Neuron

Primer

Surface-Based and Probabilistic Atlasesof Primate Cerebral Cortex

David C. Van Essen1,* and Donna L. Dierker1

1Department of Anatomy and Neurobiology, Washington University School of Medicine, St. Louis, MO 63110, USA*Correspondence: vanessen@brainvis.wustl.eduDOI 10.1016/j.neuron.2007.10.015

Brain atlases play an increasingly important role in neuroimaging, as they are invaluable for analysis,visualization, and comparison of results across studies. For both humans and macaque monkeys,digital brain atlases of many varieties are in widespread use, each having its own strengths and lim-itations. For studies of cerebral cortex there is particular utility in hybrid atlases that capitalize on thecomplementary nature of surface and volume representations, are based on a population averagerather than an individual brain, and include measures of variation as well as averages. Linking differ-ent brain atlases to one another and to online databases containing a growing body of neuroimagingdata will enable powerful forms of data mining that accelerate discovery and improve researchefficiency.

IntroductionStaggering amounts of experimental data pertaining to

brain structure and function have been obtained in recent

years using a variety of neuroimaging methods. Structural

MRI (sMRI) and functional MRI (fMRI) are especially widely

used because they allow concurrent visualization of brain

structure and function at high spatial resolution. Powerful

ways to analyze, visualize, and access neuroimaging data

are available and are rapidly evolving. Digital brain atlases

are a key component of this growing arsenal, as they pro-

vide an objective and accessible spatial framework for

representing complex experimental data sets. This review

focuses on surface-based atlases of cerebral cortex in

primates, especially humans. Cerebral cortex—the domi-

nant structure of the human brain—poses special chal-

lenges because it is highly convoluted and because the

pattern of folding varies greatly from one individual to the

next. Surface-based methods of visualization and analysis

(‘‘cortical cartography’’) are key to dealing with these

cortical convolutions, but they are most useful when asso-

ciated with complementary volumetric atlases.

In general, a brain atlas is a representation of anatomical

structure and other reference information in a spatial

framework that provides a useful repository of knowledge

and facilitates the analysis of spatially localized experi-

mental data of many types. Digital brain atlases have many

advantages over conventional print atlases, primarily be-

cause they are interactive, searchable, and extensible. In-

teractive refers to the ability to navigate quickly and seam-

lessly through complex data sets, including options to

view brain structure from many perspectives and to con-

trol what types of information are overlaid on the basic

brain anatomy. Searchable refers to options for finding rel-

evant data based on spatial coordinates, structural and

functional labels, and a variety of other search criteria. Ex-

tensible refers to options for incorporating new informa-

tion of diverse types into the atlas quickly and flexibly,

without needing to await a new print edition.

The five main objectives of this review are (1) to discuss

key structural and functional characteristics of human

cerebral cortex that profoundly impact the analysis of neu-

roimaging data and dictate desirable features of a cortical

atlas; (2) to characterize major brain atlases currently used

in human and monkey neuroimaging; (3) to illustrate the

utility of digital atlases for visualization, analysis, localiza-

tion, and data mining; (4) to compare strategies used to

compensate for individual variability and to relate these

strategies to the biological basis of variability; and (5) to

illustrate comparisons between human and macaque cor-

tex using surface-based atlases. Other recent reviews

cover a broader range of atlas-related issues (Devlin and

Poldrack, 2007; Mazziotta et al., 2001; Toga and Thomp-

son, 2001, 2002, 2003, 2005; Toga et al., 2006).

1. Individual Variability Is Large Relative toCortical Area DimensionsDeciphering the amazingly complex functional organiza-

tion of cerebral cortex requires that experimental data be

localized as accurately as possible within the convoluted

cortical sheet. It is equally important to recognize the

uncertainties and errors that inevitably occur when spec-

ifying cortical locations. To help frame these issues, it is in-

structive to consider the following questions about human

cerebral cortex. What are the dimensions of the cortex

and its functional subdivisions? What are the nature and

magnitude of individual variability? What constraints are

imposed by the spatial resolution and signal-to-noise

routinely attainable in neuroimaging studies?

Human cerebral cortex is a thin sheet (3 mm thick on av-

erage) with a surface area of �900 cm2 per hemisphere—

equivalent to a 13’’ (34 cm) pizza (Fischl and Dale, 2000;

Henery and Mayhew, 1989; Jouandet et al., 1989; Makris

Neuron 56, October 25, 2007 ª2007 Elsevier Inc. 209

Neuron

Primer

Figure 1. Shape Characteristics ofCerebral Cortex in an Individual HumanSubject(A) A parasagittal section through the righthemisphere structural MRI (sMRI). Red contourshows a slice through the fiducial surface;contour thickness varies according to howobliquely the surface is sliced.(B) The right hemisphere fiducial surface gen-erated using the SureFit algorithm in Caret,which provides an approximation to the corti-cal midthickness (layer 4).(C)Avery inflatedsurface inwhichshape charac-teristics are represented by a map of sulcaldepth (distance ofeachsurfacenodeto the near-est point in the ‘‘cerebral hull’’ (gyral crowns).(D) A flat map representation that shows theentire hemisphere in a single view.Data are accessible via http://sumsdb.wustl.edu/sums/directory.do?id=6650508.

et al., 2005; Tramo et al., 1995; Van Essen, 2005a). Exten-

sive convolutions allow this large cortical expanse to fit

into a compact cerebral volume (about 18 cm long, 13 cm

high, 14 cm total brain width). These relationships are illus-

trated for an exemplar brain (sMRI slice in Figure 1A) and

the right hemisphere surface displayed in its original 3D

(fiducial) configuration (Figure 1B). A map of ‘‘sulcal depth’’

(distance to the nearest gyrus) provides a useful measure

of the original shape when viewing the inflated and flat-

tened configurations (Figures 1C and 1D).

The cortical sheet contains a complex mosaic of cortical

areas, each having distinct anatomical and functional

characteristics. Accurate partitioning of the entire cortex

has proven difficult, mainly because the differences be-

tween areas are often subtle. Competing partitioning

schemes remain in use for most regions, leading to fre-

quent debate and confusion, and the total number of cor-

tical areas is not known for any species. Human cortex

probably contains between 100 and 200 areas in each

hemisphere (Van Essen, 2004b) arranged in a pattern that

has strong bilateral symmetry but some important hemi-

spheric asymmetries, especially in the temporal lobe (Toga

and Thompson, 2003; Van Essen, 2005a). Assuming an

intermediate value of 150 areas per hemisphere, each indi-

vidual area would occupy�6 cm2 surface area on average,

equivalent to a pepperoni-sized disk �3 cm in diameter.

A few areas are much larger (area V1 is �20 cm2 on aver-

age), while others are much smaller. Many areas are elon-

gated and as little as 5–10 mm wide, such as the archi-

tectonic areas identified in orbitofrontal cortex (Ongur

et al., 2003).

Spatial resolution in neuroimaging is limited by the size

of the individual ‘‘voxels’’ (volume elements) acquired in

each scan. Voxel dimensions are typically 1 mm3 for sMRI

(as in Figure 1A) and 3 3 3 3 3 mm or larger for human

fMRI. A typical cortical area with a volume of �1800

mm3 (�600 mm2 area 3 3 mm thickness), would thus be

equivalent to about 70 fMRI voxels in overall extent; small

or narrow areas are only a few voxels wide.

It is possible to identify and map individual cortical areas

using fMRI. The most notable successes involve single-

210 Neuron 56, October 25, 2007 ª2007 Elsevier Inc.

subject mapping of many retinotopic visual areas (V1,

V2, V3, etc.; DeYoe et al., 1996; Grill-Spector and Malach,

2004; Larsson and Heeger, 2006; Sereno et al., 1995;

Swisher et al., 2007; Wandell et al., 2005, 2007; but see

Jack et al., 2007). Extending this approach to obtain an

accurate, fine-grained mapping of the entire human cortex

is an extremely desirable objective, but an elusive one for

several reasons. The signal-to-noise ratio obtained with

fMRI is generally low, especially in regions outside early

sensory and motor areas. Signal-to-noise can be en-

hanced by within-subject spatial smoothing and by intra-

subject averaging, but each of these substantially de-

grades spatial resolution. Averaging across subjects

entails registering the data to a common spatial frame-

work, i.e., a brain atlas. The nature and magnitude of indi-

vidual variability pose major challenges, making it critical

to estimate the uncertainties and alignment errors associ-

ated with different atlases and registration strategies.

Four aspects of individual variability impact cortical reg-

istration. (1) Brain size and shape. Brains differ in total vol-

ume, linear dimensions, and overall shape within the cra-

nial cavity. (2) Folding patterns. Humans and other highly

gyrencephalic species show dramatic individual differ-

ences in the specific pattern of convolutions. This is illus-

trated with medial views of two example right hemisphere

surfaces shown in their original 3D (‘‘fiducial’’) configura-

tion and on inflated surfaces (Figures 2A and 2B). Three

major sulci, the calcarine (CaS), parieto-occipital (POS),

and cingulate (CiS) can be readily identified, but differ-

ences in shape and location are evident, particularly for

the calcarine sulcus. Importantly, the assorted local fea-

tures are the most variable. For example, the blue arrows

in Figures 2A and 2B point to small folds in one hemi-

sphere that are absent or different in orientation in the

other. Each ‘‘cortical brainprint’’ represented by one of

these sulcal depth maps is likely to be as unique as a hu-

man fingerprint pattern. (3) Areal size. Any given area dif-

fers in size by a factor of 2-fold or more across individuals.

This has been documented most extensively for area V1

(Andrews et al., 1997; Stensaas et al., 1974) but has

been demonstrated for other areas as well (Dougherty

Neuron

Primer

Figure 2. Medial Views of the Right Hemisphere in Two Individuals(A) The fiducial surface (top) and inflated surface (bottom) of Case A, with major sulci identified in the lower panel.(B) The fiducial surface (top) and inflated surface (bottom) of Case B. Blue arrows (top panels) show local features that are different in the twohemispheres or present in one but not the other. Insets show rotated views of the occipital pole, illustrating the ambiguous definition of the tip ofthe calcarine sulcus.(C and D) Lateral inflated views of left hemispheres of Cases (C) and (D), with blue arrows showing locations where intersubject correspondence isambiguous.Data are accessible via http://sumsdb.wustl.edu/sums/directory.do?id=6650508.

et al., 2003; Eickhoff et al., 2006). The variability of individ-

ual areas exceeds the estimates of variability in total cor-

tical size and surface area (Andrews et al., 1997; Elston,

2006; Henery and Mayhew, 1989; A. Kline et al., 2005,

Org. Hum. Brain Mapp., abstract). (4) Area versus folding

landmarks. Area V1 occupies most of the calcarine sulcus,

but a variable portion of it extends into neighboring gyral

and sulcal regions by up to several cm (Amunts et al.,

2000; Rademacher et al., 1993). For most other cortical

areas the correlation with cortical folding is even weaker

(Amunts et al., 2007). In considering the impact of these

different aspects of variability, it becomes important to

clarify the concept of ‘‘corresponding’’ locations in differ-

ent brains.

