows for

learning to take place, or if experience leads to morphological

maturation. Nevertheless, integrating structural MRI with

functional MRI and neurocognitive performance holds tremen-

dous promise in elucidating structure--performance relationships

of high-level cognitive functions specific to humans.

Funding

National Institute of Drug Abuse Grants (R21 DA15878

and RO1 DA017831) awarded to E.R.S.; National Institute of

Health (NIH)/National Center for Research Resources grant

(P41 RR013642); National Institute of Neurological Disorders

and Stroke grant (NS3753) awarded to A.W.T; grants

(AG016570, LM05639, EB01651, and RR019771) supported

P.M.T.; and NIH Roadmap for Medical Research grant (U54

RR021813 entitled Center for Computational Biology).

Notes

We thank Guinevere Eden’s research group for sharing the stimuli used

in the orthographic processing task. Conflict of Interest : None declared.

Address correspondence to Elizabeth R. Sowell, PhD, UCLA

Laboratory of Neuro Imaging, Department of Neurology, 635 Charles

Young Drive South, Suite 225, Los Angeles, CA 90095, USA. Email:

esowell@ucla.edu.

References

Aganj I, Sapiro G, Parikshak N, Madsen SK, Thompson PM. 2008.

Measurement of cortical thickness from MRI by minimum line inte-

grals on soft-classified tissue. Human Brain Mapping. Forthcoming.

Beckmann CF, Jenkinson M, Smith SM. 2003. General multilevel linear

modeling for group analysis in FMRI. Neuroimage. 20:1052--1063.

Booth JR, Burman DD, Meyer JR, Gitelman DR, Parrish TB, Mesulam MM.

2004. Development of brain mechanisms for processing ortho-

graphic and phonologic representations. J Cogn Neurosci. 16:

1234--1249.

Brown TT, Lugar HM, Coalson RS, Miezin FM, Petersen SE, Schlaggar BL.

2005. Developmental changes in human cerebral functional

organization for word generation. Cereb Cortex. 15:275--290.

Colby CL, Goldberg ME. 1999. Space and attention in parietal cortex.

Annu Rev Neurosci. 22:319--349.

Durston S, Casey BJ. 2006. What have we learned about cognitive

development from neuroimaging? Neuropsychologia. 44:2149--2157.

Durston S, Davidson MC, Tottenham N, Galvan A, Spicer J, Fossella JA,

Casey BJ. 2006. A shift from diffuse to focal cortical activity with

development. Dev Sci. 9:1--8.

Duvernoy HM, Cabanis EA, Vannson JL. 1991. The human brain: surface,

three-dimensional sectional anatomy and MRI. Vienna, Austria:

Springer-Verlag.

Evans AC, Kamber M, Collins DL, MacDonald D. 1994. An MRI-based

probabilistic atlas of neuroanatomy. In: Shorvon SD, editor. Magnetic

resonance scanning and epilepsy. New York: Plenum. p. 263--274.

Fan J, McCandliss BD, Fossella J, Flombaum JI, Posner MI. 2005. The

activation of attentional networks. Neuroimage. 26:471--479.

Fuster JM. 2002. Frontal lobe and cognitive development. J Neurocytol.

31:373--385.

Gaillard WD, Hertz-Pannier L, Mott SH, Barnett AS, LeBihan D,

Theodore WH. 2000. Functional anatomy of cognitive development:

fMRI of verbal fluency in children and adults. Neurology. 54:

180--185.

Gaillard WD, Sachs BC, Whitnah JR, Ahmad Z, Balsamo LM, Petrella JR,

Braniecki SH, McKinney CM, Hunter K, Xu B, et al. 2003.

Developmental aspects of language processing: fMRI of verbal

fluency in children and adults. Hum Brain Mapp. 18:176--185.

Gogtay N, Giedd JN, Lusk L, Hayashi KM, Greenstein D, Vaituzis AC,

Nugent TF, 3rd, Herman DH, Clasen LS, Toga AW, et al. 2004.

Dynamic mapping of human cortical development during childhood

through early adulthood. Proc Natl Acad Sci USA. 101:8174--8179.