Functional and Geographic Correspondences Are

Conceptually and Empirically Distinct

The overarching objective when registering individuals to

an atlas is to align corresponding cortical locations as ac-

curately as possible. Cortical locations in different individ-

uals are in functional correspondence if they are part of the

same cortical area and are matched in terms of whatever

is mapped within that area. For example, each point in the

visual field (e.g., the center of the foveal representation)

can be used to define corresponding locations in area

V1 of different individuals. In a similar vein, locations can

be considered in geographic correspondence if they rep-

resent the same ‘‘geographic’’ feature (so named because

gyri and sulci are akin to hills and valleys on the earth’s

surface) or have a consistent relationship to other geo-

graphic features. For example, the posterior tip of the cal-

carine sulcus can be considered geographically corre-

sponding in different individuals.

Three important caveats arise in applying these con-

cepts to the realities of cortical structure and function.

(1) Function-folding mismatches. Because areal bound-

aries vary relative to geographic landmarks, functional

and geographic correspondences are inherently in con-

flict. The question is not whether they are mismatched,

but by how much. The answers depend on the areas con-

sidered and on the particular individuals analyzed. For ex-

ample, the center of the foveal representation might lie

near the posterior tip of the calcarine sulcus in one individ-

ual and 3 cm lateral to it in another individual. When such

mismatches occur, either the local visuotopic maps or the

local folds can be aligned, but not both concurrently. (2)

Ambiguous geographic correspondences. Variability in

folding patterns can lead to ambiguity in the notion of geo-

graphic correspondence. For example, in Figures 2C and

2D (lateral views of inflated surfaces) the dorsal tip of the

superior temporal sulcus (STS) is well-defined in Case C,

but in Case D it has a distinct fork; the choice of which

branch should be designated is the ‘‘real’’ dorsal tip is

largely arbitrary. Similarly, the ventral-anterior tip of the

STS is well defined in Case D but is ambiguous in Case

C. (3) Folding mismatches. Given the observed degree of

variability in cortical folding patterns, the challenges in

registration go beyond just the problem of resolving am-

biguous correspondences (e.g., which is the main dorsal

STS branch in Case D). The deeper problem is that no reg-

istration that respects the topology of the cortical sheet

can successfully match every major and minor fold in

one individual with a corresponding fold in another individ-

ual. Instead, the registration process must tolerate folding

mismatches, such as the crown of a gyrus in one individual

corresponding to the fundus of a sulcus in another individ-

ual. Folding mismatches can occur at both a fine-grained

scale, such as the irregular minor gyri and sulci in Figures

2A and 2B (blue arrows) and also at a more macroscopic

scale, as with atypical ‘‘folding variants’’ in which a major

sulcus or gyrus has a very different configuration in some

Neuron 56, October 25, 2007 ª2007 Elsevier Inc. 211

Neuron

Primer

Figure 3. Shape Characteristics ofCerebral Cortex in an IndividualMacaque(A) A parasagittal section through the righthemisphere sMRI of macaque F99UA1 (a.k.a.the F99 atlas, Table 1). Red contour showsa slice through the fiducial cortical midthick-ness surface.(B) The right hemisphere fiducial surface.(C) The very inflated surface, with a map of sul-cal depth.(D) A flat map representation.Data are accessible via http://sumsdb.wustl.edu/sums/directory.do?id=6650508.

individuals compared to the most common pattern (Lyttel-

ton et al., 2007; Mangin et al., 2004; Ono et al., 1990).

These issues are very important when considering the

strengths and limitations of various registration algo-

rithms, a topic discussed in a later section.

Cortical organization has been studied extensively in

many other primate species using anatomical, physiolog-

ical, and/or neuroimaging approaches, but the macaque

monkey is by far the most intensively studied nonhuman

primate. The macaque is also particularly relevant to hu-

man neuroimaging, thanks to a growing number of ma-

caque fMRI studies that facilitate interspecies compar-

isons (Brewer et al., 2002; Denys et al., 2004; Orban

et al., 2004; Sereno and Tootell, 2005; Tsao et al., 2006;

Vincent et al., 2007). There are many fundamental similar-

ities across primate species, but also many obvious quan-

titative and qualitative differences. The surface area of

macaque cerebral cortex is only 15% of that in humans

(120 cm2 versus 900 cm2; Van Essen et al., 2005), and it

is only 1–2 mm thick (O’Kusky and Colonnier, 1982). The

overall patter of cortical convolutions is far less variable

in the macaque (Figure 3) compared to humans, even

though there is 2-fold or more variability in areal size

(Maunsell and Van Essen, 1987; Van Essen et al., 1984).

Gyral and sulcal landmarks are typically reliable to within

a few mm in predicting the location of most cortical areas

(Carmichael and Price, 1994; Lewis and Van Essen, 2000;

Van Essen, 2004a; Van Essen et al., 2005). Studies of

architecture, connections, topographic organization, and

function suggest that there are about 100 areas in the ma-

caque cortical mosaic (Van Essen, 2004a, 2004b). The av-

erage size of a macaque cortical area is about 1 cm2, but

the range across different areas approaches two orders of

magnitude (Felleman and Van Essen, 1991). Fortunately,

the smaller brain size of the macaque allows fMRI to be

carried out at higher resolution than in humans, with voxel

size routinely 2 mm or less.

The species differences between humans and ma-

caques in the variability of cortical folding and in the mag-

nitude of function-folding mismatches, while very striking,

have a plausible explanation in terms of the developmen-

tal mechanisms that give rise to cortical folds. In brief, me-

chanical tension along the axons of long-distance cortico-

cortical connections may be the primary driving force for

cortical folding during prenatal and perinatal development

(Van Essen, 1997). By this hypothesis, consistency in fold-

212 Neuron 56, October 25, 2007 ª2007 Elsevier Inc.

ing may reflect consistent patterns of connectivity among

nearby areas. Variability in folding may reflect differences

in connectivity and/or size among a mosaic of many small

‘‘balkanized’’ areas. Species differences in the overall

degree of variability may reflect a larger number of cortical

areas in humans, along with a disproportionate increase in

cortical surface area relative to the underlying white matter

and subcortical gray matter (Van Essen, 2006).

2. Probabilistic Brain Atlases AreFrameworks for the FutureIdeally, brain atlases for the 21st century should include

many attributes and many types of data (cf. Toga et al.,

2006). The following nine attributes are particularly impor-

tant for digital atlases of cerebral cortex. Resolution. The

atlas should represent brain structure at high spatial reso-

lution. Areas. It should include maps of cortical areas

based on as many as possible of the various partitioning

schemes in current use. Probabilistic. Cortical areas, cor-

tical folding patterns, and identified gyri and sulci should

be represented probabilistically whenever possible, in a

way that reflects variability in cortical convolutions and in

the size, location, and internal (e.g., topographic) organiza-

tion of cortical areas. Visualization. The atlas should be

linked to powerful, flexible, and easy-to-use visualization

tools, both surface-based and volume-based, to facilitate

navigation and display of diverse data types. Coordinates.

It should be associated with well-defined stereotaxic and

surface-based coordinate systems to facilitate objective

reporting of spatial location. Atlas-linked. It should be reg-

istered to other widely used cortical atlases and stereo-

taxic spaces, in order to facilitate comparisons across

studies. Accessible. It should be readily accessible for vi-

sualization online and after downloading. Database-linked.

It should be linked to searchable databases that allow

a rapidly growing body of experimental data to be viewed,

analyzed, and compared using the atlas framework. Exten-

sible. It should be easy to update the atlas as new experi-

mental data (e.g., newly charted cortical areas) become

available.

No single atlas is fully satisfactory in meeting all of

these criteria, but a number of useful digital atlases have

emerged over the past decade. Unlike book atlases, digital

brain atlas are not static entities with well-defined content.

Instead, they have evolved into inherently complex and

dynamic constructs in terms of their information content

Neuron

Primer

and how it is accessed. Portals to emerging atlases and

associated neuroimaging tools are accessible via the

Neuroscience Database Gateway (http://ndg.sfn.org/)

and the Neuroimaging Informatics Tools and Resources

Clearinghouse (NITRC) (http://www.nitrc.org/).

Table 1 provides a snapshot of major human and ma-

caque digital cortical atlases that are currently accessible

to the neuroimaging community for use in visualization and

analysis. The first three columns indicate the primary visu-

alization substrate for each atlas volume (column 1) and/or

surface (columns 2 and 3). The primary anatomical ‘‘tem-

plate,’’ which may be a volume and/or a surface, forms the

core of an atlas, but the atlas itself may contain extensive

additional reference data (columns 5 and 6; see below).

Some atlases have multiple row entries, reflecting their

availability in different stereotaxic spaces (column 4) or

software platforms (column 8). We consider first the ana-

tomical templates and visualization substrates for single-

brain and population-average human atlases (columns

1–3).

The two single-subject atlases in widespread use differ

in many respects. The classical Talairach atlas (Talairach

and Tournoux, 1988 [‘‘T88’’]) defined a standard stereo-

taxic space that attained well-deserved prominence as

a spatial framework for reporting coordinates of neuroi-

maging results (Fox et al., 1985, 2005). It is based on brain

slice drawings from a single postmortem brain that are

digitally accessible via the Sleuth (BrainMap) database.

The ‘‘Colin27’’ atlas is based on high-resolution imaging

of a single subject initially generated in MNI space (Fig-

ure 4A) but also available after transformation to several

other stereotaxic spaces. Available fiducial surfaces for

the Colin27 atlas include representations of the gray-white

(GW) boundary (Figure 4B), the pial surface (data not

shown), and the cortical midthickness (CMT) representa-

tion (Figure 4C). The GW and pial surface representations

can be used jointly to compute cortical thickness (Fischl

and Dale, 2000). The CMT representation provides a

more balanced representation of gyral and sulcal regions,

which is important for analyses that involve measure-

ments of cortical surface area (Van Essen et al., 2001).

Population-average atlases are increasingly widely used

because they avoid the biases associated with the idio-

syncratic cortical folding pattern of any particular brain,

including T88 and Colin27. In a volume-averaged popula-

tion atlas such as the ICBM 152 (a.k.a. MNI152, Figure 4D),

residual variability is manifested by pronounced blurring

between gray matter and white matter, especially in corti-

cal regions.

Volume-averaged atlases have the advantage of com-

plete coverage of the brain, regularity of spatial sampling

(uniform voxel size), and the availability of many voxel-

based visualization and analysis tools. Their major limita-

tions relate to the difficulty of displaying complex, spatially

distributed data and to the alignment errors associated

with volume-based registration. Surface-averaged atlases

have the advantage of flexible cortical visualization options

and greater fidelity of surface-based registration. A major

limitation is the exclusion of subcortical structures. Also,

capitalizing fully on the alignment capabilities of surface-

based registration requires high-quality surface recon-

structions of each individual subject. Even with recent

improvements in automated segmentation and error cor-

rection, insuring adequate quality control remains an im-

portant issue.

Given these inherent complementarities, a hybrid atlas

that supports concurrent utilization of surfaces and vol-

umes for visualization and analysis is naturally attractive.

This is the objective of the PALS (Population-Average,

Landmark, and Surface-based) atlas concept and the

specific PALS-B12 atlas data set that is based on sMRI

volumes from 12 normal young adults (Van Essen, 2005a).

The PALS-B12 atlas volume was generated by averaging

the individual sMRI volumes after registration to stereo-

taxic space (Figure 5A). The PALS-B12 atlas surfaces

were generated by reconstructing the fiducial surface of

each hemisphere, registering each surface to the atlas,

and generating ‘‘average fiducial’’ surface representations

for the left and right hemispheres (Figure 5B). In regions of

low variability such as the central sulcus (CeS), major gyral

and sulcal features are largely preserved, but in regions of

high variability many features are averaged out and the

average fiducial surface runs approximately midway along

the depth of the major sulci, as evidenced in the surface

contours visible in Figure 5A. The PALS atlas provides

flexible visualization options that include inflated, very

inflated, flat map, and spherical configurations (Figures

5C–5F; Table 1). In each surface configuration, a map of

average sulcal depth provides an objective measure of

cortical shape that has well-defined features in regions

of low variability and is blurred in regions of high variability.

A key aspect of surface-based registration involves

representing each individual hemisphere’s surface by

a ‘‘standard mesh’’ (Saad et al., 2004) that contains a fixed

number of surface nodes. A standard mesh can implicitly

define what constitutesgeographiccorrespondence,based

on the atlas and registration algorithm used. For example,

the more posterior of the highlighted nodes (black) in Fig-

ure 5 is consistently on the postcentral gyrus of the PALS

atlas surfaces (Figures 5A–5F) and in the two individual

hemispheres (Figure 5G). The moreanteriornode, ina region

of higher variability, lies on the precentral gyrus in one

individual but in the precentral sulcus in another.

Another important aspect of correspondence involves

comparisons between the left and right hemispheres. In

the PALS atlas, both the left and right hemispheres were

registered to an unbiased target (based on landmarks

derived from both hemispheres). This implicitly defines

geographically corresponding locations in the two hemi-

spheres (see highlighted nodes in Figure 5) and facilitates

analyses of hemispheric symmetries and asymmetries

(Van Essen, 2005a; Van Essen et al., 2006).

Study-Specific Averages and Age-Specific Atlases

An alternative population-average strategy is to generate

study-specific population-average surfaces that are based

on a group of subjects within a given study but not linked to

Neuron 56, October 25, 2007 ª2007 Elsevier Inc. 213

Table

Visuali

Citation

Volume

(Templ

Single-

1. Tala c Talairach and Tournoux, 1988;

Lancaster et al., 2000

Lancaster et al., 2000

2. Colin ,j Tzourio-Mazoyer et al., 2002;

Eickhoff et al., 2005, 2007

Caretl Bm Van Essen, 2002, 2004a

FreeSu Dale et al., 1999

SUMA Saad et al., 2004

Popula

3. MNI

4. ICBM M,f Mazziotta et al., 2001;

Eickhoff et al., 2005, 2007;

Lancaster et al., 2000

5. ICBM

6. LON Shattuck et al., 2007

7. LON Dinov et al., 2000;

Mega et al., 2005

8. 711- Buckner et al., 2004

9. PAL aretz Van Essen, 2005a

10. Fre

averag

Fischl et al., 1999

Single-

11. F99 Bm Van Essen, 2002, 2004a

12. PH Paxinos et al., 2000;

Van Essen, 2002

13. N2 Bm Black et al., 2001

Neu

ron

Prim

er

214

Neuro

n56,

Octo

ber

25,2007

ª2007

Els

evie

rIn

c.

1. Digital Atlases of Human and Macaque Cerebral Cortex

zation Substrates (Templates) Related Data

Softwareate)

Fiducial

Surface

Other

Configurations

Stereotaxic

Space Source Content

Brain Atlases: Human

irach T88 TDa gyri; Brod Sleuth,b AFNI

MNI TD,a WPAd gyri; Brod FSL,e SPMf

27 sMRIg MNI AAL,h JAMi gyri; areas SPMf; MRIcro

Caret,k AFNIc

CMT inflated, flat,

sphere

MNI, T88,

711-2B

— sulci; Brod, Vis,

OrbFront

Caretk/SumsD

rfern GW, pial sphere MNI — gyri; subcortical FreeSurfero

p GW, pial inflated MNI JAMi areas SUMAp

tion Atlases: Human

305q BICr

152s MNI JAM,i TA,t HOAt gyri; areas BIC,r AIR,u SP

FSL,e AFNIc

452v MNI

I (LPBA40w) MNI PubBrainx GM, WM (P)

I ADPA (AIR)u SVPAy areas

2B 711-2B Caretk

S-B12z CMT

(average,

individuals)

inflated, flat,

sphere

MNI, MNI,

T88, 711-2B

JAMi sulci, areas;

Brod, visual,

OrbFront

Caretk; WebC

eSurfer

e 40aasphere gyri FreeSurfero

Brain Atlases: Macaque

z CMT inflated, flat,

sphere

AC-PC sulci; areas Caretk/SumsD

Tz CMT interaural areas

Kbb Caretk/SumsD

14. UCLA

ne

Monkey 2Dd

Cannestra et al., 1997

15

m

rgee Mikula et al., 2007

16 Saleem and Logothetis, 2006

P

17 DBm Vincent et al., 2007

U imer’s Disease Probabilistic Atlas; Brod, Brod-

m eas; SVPA, Sub-Volume Probabilistic Atlas; TA,

Taa hb

c hd

e hf hg

h hi hj hk hl hm

n ho

p

q

r hs ht hu hv hw

x hy hz haa

bb

cc

dd

ee

ff

Neu

ron

Prim

er

Neuro

n56,O

cto

ber

25,2007

ª2007

Els

evie

rIn

c.

215

mestrinacc,dd Atlas Viewerd

. UC-Davisee

acaque

histological

sections

none geography, areas BrainMaps.o

. D99ff macaque EBZ geography, areas SumsDBm

opulation-Average Atlas: Macaque

. F6z CMT inflated, flat AC sulci Caretk/Sums

nderscore = probabilistic. Abbreviations: AAL, automatic anatomical labeler; AC-PC, anterior-commissure/posterior-commissure; ADPA, Alzhe

ann areas; CMT, cortical midthickness; GW, gray/white; HOA, Harvard/Oxford Atlas; JAM, Juelich Architectonic Maps; OrbFront, orbitofrontal ar

lairach Atlas; TD, Talairach Daemon; WPA, WFU_PickAtlas, Wake Forest Pick-Atlas.

ttp://ric.uthscsa.edu/new/resources/talairachdaemon/talairachdaemon.html

http://brainmap.org/

ttp://afni.nimh.nih.gov/afni/

http://www.fmri.wfubmc.edu/cms/software

ttp://www.fmrib.ox.ac.uk/fsl/

ttp://www.fil.ion.ucl.ac.uk/spm/

http://www.bic.mni.mcgill.ca/brainweb/

ttp://www.cyceron.fr/freeware/

ttp://www.fz-juelich.de/inb/inb-3/spm_anatomy_toolbox

ttp://www.sph.sc.edu/comd/rorden/mricro.html

ttp://brainmap.wustl.edu/caret/

ttp://sumsdb.wustl.edu/sums/directory.do?id=6363287&dir_name=COLIN_CEREBRAL

http://sumsdb.wustl.edu/sums/

ttp://sourceforge.net/project/shownotes.php?release_id=325132

http://surfer.nmr.mgh.harvard.edu/fswiki/DownloadAndInstall

http://afni.nimh.nih.gov/afni/suma

http://www.bic.mni.mcgill.ca/software/mni_autoreg/

ttp://www.bic.mni.mcgill.ca/software/

ttp://www.bic.mni.mcgill.ca/cgi/icbm_view/

ttp://www.fmrib.ox.ac.uk/fsl/fslview/atlas-descriptions.html

ttp://bishopw.loni.ucla.edu/AIR5/index.html

ttp://www.loni.ucla.edu/ICBM/Downloads/Downloads_Atlases.shtml

http://www.loni.ucla.edu/Atlases/Atlas_Detail.jsp?atlas_id=12

ttp://www.pubbrain.org/

ttp://www.loni.ucla.edu/ad/AD_ASVPA.html

ttp://sumsdb.wustl.edu/sums/directory.do?id=6585200&dir_name=CARET_TUTORIAL_SEPT-06

http://surfer.nmr.mgh.harvard.edu/

http://www.nil.wustl.edu/labs/kevin/ni/n2k/p1.htm

http://www.loni.ucla.edu/Atlases/Atlas_Detail.jsp?atlas_id=2

http://www.loni.ucla.edu/Research/Atlases/Data/monkey/MonkeyAtlasViewer.shtml

http://brainmaps.org/

http://sumsdb.wustl.edu/sums/directory.do?id=6653275

Neuron

Primer

Figure 4. Single-Brain and Population-Average Atlas Examples(A) High-resolution sMRI Colin27 individual-brain atlas.(B) Colin27 fiducial surface generated from thegray-white (GW) boundary (SUMA).(C) Colin27 fiducial surface generated from thecortical midthickness (CMT) surface (Caret; VanEssen, 2002). The CMT surface is generatedautomatically in Caret, but it can also be com-puted by averaging the GW and pial surfaces.(D) Coronal slice through the ICBM-152 volume-average population atlas (Mazziotta et al.,2001).Data are accessible via http://sumsdb.wustl.edu/sums/directory.do?id=6650508.

any particular atlas (Argall et al., 2006; Goebel et al., 2006;

Lyttelton et al., 2007). This successfully reduces intersub-

ject variability and can facilitate a variety of within-group

analyses (Figure 6A; see also below). However, it is less

amenable to quantitative comparisons of results across

studies unless and until the data are subsequently regis-

tered to a common surface-based atlas. A related con-

cept involves age-specific template atlases that represent

shape characteristics within defined age groups, ranging

from infants (J. Hill et al., 2007, Soc. Neurosci., abstract)

to elderly adults (Thompson et al., 2001). Mappings be-

tween different age-specific atlases, when available, allow

quantitative comparisons across ages.