Holland SK, Plante E, Weber Byars A, Strawsburg RH, Schmithorst VJ,

Ball WS, Jr. 2001. Normal fMRI brain activation patterns in children

performing a verb generation task. Neuroimage. 14:837--843.

Jenkinson M, Bannister P, Brady M, Smith S. 2002. Improved

optimization for the robust and accurate linear registration and

motion correction of brain images. Neuroimage. 17:825--841.

Jenkinson M, Smith S. 2001. A global optimisation method for robust

affine registration of brain images. Med Image Anal. 5:143--156.

Kastner S, Ungerleider LG. 2000. Mechanisms of visual attention in the

human cortex. Annu Rev Neurosci. 23:315--341.

Lu L, Leonard C, Thompson P, Kan E, Jolley J, Welcome S, Toga A,

Sowell E. 2007. Normal developmental changes in inferior frontal gray

matter are associated with improvement in phonological processing:

a longitudinal MRI analysis. Cereb Cortex. 17:1092--1099.

MacDonald D, Avis D, Evans A. 1994. Multiple surface identification and

matching in magnetic resonance images. Proc Vis Biomed Comput.

2359:160--169.

Mazziotta JC, Toga AW, Evans A, Fox P, Lancaster J. 1995. A probabilistic

atlas of the human brain: theory and rationale for its development.

The International Consortium for Brain Mapping (ICBM). Neuroimage.

2:89--101.

Nichols T, Holmes A. 2004. Nonparametric permutation tests for

functional neuroimaging. In: Frackowiak RSJ, Friston KJ, Frith CD,

Dolan RJ, Price CJ, Zeki S, Ashburner J, Penny W, editors. Human

brain function. 2nd edn. New York: Elsevier. p. 887--910.

Oldfield RC. 1971. The assessment and analysis of handedness: The

Edinburgh inventory. Neuropsychologia. 9:97--113.

Ono M, Kubik S, Abernathey CD. 1990. Atlas of the cerebral sulci.

Stuttgart: G. Thieme Verlag.

Cerebral Cortex November 2009, V 19 N 11 2603

Ratnanather JT, Wang L, Nebel MB, Hosakere M, Han X, Csernansky JG,

Miller MI. 2004. Validation of semiautomated methods for quanti-

fying cingulate cortical metrics in schizophrenia. Psychiatry Res.

132:53--68.

Raz A, Buhle J. 2006. Typologies of attentional networks. Nat Rev

Neurosci. 7:367--379.

Rivera SM, Reiss AL, Eckert MA, Menon V. 2005. Developmental changes

in mental arithmetic: evidence for increased functional specializa-

tion in the left inferior parietal cortex. Cereb Cortex. 15:1779--1790.

Sapiro G. 2001. Geometric partial differential equations and image

analysis. Cambridge (UK): Cambridge University Press.

Schlaggar BL, Brown TT, Lugar HM, Visscher KM, Miezin FM,

Petersen SE. 2002. Functional neuroanatomical differences between

adults and school-age children in the processing of single words.

Science. 296:1476--1479.

Shattuck DW, Sandor-Leahy SR, Schaper KA, Rottenberg DA, Leahy RM.

2001. Magnetic resonance image tissue classification using a partial

volume model. Neuroimage. 13:856--876.

Shaw P, Greenstein D, Lerch J, Clasen L, Lenroot R, Gogtay N, Evans A,

Rapoport J, Giedd J. 2006. Intellectual ability and cortical

development in children and adolescents. Nature. 440:676--679.

Smith SM. 2002. Fast robust automated brain extraction. Hum Brain

Mapp. 17:143--155.

Sowell ER, Peterson BS, Thompson PM, Welcome SE, Henkenius AL,

Toga AW. 2003. Mapping cortical change across the human life

span. Nat Neurosci. 6:309--315.

Sowell ER, Thompson PM, Holmes CJ, Batth R, Jernigan TL, Toga AW.

1999. Localizing age-related changes in brain structure between

childhood and adolescence using statistical parametric mapping.

Neuroimage. 9:587--597.

Sowell ER, Thompson PM, Holmes CJ, Jernigan TL, Toga AW. 1999. In

vivo evidence for post-adolescent brain maturation in frontal and

striatal regions. Nat Neurosci. 2:859--861.

Sowell ER, Thompson PM, Leonard CM, Welcome SE, Kan E, Toga AW.

2004. Longitudinal mapping of cortical thickness and brain growth

in normal children. J Neurosci. 24:8223--8231.

Sowell ER, Thompson PM, Rex D, Kornsand D, Tessner KD, Jernigan TL,

Toga AW. 2002. Mapping sulcal pattern asymmetry and local cortical

surface gray matter distribution in vivo: maturation in perisylvian

cortices. Cereb Cortex. 12:17--26.

Sowell ER, Thompson PM, Tessner KD, Toga AW. 2001. Mapping

continued brain growth and gray matter density reduction in dorsal

frontal cortex: Inverse relationships during postadolescent brain

maturation. J Neurosci. 21:8819--8829.

Sowell ER, Thompson PM, Toga AW. 2004. Mapping changes in the

human cortex throughout the span of life. Neuroscientist.

10:372--392.

Sprenger-Charolles L, Siegel LS, Bechennec D, Serniclaes W. 2003.

Development of phonological and orthoraphic processing in

reading aloud, in silent reading, and in spelling: a four-year

longitudinal study. J Exp Child Psychol. 84:194--217.

Thomas KM, King SW, Franzen PL, Welsh TF, Berkowitz AL, Noll DC,

Birmaher V, Casey BJ. 1999. A developmental functional MRI study

of spatial working memory. Neuroimage. 10:327--338.

Thompson PM, Hayashi KM, de Zubicaray G, Janke AL, Sowell ER,

Rose SE, Semple J, Herman D, Hong MS, Dittmer SS, et al. 2003.

Dynamic mapping of Alzheimer’s disease. Proceedings of the 19th

Colloque Medecine et Recherche. IPSEN Foundation. Paris (France):

Springer-Verlag.

Thompson PM, Hayashi KM, Sowell ER, Gogtay N, Giedd JN,

Rapoport JL, de Zubicaray GI, Janke AL, Rose SE, Semple J, et al.

2004. Mapping cortical change in Alzheimer’s disease, brain

development, and schizophrenia. Neuroimage. 23(Suppl. 1):S2--18.

Thompson PM, Mega MS, Vidal C, Rapoport JL, Toga AW, editors. 2001.

Detecting disease-specific patterns of brain structure using cortical

pattern matching and a population-based probablistic brain atlas.

New York: Springer-Verlag.

Turkeltaub PE, Gareau L, Flowers DL, Zeffiro TA, Eden GF. 2003.

Development of neural mechanisms for reading. Nat Neurosci.

6:767--773.

Von Economo CV. 1929. The cytoarchitectonics of the human cerebral

cortex. London: Oxford Medical Publications.

Wagner RK, Torgesen JK, Rashotte CA. 1999. Comprehensive test of

phonological processing. Austin (TX): Pro-Ed.

Wagner RK, Torgesen JK, Rashotte CA, Hecht SA, Barker TA, Burgess SR,

Donahue J, Garon T. 1997. Changing relations between phonolog-

ical processing abilities and word-level reading as children develop

from beginning to skilled readers: a 5-year longitudinal study. Dev

Psychol. 33:468--479.

Wechsler D. 2003. Wechsler Intelligence Scale for Children. San

Antonio (TX): The Psychological Corporation.

Wolf M, Bowers PG. 1999. The double-deficit hypothesis for the

developmental dyslexias. J Educ Psychol. 91:415--438.

Woods RP, Mazziotta JC, Cherry SR. 1993. MRI-PET registration with

automated algorithm. J Comput Assist Tomogr. 17:536--546.

Woolrich MW, Behrens TE, Beckmann CF, Jenkinson M, Smith SM. 2004.

Multilevel linear modelling for FMRI group analysis using Bayesian

inference. Neuroimage. 21:1732--1747.

Woolrich MW, Ripley BD, Brady M, Smith SM. 2001. Temporal

autocorrelation in univariate linear modeling of FMRI data. Neuro-

image. 14:1370--1386.

Worsley KJ, Evans AC, Marrett S, Neelin P. 1992. A three-dimensional

statistical analysis for CBF activation studies in human brain. J Cereb

Blood Flow Metab. 12:900--918.

2604 Brain Activation--Structure Relationships in Children d Lu et al.