Another class of study-specific population-average at-

las involves surface representations of just the exposed

(gyral) portions of cerebral cortex (Shi et al., 2007; Thomp-

son et al., 2000; Toga and Thompson, 2002; Toga et al.,

2001). Because these ‘‘gyral maps’’ (Figures 6B and 5C)

do not explicitly represent buried cortex (which constitute

about two-thirds of total cortical surface area), they differ

from full-hemisphere atlases (e.g., PAS-B12 and FreeSur-

fer Average-40) in terms of topology and completeness of

the cortical representation. Nonetheless, it should be fea-

216 Neuron 56, October 25, 2007 ª2007 Elsevier Inc.

sible to generate mappings between gyral map atlases

and the corresponding gyral portions of full-hemisphere

atlases, thereby facilitating migration of data between

these two widely used classes of surface-based atlas.

Location, Location, Location!

The realtors’ mantra about the paramount importance of

location has a parallel in neuroimaging, where many differ-

ent options are used to specify location within the brain.

Cartographers describe locations on the earth’s surface

using spatial coordinates, geographic labels, and political

labels. Analogous strategies are used within the brain; the

main distinction is between coordinate-based and region-

based descriptions of location.

Coordinate-Based Localization

Spatial coordinates provide a concise, precise, and objec-

tive way to express location in a brain atlas relative to an

anatomically defined origin (the anterior commissure in the

Talairach and other human atlases). A ‘‘template’’ or atlas

represents the underlying structure, and a registration

process uses a particular algorithm to match any individ-

ual brain to the atlas. Available volumetric atlases differ

in the dimensions and shape of the template, in the regis-

tration process by which the template was generated

Figure 5. The PALS-B12 Atlas(A) Average sMRI volume from 12 young adultsubjects plus surface contours of the left andright hemisphere average fiducial surfaces.(B) Average fiducial surface generated byregistering each surface to the atlas by a land-mark-constrained deformation algorithm ap-plied to spherical maps.(C–E) Inflated, very inflated, and flat map con-figurations shaded by maps of average sulcaldepth. Highlighted nodes (black) representcorresponding locations in precentral andpostcentral regions of the left and righthemispheres.(F) Spherical map used in registration of individ-uals to the atlas.(G) Standard-mesh representations of fiducialsurfaces from two individuals, showing loca-tions that are defined as geographically corre-sponding via the registration process.Data are accessible via http://sumsdb.wustl.edu/sums/directory.do?id=6650508.

Neuron

Primer

Figure 6. Study-Specific Average Surfaces(A) Population-average tactile fMRI responses viewed on a study-specific right hemisphere average surface after inflation (Beauchamp et al., 2007).Averaging the fMRI data on the surface reduces individual variability (Argall et al., 2006; see below), but displaying results on a study-specific averageimpedes comparisons across studies. Reproduced with permission from Beauchamp et al. (copyright 2007, the Society for Neuroscience). (B) A pop-ulation-average cerebral hull model generated by sulcal landmark-constrained registration applied to implicit surfaces (Shi et al., 2007). (C) A map ofvariability among the nine contributing subjects. (B) and (C) are reproduced with permission from Shi et al. (copyright 2007, Elsevier).

(sometimes a complex and iterative process), and the

registration algorithms available for registering individual

subjects to the template. There is a dichotomy between

templates that approximate the dimensions of the original

single-brain Talairach atlas (‘‘T88’’ space) and those

based on the MNI305 target (‘‘MNI’’ space), which are

actually larger than average brain dimensions (Lancaster

et al., 2007). The differences between Talairach and MNI-

based templates can exceed 10 mm in the coordinates

of a given geographic locus; even within atlases in MNI

space, differences can be up to 5 mm when comparing

linear to nonlinear algorithms (Lancaster et al., 2007; Van

Essen and Dierker, 2007). Proposals that the community

voluntarily converge on an existing stereotaxic space

(Devlin and Poldrack, 2007) are laudable, but the number

of widely used atlases and sterotaxic spaces seems more

likely to increase than to decrease in the coming decade.

Given the diversity of atlases, methods to map from one

to another are essential in order to avoid errors and confu-

sion when comparing data reported on different atlases.

To this end, matrix transformations that convert from

one space to another are available for ones that are based

on affine or other low-dimensional algorithms (e.g., Lan-

caster et al., 2007). An alternative strategy involves ‘‘sur-

face atlas mediation,’’ using average fiducial surfaces that

were created separately for each target atlas and are part

of the PALS atlas data set (cf. Figure 4B; Table 1). This

provides an efficient and unbiased way to map diverse

types of data (fMRI volumes, etc.) onto atlas surfaces

(see below).

For cerebral cortex, spatial coordinates can also be re-

ported in a 2D spherical coordinate system of latitude and

longitude (Fischl et al., 1999; Van Essen, 2005a; see also

Clouchoux et al., 2005). Because spherical coordinates

respect cortical surface topology, locations with similar

spherical coordinates are always in close proximity along

the cortical sheet. In contrast, locations on opposite banks

of a sulcus may have similar 3D stereotaxic coordinates

yet be far apart within the cortex. Multiple versions of

spherical space are available, and in some cases transfor-

mations between spherical atlas spaces are available,

such as between PALS-B12 and FreeSurfer Average-40

(Van Essen, 2005a).

Region-Based Localization

Atlases gain in usefulness when locations can be ex-

pressed in relation to identified geographic and/or func-

tional regions. Sources of geographical information de-

rived from single subjects and available in one or more

atlases include cortical gyri and classical Brodmann areas

derived from the original Talairach atlas, Talairach Dae-

mon (Table 1, ‘‘TD’’ in column 5); anatomically defined

subregions of the Colin27 brain in the Automated Anatom-

ical Labeling (AAL) tool; and Brodmann, visuotopic, and

orbitofrontal areas on the Colin27 surface (Caret).

Probabilistic maps that represent the range of variability

in a standard population provide an increasingly important

way to cope with individual variability. Available probabi-

listic representations of cortical geography include corti-

cal and subcortical probability maps in the FSL Harvard/

Oxford Atlas (HOA), the LONI Sub-Volume Probabilistic

Atlas (SVPA), and maps of identified sulci in the PALS

atlas.

A powerful alternative strategy for geographic localiza-

tion involves automated sulcal/gyral identification algo-

rithms that can be applied to individual subjects registered

to atlas space. BrainVisa (http://brainvisa.info/) and Free-

Surfer software currently provide such options. Efforts to

improve automated identification of cortical and subcorti-

cal structures are ongoing in many laboratories (Cachia

et al., 2003; Desikan et al., 2006; Fischl et al., 2004; Klein

et al., 2005; Mega et al., 2005). While these processes are

not perfect, given the complexity and variability of convo-

lutions (Ono et al., 1990; see above), they are likely to pro-

vide better estimates for any given individual than those

derived from probabilistic maps of a standard population.

A growing collection of probabilistic architectonic maps

have been generated from architectonic analyses of post-

mortem brains by the Zilles laboratory in Juelich. This pro-

cess involves objective charting of areal boundaries, reg-

istration of individual hemispheres to the Colin27 brain

template, and summation across individual cases to gen-

erate population-average architectonic maps (Amunts

Neuron 56, October 25, 2007 ª2007 Elsevier Inc. 217

Neuron

Primer

Figure 7. Cross-Platform Analyses offMRI Data(A–C) Visualization of fMRI activations fromthree published studies carried out usingvolume-averaged analyses in different stereo-taxic spaces, then mapped to the PALS atlassurface using space-specific multi-fiducialmapping to compensate for individual variabil-ity and for the differences between spaces (VanEssen, 2005a). Visuotopic area boundaries aremainly from the fMRI study of Hadjikhani et al.(1998). Data are accessible via http://sumsdb.wustl.edu/sums/directory.do?id=6650508.

et al., 2007; Eickhoff et al., 2005, 2007). These ‘‘Juelich ar-

chitectonic maps’’ (‘‘JAM’’ in Table 1, column 5) are asso-

ciated with several atlases and are accessible via several

software platforms. Though extremely valuable, the avail-

able probabilistic architectonic maps currently cover only

a minority of the cortical expanse. Registration to the

Colin27 brain has been carried out using both linear and

nonlinear algorithms. The high-dimensional nonlinear reg-

istration achieves tighter clustering of each area in the

single-subject target atlas, but there is a potential disad-

vantage in using a registration strategy that differs from

the linear (e.g., FSL’s FLIRT) or low-dimensional nonlinear

(e.g., SPM2) algorithms typically used in fMRI studies (Pol-

drack and Devlin, 2007; Van Essen and Dierker, 2007).

When using probabilistic architectonic maps to infer the

cortical area(s) associated with any given fMRI activation

pattern, the choice of which registration method provides

greatest accuracy is an empirical question that warrants

further investigation.

3. Atlases Are Key to Data Mappingand Data MiningDigital brain atlases are invaluable for the analysis and

visualization of vast amounts of structural and functional

neuroimaging data obtained in ongoing experimental

studies. The specific analysis options available are too

numerous, diverse, and platform dependent to be covered

systematically here, but it is useful to draw attention to two

general ways in which atlas utility can be enhanced. These

involve data migration across platforms and data mining

across studies.

Migration across Platforms

No single software platform is comprehensive in its data

analysis and visualization capabilities, and many studies

can benefit from analyses that make use of multiple plat-

forms. Cross-platform data migration can occur at many

analysis stages, such as intensity normalization on one

platform (FSL), segmentation on another (FreeSurfer), and

surface-based registration on a third (Caret/PALS). One

increasingly common two-stage approach is to ‘‘analyze

volumes, visualize surfaces.’’ More specifically, this en-

tails carrying out primary analyses of fMRI data in the

volume domain, thereby capitalizing on sophisticated vol-

umetric analysis tools (SPM, AFNI, etc.). Results are then

mapped onto a surface-based atlas (e.g., PALS) to aid in

218 Neuron 56, October 25, 2007 ª2007 Elsevier Inc.

visualization and further analysis. For example, Figure 7

shows fMRI activation patterns from three studies initially

analyzed in distinct stereotaxic spaces (T88, MNI, and

711-2B). Volume-averaged results were mapped to the

appropriate space-specific PALS atlas by an easily exe-

cuted process of ‘‘multi-fiducial mapping’’ that minimizes

biases arising from folding variability (Van Essen, 2005a).

Each set of fMRI activations is displayed in relation to the

boundaries of visuotopic areas (colored contours) that

were themselves mapped to the PALS atlas by surface-

based registration. This facilitates objective assessment

of the degree of overlap between motion-selective (Fig-

ure 7A; Lewis et al., 2000) and object-selective (Figure 7B;

Denys et al., 2004) activations to one another, to area MT

(red contours) and other cortical areas, and to the vibro-

tactile activation pattern, in early blind individuals (Fig-

ure 7C; Burton et al., 2004).

Cross-platform data migration and analysis can be facil-

itated by increased use of common data formats. For vol-

ume data, NIfTI (NeuroImaging Informatics Technology

Initiative, http://nifti.nimh.nih.gov/) is an emerging stan-

dard that is supported by a growing number of neuroimag-

ing software platforms. An analogous effort for surface

representations (GIfTI) began more recently but will hope-

fully become widely adopted as well.

Data Mining

Investigators interested in relating their current neuroi-

maging results to what is already known must extract

relevant findings from a neuroimaging literature that has

mushroomed to thousands of studies in dozens of major

neuroscience journals, with specific results displayed in

many formats. It is increasingly difficult to carry out thor-

ough, careful, objective comparisons of results residing

in this expanding cacophony of neuroimaging studies.

To address this problem, progress is needed on multiple

fronts, but especially in the enhancement of neuroimaging

databases and data mining tools.

One domain that is ripe for enhanced data mining

involves the stereotaxic coordinates frequently used to

report the centers of activation foci and other regions of

interest. An estimated 3000 neuroimaging studies had re-

ported about 105 stereotaxic coordinates as of 2004 (Fox

et al., 2005), and the number of new studies reporting co-

ordinate data is now growing by perhaps 1000 per year.

Because tables of stereotaxic coordinates scattered in

Neuron

Primer

Figure 8. Meta-Analysis Results from Neuroimaging Studies Extracted from Databases and Displayed on Atlases(A) Mental rotation studies from the BrainMap database displayed on the Talairach atlas slice views using Sleuth software (http://brainmap.org/).(B) Mental rotation meta-analysis (Zacks, 2007) stored in SumsDB and displayed on the PALS atlas inflated surfaces using Caret (as shown here) oronline using WebCaret. Using data sets accessible via http://sumsdb.wustl.edu/sums/directory.do?id=6650508, metadata for individual studies canbe identified using color key (far right) and Identify Window (lower right), including links to the original online articles.

various journal articles are not readily searchable, stereo-

taxic coordinates need to be deposited in a searchable

database to allow effective data mining. Currently, two

complementary databases each house a substantial and

growing portion of the relevant literature. The BrainMap

database (http://brainmap.org/) currently contains �43,000

coordinates from �1200 studies, searchable using the

downloadable Sleuth application. The SumsDB database

currently contains �13,000 searchable coordinates from

�450 studies; search results can be viewed online on

the PALS atlas (via WebCaret) or downloaded for offline

analysis using Caret.

The complementary nature of these two databases can

be illustrated by searching each for studies related to a

particular functional task such as mental rotation. Using

Sleuth, this identifies hundreds of coordinates from 29

studies, displayed on slice views relative to the Talairach

atlas outline (Figure 8A). Extensive metadata relating to

each study are available within the BrainMap database.

Flexible meta-analysis options, including Activation Like-

lihood Estimation (Turkeltaub et al., 2002; Laird et al.,

2005) facilitate meta-analysis studies (Fox et al., 2005).

In SumsDB a meta-analysis of mental rotation studies

(Zacks, 2007) identified 319 coordinates from 32 studies

that can be displayed on the PALS atlas surface online

using WebCaret or offline in Caret (Figure 8B).

Cross-study comparisons and meta-analyses like those

illustrated in Figures 7 and 8 must of course be carried out

with careful attention to potential confounds. These in-

clude differences in task design or other methodological

details, spatial biases in the primary data, and a variety

of other differences across studies. On the other hand,

there are obvious downsides to failing to make adequate

comparisons with previously published results. Data min-

ing tools that facilitate access to key data and to the asso-

ciated primary literature will improve research efficiency

and allow investigators to concentrate on data analysis

and interpretation rather than finding and formatting the

relevant published data.

Stereotaxic coordinates provide only a sparse represen-

tation of the complex fMRI activation patterns reported in

any given study, and there is a growing need for databases

that can handle complex volumetric and surface-based

neuroimaging data, such as the examples shown in Fig-

ure 7. SumsDB and WebCaret are customized for com-

bined surface and volume storage and visualization. A ma-

jor emphasis is on specific data sets associated with

published figures that include WebCaret ‘‘scenes’’ exactly

replicating figure contents. The figure legends in the pres-

ent review include links to SumsDB/WebCaret that dem-

onstrate this feature.

The precision with which stereotaxic coordinate points

(‘‘foci’’) are displayed in Figure 8 belies the fact that the

underlying experimental fMRI activations are associated

with major spatial uncertainties and biases. These uncer-

tainties depend on many factors, but a particularly impor-

tant one is the fidelity of registration from individuals to the

atlas. This motivates further discussion of different ap-

proaches used to improve intersubject alignment.

4. Surface-Based Registration OutperformsVolume-Based RegistrationCurrently, most fMRI studies are analyzed largely or

entirely in the volume domain using well-established

software platforms (SPM, AFNI, etc.). However, surface-

based analyses of individual subjects, accompanied by

surface-based registration (SBR) to improve intersubject

alignment, are increasingly common. In theory, SBR has

an intrinsic advantage over volume-based registration

(VBR) because it respects the topology of the cortical

sheet. A key issue is whether in practice the gains in

intersubject alignment are sufficient to warrant more

Neuron 56, October 25, 2007 ª2007 Elsevier Inc. 219

Neuron

Primer

Figure 9. SBR Outperforms VBR inAligning Cortical Sulci(A and B) Medial views of the occipital lobe insurface reconstructions of two right hemi-spheres (Cases A and B, same as Figure 2)after VBR to 711-2B stereotaxic space. Thecalcarine sulcus (CaS) has very different trajec-tories in the two cases.(C and D) The different CaS trajectories arealso evident in parasagittal sMRI slices throughCases A and B.(E) Segmentation-based map of the dorsalbank of the calcarine sulcus (CaSd) in CasesA and B.(F) Volume-averaged probabilistic map of theCaSd in 12 individuals, revealing only moder-ately good alignment (many dark red regions).(G) Map of the CaSd from Cases A and B afterSBR to the PALS-B12 atlas surface, showinggood alignment in most regions.(H) Probabilistic map of the CaSd from all 12individuals after SBR, showing good overallalignment (bright red regions). Data are ac-cessible via http://sumsdb.wustl.edu/sums/directory.do?id=6650508.

widespread adoption of the SBR approach. Empirically,

many studies have now demonstrated that SBR indeed

can achieve substantially better alignment than VBR for

several types of experimental data, including fMRI activa-

tion patterns (A. Anticevic et al., 2007, Cogn. Neurosci.

Soc., abstract; Argall et al., 2006; Desai et al., 2005; Fischl

et al., 1999, 2004; Goebel et al., 2006; van Atteveldt et al.,

2004), simulated fMRI data (Jo et al., 2007), identified

cortical sulci (A. Anticevic et al., 2007, Cogn. Neurosci.

Soc., abstract; Nordahl et al., 2007; Van Essen, 2005a),

and architectonic areas (Yeo et al., 2007).

Figure 9 illustrates why SBR outperforms VBR and sets

the stage for comparing SBR algorithms. Using the same

example hemispheres as in Figure 2, the very different tra-

jectories of the calcarine sulcus (CaS) in Cases A and B are

evident in medial surface views (Figures 9A and 9B), in

parasagittal sMRI slices (Figures 9C and 9D), and in seg-

mented maps of its dorsal bank (CaSd, Figure 9E). In the

volume-averaged population map for all 12 contributing

subjects (Figure 9F), the CaSd showed considerable dis-

persion, signifying only moderately consistent alignment.

In contrast, surface-based registration to the PALS atlas

(using the ‘‘PALS-SBR’’ process) yielded much better

alignment of the CaSd for the two individuals (Figure 9G)

and for the population average (Figure 9H). By a quantita-

tive measure, the alignment was 2.5-fold better for SBR

than for VBR for the calcarine sulcus and on average

1.7-fold better for the 18 sulci tested (Van Essen, 2005a).

In general, the aim of SBR should be to align sulci that

are relatively consistent in shape but not to force large

local deformations in regions where variability is high

and folding mismatches are to be expected. In practice,

various SBR methods differ markedly in the degree of lo-

cal deformation tolerated in the registration process. At

one end of the spectrum, the SUMA-SBR method (Argall

et al., 2006) introduces no local deformations to improve

alignment and is therefore unlikely to be optimal in aligning

220 Neuron 56, October 25, 2007 ª2007 Elsevier Inc.

major landmarks like the central sulcus and calcarine sul-

cus. In contrast, the global energy-minimization approach

of FreeSurfer-SBR (Fischl et al., 1999) achieves good

alignment of major sulci, but it results in pronounced local

distortions when matching local features in regions where

folding mismatches are to be expected (Figure 2 in Wisco

et al., 2007; D.C.V.E., unpublished data). PALS-SBR is in-

termediate between these two and may be closer to opti-

mal, insofar as it uses a small number of explicit landmarks

to align major sulci and aims to minimize areal distortions

in the intervening regions. Careful testing using common

data sets is needed to better characterize the strengths

and limitations of these and several other available SBR

methods (Chung et al., 2003, 2005; Clouchoux et al., 2005;

Goebel et al., 2006; Lyttelton et al., 2007) and to help guide

algorithmic refinements that further improve intersubject

alignment. Also, in order to fully capitalize on the advan-

tages of intersubject SBR analyses, it is important to have

surface-based statistical analysis methods that allow sig-

nificance to be evaluated quantitatively. Such methods

are under active development, but many are already avail-

able for both morphometric and fMRI studies using major

software packages, including FreeSurfer, SUMA, Caret,

Brain Voyager (see Table 1), and SURFTRACC (Lyttelton

et al., 2007; Robbins et al., 2004).

SBR approaches will have an even greater impact as

techniques emerge that allow routine identification of

cortical areas in individual subjects, as this will allow

intersubject alignment to be constrained by cortical area

boundaries (Dougherty et al., 2003; Larsson and Heeger,

2006). There has been encouraging recent progress in

cortical areal delineation using diffusion tractography

(Behrens et al., 2006; Klein et al., 2007) and spatial corre-

lations of resting-state fMRI fluctuations (Margulies et al.,

2007).

‘‘Functional localizers’’ offer an alternative way to com-

pensate for individual variability using SBR. Typically,

Neuron

Primer

Figure 10. Interspecies Comparisons of Cortical Organization(A) Macaque atlas (fiducial surface, lateral view) with architectonic areas from Lewis and Van Essen (2000) displayed.(B) Macaque areas registered to the human PALS atlas using interspecies SBR and 23 functionally defined landmarks (Orban et al., 2004).(C) Brodmann (1909) architectonic areas displayed on the inflated PALS atlas surface.(D) Map of cortical expansion based on the registration between macaque and human, showing hotspots of high expansion near the temporo-parietaljunction and the dorsolateral prefrontal cortex. Data are accessible via http://sumsdb.wustl.edu/sums/directory.do?id=6650508.

a functional localizer task is used to identify a region (clus-

ter of voxels) in each subject that is related to a particular

function (e.g., face selectivity) and can be used to con-

strain subsequent analyses of independently acquired

data (Saxe et al., 2006). By defining functional localizers

on surfaces rather than volumes and using them in addi-

tion to or instead of local geographic features as registra-

tion constraints, improvement in intersubject alignment of

nearby functional regions might be attained. On the other

hand, it is critical to recognize that functional localizers

cannot in general be equated with functionally specialized

areas (Friston et al., 2006; Friston and Henson, 2006).

Thus, while this approach does have potential, it also has

major pitfalls that warrant careful evaluation in any empir-

ical tests.

Given the inherent limitations of affine and other low-

dimensional volume registration algorithms, efforts have

been made to improve in intersubject alignment in humans

using high-dimensional VBR algorithms (Eickhoff et al.,

2006; Thompson et al., 2000). However, the local defor-

mation patterns required in the volume domain in order

to respect surface topology are enormously more com-

plex than are needed when registering spherical surfaces.

Consequently, high-dimensional VBR should be applied

and interpreted with great caution as a method for reduc-

ing intersubject variability in human cerebral cortex.

The limitations of high-dimensional VBR and energy-

based SBR arising in connection with intersubject align-

ment of human cortex are much less of a concern when

dealing with shapes that are less variable, even if still very

complex. This applies to intersubject alignment of cere-

bral cortex in the more consistently folded macaque mon-

key (Chef d’Hotel et al., 2002; Nelissen et al., 2005) and to

intrasubject alignment of individual human brains that are

followed longitudinally. Indeed, for longitudinal studies in

general, high-dimensional VBR and SBR should in prin-

ciple have a significant advantage over landmark-based

SBR. Also, high-dimensional VBR has major advantages

for cross-model intrasubject registration, especially when

registering EPI volumes (which can be highly distorted) to

sMRI volumes.

5. Surface-Based Analyses FacilitateMonkey-Human ComparisonsLike their human counterparts, macaque brain atlases

come in several varieties (Table 1, rows 11–17). Digital

sMRI-based atlases include several single-subject volu-

metric atlases, plus the F99 hybrid surface-volume atlas

(Van Essen, 2002, 2004a). The population-average sur-

face-volume F6 atlas serves as a useful target for fMRI

studies (Vincent et al., 2007) and has been registered to

the F99 atlas so that data can migrate from one to the

other. The F99 and F6 atlases also contain a large reper-

toire of associated reference data available in the SumsDB

database (see Figure 10 legend), including 12 areal parti-

tioning schemes (one of which is a probabilistic architec-

tonic map), and connectivity data plus links to the CoCo-

Mac connectivity database (http://www.cocomac.org/).

The Saleem and Logothetis (2006) printed macaque atlas

includes MRI images, histological sections, and areal par-

titioning schemes, and the associated ‘‘D99’’ MRI volume

is accessible online (Table 1, row 16). Another useful re-

source is the labeled histological sections of macaque,

human, and other species that are available at http://

brainmaps.org/.

A substantial minority of macaque cortical areas have

convincing or strongly suspected homologs in humans,

based on similarities in architecture and/or functional or-

ganization (Ongur et al., 2003; Orban et al., 2004; Petrides,

2005; Sereno and Tootell, 2005; Tootell et al., 2003; Van

Essen, 2005b). Surface-based atlases can play a pivotal

role in evaluating candidate homologies and facilitating

objective quantitative comparisons of cortical organiza-

tion in monkeys and humans. This is because the pro-

found species differences in folding and in area-folding

relationships make geographic (shape) features generally

unsatisfactory as a constraint on registration and instead

require landmarks based on known or suspected homolo-

gies. Using landmark-constrained interspecies registra-

tion involving 23 presumed homologies, Figure 10 shows

a map of architectonically delineated cortical areas in the

macaque (Figure 10A), the same macaque areas regis-

tered to human cortex (Figure 10B), and a map of human

Neuron 56, October 25, 2007 ª2007 Elsevier Inc. 221

Neuron

Primer

architectonic areas for comparison (Figure 10C). A map of

cortical expansion in humans versus macaque based on

this registration shows several hotspots of particularly

high expansion (Figure 10D). Future studies capitalizing

on additional experimental information brought into this

atlas framework and analyzed using interspecies surface-

based registration may help resolve the key evolutionary

issue of whether local cortical expansion has occurred

mainly by the emergence of entirely new areas or mainly

by differential expansion of existing areas in a common

ancestor.

Concluding RemarkThe roles played by brain atlases in the study of the cere-

bral cortex will continue to expand and evolve rapidly.

These infrastructural advances will greatly improve our

ability to tackle questions that get at the core of what

makes us uniquely human, what characterizes our individ-

ual personalities and capabilities, and what goes wrong in

countless brain diseases and disorders.

ACKNOWLEDGMENTS

We thank J. Harwell and P. Gu for superb software development; E.Reid for technical assistance; S. Danker for assistance in manuscriptpreparation; and K. Hansen, J. Zacks, A. Anticevic, J. Van Horn, andothers for helpful comments. Supported by NIH Grant R01-MH-60974, funded by the National Institute of Mental Health, the NationalInstitute for Biomedical Imaging and Bioengineering, and the NationalScience Foundation.

REFERENCES

Amunts, K., Malikovic, A., Mohlberg, H., Schormann, T., and Zilles, K.(2000). Brodmann’s areas 17 and 18 brought into stereotaxic space—Where and how variable? Neuroimage 11, 66–84.

Amunts, K., Schleicher, A., and Zilles, K. (2007). Cytoarchitecture ofthe cerebral cortex—More than localization. Neuroimage 37, 1061–1065.

Andrews, T.J., Halpern, S.D., and Purves, D. (1997). Correlated sizevariations in human visual cortex, lateral geniculate nucleus, and optictract. J Neurosci. 17, 2859–2868.

Argall, B.D., Saad, Z.S., and Beauchamp, M.S. (2006). Simplified inter-subject averaging on the cortical surface using SUMA. Hum. BrainMapp. 27, 14–27.

Beauchamp, M.S., Yasar, N.E., Kishan, N., and Ro, T. (2007). HumanMST but not MT responds to tactile stimulation. J. Neurosci. 27,8261–8267.

Behrens, T.E., Jenkinson, M., Robson, M.D., Smith, S.M., and Johan-sen-Berg, H. (2006). A consistent relationship between local whitematter architecture and functional specialisation in medial frontalcortex. Neuroimage 30, 220–227.

Black, K.J., Snyder, A.Z., Koller, J.M., Gado, M.H., and Perlmutter, J.S.(2001). Template images for nonhuman primate neuroimaging: 1.Baboon. Neuroimage 14, 736–743.

Brewer, A.A., Press, W.A., Logothetis, N.K., and Wandell, B.A. (2002).Visual areas in macaque cortex measured using functional magneticresonance imaging. J. Neurosci. 22, 10416–10426.

Brodmann, K. (1909). Vergleichende Lokalisationslehre der Grosshirn-rinde in ihren Prinzipien dargestellt auf Grund des Zellenbaues (Leip-zig: J.A. Barth).

222 Neuron 56, October 25, 2007 ª2007 Elsevier Inc.

Buckner, R.L., Head, D., Parker, J., Fotenos, A.F., Marcus, D., Morris,J.C., and Snyder, A.Z. (2004). A unified approach for morphometricand functional data analysis in young, old, and demented adults usingautomated atlas-based head size normalization: Reliability and valida-tion against manual measurement of total intracranial volume. Neuro-image 23, 724–738.

Burton, H., Sinclair, R.J., and McLaren, D.G. (2004). Cortical activity tovibrotactile stimulation: An fMRI study in blind and sighted individuals.Hum. Brain Mapp. 23, 210–212.

Cachia, A., Mangin, J.F., Riviere, D., Kherif, F., Boddaert, N., Andrade,A., Papadopoulos-Orfanos, D., Poline, J.B., Bloch, I., Zilbovicius, M.,et al. (2003). A primal sketch of the cortex mean curvature: A morpho-genesis based approach to study the variability of the folding patterns.IEEE Trans. Med. Imaging 22, 754–765.

Cannestra, A.F., Santori, E.M., Holmes, C.J., and Toga, A.W. (1997). Athree-dimensional multimodality brain map of the nemestrina monkey.Brain Res. Bull. 43, 141–148.

Carmichael, S.T., and Price, J.L. (1994). Architectonic subdivision ofthe orbital and medial prefrontal cortex in the macaque monkey. J.Comp. Neurol. 346, 366–402.

Chef d’Hotel, C., Hermosillo, G., and Faugeras, O. (2002). Flows of dif-feomorphisms for multimodal image registration. Proc. IEEE Int. Symp.Biomed. Imag., 753–756.

Chung, M.K., Worsley, K.J., Robbins, S., Paus, T., Taylor, J., Giedd,J.N., Rapoport, J.L., and Evans, A.C. (2003). Deformation-based sur-face morphometry applied to gray matter deformation. Neuroimage18, 198–213.

Chung, M.K., Robbins, S.M., Dalton, K.M., Davidson, R.J., Alexander,A.L., and Evans, A.C. (2005). Cortical thickness analysis in autism withheat kernel smoothing. Neuroimage 25, 1256–1265.

Clouchoux, C., Coulon, O., Riviere, D., Cachia, A., Mangin, J.F., andRegis, J. (2005). Anatomically constrained surface parameterizationfor cortical localization. Med. Image Comput. Comput. Assist. Interv.Int. Conf. Med. Image. Comput. Comput. Assist. Interv. 8, 344–351.

Dale, A.M., Fischl, B., and Sereno, M.I. (1999). Cortical surface-basedanalysis. I. Segmentation and surface reconstruction. Neuroimage 9,179–194.

Denys, K., Vanduffel, W., Fize, D., Nelissen, K., Peuskens, H., Van Es-sen, D., and Orban, G.A. (2004). The processing of visual shape in thecerebral cortex of human and non-human primates: A functional mag-netic resonance imaging study. J. Neurosci. 24, 2551–2565.

Desai, R., Liebenthal, E., Possing, E.T., Waldron, E., and Binder, J.R.(2005). Volumetric vs. surface-based alignment for localization of audi-tory cortex activation. Neuroimage 26, 1019–1029.

Desikan, R.S., Segonne, F., Fischl, B., Quinn, B.T., Dickerson, B.C.,Blacker, D., Buckner, R.L., Dale, A.M., Maguire, R.P., Hyman, B.T.,et al. (2006). An automated labeling system for subdividing the humancerebral cortex on MRI scans into gyral based regions of interest. Neu-roimage 31, 968–980.

Devlin, J.T., and Poldrack, R.A. (2007). In praise of tedious anatomy.Neuroimage 37, 1033–1034.

DeYoe, E.A., Carman, G.J., Bandettini, P., Glickman, S., Wieser, J.,Cox, R., Miller, D., and Neitz, J. (1996). Mapping striate and extrastriatevisual areas in human cerebral cortex. Proc. Natl. Acad. Sci. USA 93,2382–2386.

Dinov, I.D., Mega, M.S., Thompson, P.M., Lee, L., Woods, R.P.,Holmes, C.J., Sumners, D.W., and Toga, A.W. (2000). Analyzing func-tional brain images in a probabilistic atlas: A validation of subvolumethresholding. J. Comput. Assist. Tomogr. 24, 128–138.

Dougherty, R.F., Koch, V.M., Brewer, A.A., Fischer, B., Modersitzki, J.,and Wandell, B.A. (2003). Visual field representations and locations ofvisual areas V1/2/3 in human visual cortex. J. Vis. 3, 586–598.

Eickhoff, S., Stephan, K.E., Mohlberg, H., Grefkes, C., Fink, G.R.,Amunts, K., and Zilles, K. (2005). A new SPM toolbox for combining

Neuron

Primer

probabilistic cytoarchitectonic maps and functional imaging data.Neuroimage 25, 1325–1335.

Eickhoff, S.B., Amunts, K., Mohlberg, H., and Zilles, K. (2006). The hu-man parietal operculum. II. Stereotaxic maps and correlation withfunctional imaging results. Cereb. Cortex 16, 268–279.

Eickhoff, S.B., Paus, T., Caspers, S., Grosbras, M.H., Evans, A.C.,Zilles, K., and Amunts, K. (2007). Assignment of functional activationsto probabilistic cytoarchitectonic areas revisited. Neuroimage 36, 511–521.

Elston, G.N. (2006). Specialization of the neocortical pyramidal cellduring primate evolution. In Evolution of Nervous Systems, J.H.Kaas, ed. (Oxford: Academic Press), pp. 191–241.

Felleman, D.J., and Van Essen, D.C. (1991). Distributed hierarchicalprocessing in primate cerebral cortex. Cereb. Cortex 1, 1–47.

Fischl, B., and Dale, A.M. (2000). Measuring the thickness of thehuman cerebral cortex from magnetic resonance images. Proc. Natl.Acad. Sci. USA 97, 11050–11055.

Fischl, B., Sereno, M.I., Tootell, R.B., and Dale, A.M. (1999). High-res-olution intersubject averaging and a coordinate system for the corticalsurface. Hum. Brain Mapp. 8, 272–284.

Fischl, B., van der Kouwe, A., Destrieux, C., Halgren, E., Segonne, F.,Salat, D.H., Busa, E., Seidman, L.J., Goldstein, J., Kennedy, D., et al.(2004). Automatically parcellating the human cerebral cortex. Cereb.Cortex 14, 11–22.

Fox, P.T., Perlmutter, J.S., and Raichle, M.E. (1985). A stereotacticmethod of anatomical localization for positron emission tomography.J. Comput. Assist. Tomogr. 9, 141–153.

Fox, M.D., Snyder, A.Z., Vincent, J.L., Corbetta, M., Van Essen, D.C.,and Raichle, M.E. (2005). The human brain is intrinsically organized intodynamic, anticorrelated functional networks. Proc. Natl. Acad. Sci.USA 102, 9673–9678.

Friston, K.J., and Henson, R.N. (2006). Commentary on: Divide andconquer; a defence of functional localisers. Neuroimage 30, 1097–1099.

Friston, K.J., Rotshtein, P., Geng, J.J., Sterzer, P., and Henson, R.N.(2006). A critique of functional localisers. Neuroimage 30, 1077–1087.

Goebel, R., Esposito, F., and Formisano, E. (2006). Analysis of func-tional image analysis contest (FIAC) data with brainvoyager QX:From single-subject to cortically aligned group general linear modelanalysis and self-organizing group independent component analysis.Hum. Brain Mapp. 27, 392–401.

Grill-Spector, K., and Malach, R. (2004). The human visual cortex.Annu. Rev. Neurosci. 27, 649–677.

Hadjikhani, N., Liu, A.K., Dale, A.M., Cavanagh, P., and Tootell, R.B.H.(1998). Retinotopy and color sensitivity in human visual cortical areaV8. Nat. Neurosci. 1, 235–241.

Henery, C.C., and Mayhew, T. (1989). The cerebrum and cerebellum ofthe fixed human brain: Efficient and unbiased estimates of volumesand cortical surface areas. J. Anat. 167, 167–180.

Jack, A.I., Patel, G.H., Astafiev, S.V., Snyder, A.Z., Akbudak, E., Shul-man, G.L., and Corbetta, M. (2007). Changing human visual field orga-nization from early visual to extra-occipital cortex. PLoS ONE 2, e452.10.1371/journal.pone.0000452.

Jo, H.J., Lee, J.M., Kim, J.H., Shin, Y.W., Kim, I.Y., Kwon, J.S., andKim, S.I. (2007). Spatial accuracy of fMRI activation influenced by vol-ume- and surface-based spatial smoothing techniques. Neuroimage34, 550–556.

Jouandet, M., Tramo, M., Herron, D., Hermann, A., Loftus, W., Bazell,J., and Gazzaniga, M. (1989). Brainprints: Computer-generated two-dimensional maps of the human cerebral cortex in vivo. J. Cogn. Neu-rosci. 1, 88–117.

Klein, A., Mensh, B., Ghosh, S., Tourville, J., and Hirsch, J. (2005).Mindboggle: Automated brain labeling with multiple atlases. BMCMed. Imaging 5, 7.

Klein, J.C., Behrens, T.E.J., Robson, M.D., Mackay, C.E., Higham,D.J., and Johansen-Berg, H. (2007). Connectivity-based parcellationof human cortex using diffusion MRI: Establishing reproducibility,validity and observer independence in BA 44/45 and SMA/pre-SMA.Neuroimage 34, 204–211.

Laird, A.R., McMillan, K.M., Lancaster, J.L., Kochunov, P., Turkeltaub,P.E., Pardo, J.V., and Fox, P.T. (2005). A comparison of label-based re-view and activation likelihood estimation in the Stroop task. Hum. BrainMapp. 25, 6–21.

Lancaster, J.L., Woldorff, M.G., Parsons, L.M., Liotti, M., Freitas, C.S.,Rainey, L., Kochunov, P.V., Nickerson, D., Mikiten, S.A., and Fox, P.T.(2000). Automated Talairach atlas labels for functional brain mapping.Hum. Brain Mapp. 10, 120–131.

Lancaster, J.L., Tordesillas-Gutierrez, D., Martinez, M., Salinas, F.,Evans, A., Zilles, K., Mazziotta, J.C., and Fox, P.T. (2007). Bias be-tween MNI and Talairach coordinates analyzed using the ICBM-152brain template. Hum. Brain Mapp. Published online Janary 31, 2007.10.1002/hbm.20345.

Larsson, J., and Heeger, D.J. (2006). Two retinotopic visual areas inhuman lateral occipital cortex. J. Neurosci. 26, 13128–13142.

Lewis, J.W., and Van Essen, D.C. (2000). Mapping of architectonicsubdivisions in the macaque monkey, with emphasis on parieto-occip-ital cortex. J. Comp. Neurol. 428, 79–111.

Lewis, J.W., Beauchamp, M.S., and DeYoe, E.A. (2000). A comparisonof visual and auditory motion processing in human cerebral cortex. Ce-rebral Cortex. 10, 873–888.

Lyttelton, O., Boucher, M., Robbins, S., and Evans, A. (2007). An unbi-ased iterative group registration template for cortical surface analysis.Neuroimage 34, 1535–1544.

Makris, N., Schlerf, J.E., Hodge, S.M., Haselgrove, C., Albaugh, M.D.,Seidman, L.J., Rauch, S.L., Harris, G., Biederman, J., Caviness, V.S.,Jr., et al. (2005). MRI-based surface-assisted parcellation of humancerebellar cortex: An anatomically specified method with estimate ofreliability. Neuroimage 25, 1146–1160.

Mangin, J.F., Riviere, D., Cachia, A., Duchesnay, E., Cointepas, Y., Pa-padopoulos-Orfanos, D., Scifo, P., Ochiai, T., Brunelle, F., and Regis,J. (2004). A framework to study the cortical folding patterns. Neuro-image 23 (Suppl 1), S129–S138.

Margulies, D.S., Kelly, A.M., Uddin, L.Q., Biswal, B.B., Castellanos,F.X., and Milham, M.P. (2007). Mapping the functional connectivity ofanterior cingulate cortex. Neuroimage 37, 579–588.

Maunsell, J.H.R., and Van Essen, D.C. (1987). The topographic organi-zation of the middle temporal visual area in the macaque monkey: rep-resentational biases and the relationship to callosal connections andmyeloarchitectonic boundaries. J. Comp. Neurol. 266, 535–555.

Mazziotta, J., Toga, A., Evans, A., Fox, P., Lancaster, J., Zilles, K.,Woods, R., Paus, T., Simpson, G., Pike, B., et al. (2001). A probabilisticatlas and reference system for the human brain: International Consor-tium for Brain Mapping (ICBM). Philos. Trans. R. Soc. Lond. B Biol. Sci.356, 1293–1322.

Mega, M.S., Dinov, I.D., Mazziotta, J.C., Manese, M., Thompson,P.M., Lindshield, C., Moussai, J., Tran, N., Olsen, K., Zoumalan, C.I.,et al. (2005). Automated brain tissue assessment in the elderly and de-mented population: Construction and validation of a sub-volume prob-abilistic brain atlas. Neuroimage 26, 1009–1018.

Mikula, S., Trotts, I., Stone, J.M., and Jones, E.G. (2007). Internet-en-abled high-resolution brain mapping and virtual microscopy. Neuro-image 35, 9–15.

Nelissen, K., Luppino, G., Vanduffel, W., Rizzolatti, G., and Orban, G.A.(2005). Observing others: Multiple action representation in the frontallobe. Science 310, 332–336.

Neuron 56, October 25, 2007 ª2007 Elsevier Inc. 223

Neuron

Primer

Nordahl, C.W., Dierker, D., Mostafavi, I., Schumann, C.M., Rivera, S.,Amaral, D.G., and Van Essen, D.C. (2007). Cortical folding abnormali-ties in children with autism revealed by surface-based morphometry. J.Neurosci., in press.

O’Kusky, J., and Colonnier, M.A. (1982). Laminar analysis of the num-ber of neurons, glia, and synapses in the adult cortex (area 17) of adultmacaque monkeys. J. Comp. Neurol. 210, 278–290.

Ongur, D., Ferry, A.T., and Price, J.L. (2003). Architectonic subdivisionof the human orbital and medial prefrontal cortex. J. Comp. Neurol.460, 425–449.

Ono, M., Kubick, S., and Abernathey, C.D. (1990). Atlas of the CerebralSulci (New York: Thieme Medical).

Orban, G.A., Van Essen, D., and Vanduffel, W. (2004). Comparativemapping of higher visual areas in monkeys and humans. TrendsCogn. Sci. 8, 315–324.

Paxinos, G., Huang, X.-F., and Toga, A.W. (2000). The Rhesus MonkeyBrain in Stereotaxic Coordinates (London: Academic Press).

Petrides, M. (2005). Lateral prefrontal cortex: Architectonic and func-tional organization. Philos. Trans. R. Soc. Lond. B Biol. Sci. 360,781–789.

Poldrack, R.A., and Devlin, J.T. (2007). On the fundamental role ofanatomy in functional imaging: Reply to commentaries on ‘‘In praiseof tedious anatomy’’. Neuroimage 37, 1066–1068.

Rademacher, J., Caviness, V.S., Jr., Steinmetz, H., and Galaburda,A.M. (1993). Topographical variation of the human primary cortices:Implications for neuroimaging, brain mapping, and neurobiology.Cereb. Cortex 3, 313–329.

Robbins, S., Evans, A.C., Collins, D.L., and Whitesides, S. (2004). Tun-ing and comparing spatial normalization methods. Med. Image Anal. 8,311–323.

Saad, Z.S., Reynolds, R.C., Argall, B., Japee, S., and Cox, R.W. (2004).SUMA: An interface for surface-based intra- and inter-subject analysiswith AFNI. Proc. 2004 IEEE Int. Symp. Biomed. Imaging, 1510–1513.

Saleem, K.S., and Logothetis, N.K. (2006). A Combined MRI and His-tology Atlas of the Rhesus Monkey Brain (London: Academic Press).

Saxe, R., Brett, M., and Kanwisher, N. (2006). Divide and conquer: Adefense of functional localizers. Neuroimage 30, 1088–1096.

Sereno, M.I., and Tootell, R.B. (2005). From monkeys to humans: Whatdo we now know about brain homologies? Curr. Opin. Neurobiol. 15,135–144.

Sereno, M.I., Dale, A.M., Reppas, J.B., Kwong, K.K., Belliveau, J.W.,Brady, T.J., Rosen, B.R., and Tootell, R.B. (1995). Borders of multiplevisual areas in humans revealed by functional magnetic resonance im-aging. Science 268, 889–893.

Shattuck, D.W., Mirza, M., Adisetiyo, V., Hojatkashani, C., Salamon,G., Narr, K.L., Poldrack, R.A., Bilder, R.M., and Toga, Q.W. (2007).Construction of a 3D probabilistic atlas of human brain. Neuroimage,in press.

Shi, Y., Thompson, P.M., Dinov, I., Osher, S., and Toga, A.W. (2007).Direct cortical mapping via solving partial differential equations on im-plicit surfaces. Med. Image Anal. 11, 207–223.

Stensaas, S.S., Eddington, D.K., and Dobelle, W.H. (1974). The topog-raphy and variability of the primary visual cortex in man. J. Neurosurg.40, 747–755.

Swisher, J.D., Halko, M.A., Merabet, L.B., McMains, S.A., and Somers,D.C. (2007). Visual topography of human intraparietal sulcus. J. Neuro-sci. 27, 5326–5337.

Talairach, J., and Tournoux, P. (1988). Coplanar Stereotaxic Atlas ofthe Human Brain (New York: Thieme Medical).

Thompson, P.M., Woods, R.P., Mega, M.S., and Toga, A.W. (2000).Mathematical/computational challenges in creating defromable andprobabilistic atlases of the human brain. Hum. Brain Mapp. 9, 81–92.

224 Neuron 56, October 25, 2007 ª2007 Elsevier Inc.

Thompson, P.M., Vidal, C., Giedd, J.N., Gochman, P., Blumenthal, J.,Nicolson, R., Toga, A.W., and Rapoport, J.L. (2001). Mapping adoles-cent brain change reveals dynamic wave of accelerated gray matterloss in very early-onset schizophrenia. Proc Natl Acad Sci U S A 98,11650–11655.

Toga, A.W., and Thompson, P.M. (2001). Maps of the brain. Anat. Rec.265, 37–58.

Toga, A.W., and Thompson, P.M. (2002). New approaches in brainmorphometry. Am. J. Geriatr. Psychiatry 10, 13–23.

Toga, A.W., and Thompson, P.M. (2003). Temporal dynamics of BrainAnatomy. Annu. Rev. Biomed. Eng. 5, 119–145.

Toga, A.W., and Thompson, P.M. (2005). Brain atlases of normal anddiseased populations. Int. Rev. Neurobiol. 66, 1–54.

Toga, A.W., Thompson, P.M., Mega, M.S., Narr, K.L., and Blanton,R.E. (2001). Probabilistic approaches for atlasing normal and dis-ease-specific brain variations. Anat. Embryol. (Berl.) 204, 267–282.

Toga, A.W., Thompson, P.M., Mori, S., Amunts, K., and Zilles, K.(2006). Towards multimodal atlases of the human brain. Nat. Rev. Neu-rosci. 7, 952–966.

Tootell, R.B., Tsao, D., and Vanduffel, W. (2003). Neuroimaging weighsin: Humans meet macaques in ‘‘primate’’ visual cortex. J. Neurosci. 23,3981–3989.

Tramo, M.J., Loftus, W.C., Thomas, C.E., Green, R.L., Mott, L.A., andGazzaniga, M.S. (1995). Surface area of human cerebral cortex and itsgross morphological subdivisions: In vivo measurements in monozy-gotic twins suggest differential hemisphere effects of genetic factors.J. Cogn. Neurosci. 7, 292–302.

Tsao, D.Y., Freiwald, W.A., Tootell, R.B., and Livingstone, M.S. (2006).A cortical region consisting entirely of face-selective cells. Science311, 670–674.

Turkeltaub, P.E., Eden, G.F., Jones, K.M., and Zeffiro, T.A. (2002).Meta-analysis of the functional neuroanatomy of single-word reading:Method and validation. Neuroimage 16, 765–780.

Tzourio-Mazoyer, N., Landeau, B., Papathanassiou, D., Crivello, F.,Etard, O., Delcroix, N., Mazoyer, B., and Joliot, M. (2002). Automatedanatomical labeling of activations in SPM using a macroscopic ana-tomical parcellation of the MNI MRI single-subject brain. Neuroimage1, 273–289.

van Atteveldt, N., Formisano, E., Goebel, R., and Blomert, L. (2004).Integration of letters and speech sounds in the human brain. Neuron43, 271–282.

Van Essen, D.C. (1997). A tension-based theory of morphogenesis andcompact wiring in the central nervous system. Nature 385, 313–318.

Van Essen, D.C. (2002). Windows on the brain. The emerging role of at-lases and databases in neuroscience. Curr. Opin. Neurobiol. 12, 574–579.

Van Essen, D.C. (2004a). Surface-based approaches to spatial locali-zation and registration in primate cerebral cortex. Neuroimage 23,S97–S107.

Van Essen, D.C. (2004b). Organization of visual areas in macaque andhuman cerebral cortex. In The Visual Neurosciences, L. Chalupa andJ.S. Werner, eds. (Cambridge, MA: MIT Press), pp. 507–521.

Van Essen, D.C. (2005a). A population-average, landmark- and sur-face-based (PALS) atlas of human cerebral cortex. Neuroimage 28,635–662.

Van Essen, D.C. (2005b). Surface-based comparisons of macaqueand cortical organization. In From Monkey to Brain: Proceedings ofa Fyssen Foundation Colloquium, June, 2003, S. Dehaene, ed. (Cam-bridge, MA: MIT Press), pp. 3–20.

Van Essen, D.C. (2006). Cerebral cortical folding patterns in primates:Why they vary and what they signify. In Evolution of Nervous Systems,J.H. Kaas, ed. (Oxford: Academic Press), pp. 267–276.

Neuron

Primer

Van Essen, D.C., and Dierker, D. (2007). On navigating the human cor-tex. Neuroimage 37, 1050–1054.

Van Essen, D.C., Newsome, W.T., and Maunsell, J.H.R. (1984). Thevisual field representation in striate cortex of the macaque monkey:Asymmetries, anisotropies and individual variability. Vision Res. 24,429–448.

Van Essen, D.C., Dickson, J., Harwell, J., Hanlon, D., Anderson, C.H.,and Drury, H.A. (2001). An integrated software system for surface-based analyses of cerebral cortex. J. Am. Med. Inform. Assoc. 8,443–459.

Van Essen, D.C., Harwell, J., Hanlon, D., and Dickson, J. (2005).Surface-Based Atlases and a Database of Cortical Structure andFunction. In Databasing the Brain: From Data to Knowledge (Neuroin-formatics), S.H. Koslow and S. Subramaniam, eds. (Hoboken, NJ:John Wiley & Sons), pp. 369–387.

Van Essen, D.C., Dierker, D., Snyder, A., Raichle, M.E., Reiss, A., andKorenberg, J. (2006). Symmetry of cortical folding abnormalities inWilliams syndrome revealed by surface-based analyses. J. Neurosci.26, 5470–5483.

Vincent, J.L., Patel, G.H., Fox, M.D., Snyder, A.Z., Baker, J.T., VanEssen, D.C., Zempel, J.M., Snyder, L.H., Corbetta, M., and Raichle,M.E. (2007). Intrinsic functional architecture in the anesthetized mon-key brain. Nature 447, 83–86.

Wandell, B.A., Brewer, A.A., and Dougherty, R.F. (2005). Visual fieldmap clusters in human cortex. Philos. Trans. R. Soc. Lond. B Biol.Sci. 360, 693–707.

Wandell, B.A., Dumoulin, S.O., and Brewer, A.A. (2007). Visual fieldmaps in human cortex. Neuron 56, this issue, 366–383.

Wisco, J.J., Kuperberg, G., Manoach, D., Quinn, B.T., Busa, E., Fischl,B., Heckers, S., and Sorensen, A.G. (2007). Abnormal cortical foldingpatterns within Broca’s area in schizophrenia: Evidence from structuralMRI. Schizophr. Res. 94, 317–327.

Yeo, B.T.T., Sabuncu, M., Mohlberg, H., Amunts, K., Zilles, K., Golland,P., and Fischl, B. (2007). Proceedings of the IEEE Computer SocietyWorkshop on Mathematical Methods in Biomedical Image Analysis(MMBIA 2007).

Zacks, J. (2007). Neuroimaging studies of mental rotation: A meta-analysis and review. J. Cogn. Neurosci., in press.

Neuron 56, October 25, 2007 ª2007 Elsevier Inc. 225