Maps of the Brainusers.loni.ucla.edu/~thompson/PDF/TT_NewAnat2001.pdfREVIEW Maps of the Brain ARTHUR W. TOGA* AND PAUL M. THOMPSON We review recent developments in brain mapping and

REVIEW

Maps of the BrainARTHUR W. TOGA* AND PAUL M. THOMPSON

We review recent developments in brain mapping and computational anatomy that have greatly expanded our abilityto analyze brain structure and function. The enormous diversity of brain maps and imaging methods has spurred thedevelopment of population-based digital brain atlases. These atlases store information on how the brain varies acrossage and gender, across time, in health and disease, and in large human populations. We describe how brain atlases,and the computational tools that align new datasets with them, facilitate comparison of brain data acrossexperiments, laboratories, and from different imaging devices. The major methods are presented for the constructionof probabilistic atlases, which store information on anatomic and functional variability in a population. Algorithms arereviewed that create composite brain maps and atlases based on multiple subjects. We show that group patterns ofcortical organization, asymmetry, and disease-specific trends can be resolved that may not be apparent in individualbrain maps. Finally, we describe the creation of four-dimensional (4D) maps that store information on the dynamicsof brain change in development and disease. Digital atlases that correlate these maps show considerable promise inidentifying general patterns of structural and functional variation in human populations, and how these featuresdepend on demographic, genetic, cognitive, and clinical parameters. Anat Rec (New Anat) 265:37–53, 2001.© 2001 Wiley-Liss, Inc.

KEY WORDS: biomedical imaging; brain mapping; cryosection imaging; neuroanatomy; atlas; neuroimaging; Alzheimerdisease

The widespread collection of normaland diseased brain images in vivo andex vivo enables a tremendous increasein the number of investigations focus-ing on the structural and functionalorganization of the brain. The com-plexity and variability of human brain(in particular) across subjects is sogreat that reliance on maps is essen-tial to effectively analyze and interpret

brain data. Design of appropriatemaps for human brain data presentsconsiderable challenges, becausethese systems must capture how brainstructure and function vary across ageand gender, in different disease states,across imaging modalities, and evenacross species.

This study introduces brain maps asapplied to a variety of research areasin health and disease. It includes abrief survey of the types of maps andto what questions they are applied.First, there is an overview of brainmapping, describing the elements of amap as distinct from an image. Next,there follows a discussion of coordi-nate systems and how they referencesource data from individuals, fol-lowed by a description of deformableatlases and how they can adapt to andmeasure variability across subjectsand systems. We introduce maps thatdescribe relationships between differ-ent observations, such as across modal-ities or with a changing morphology.Finally, we describe the categoriza-tion of subpopulations.

MAPS AND ATLASES

Diversity of Brain Maps

Comprehensive maps of brain struc-ture have been created at a variety ofspatial scales. These are based upon3D tomographic images (Damasio,1995), anatomic specimens (Talairachand Szikla, 1967; Talairach and Tour-noux, 1988; Ono et al., 1990; Duver-noy, 1991) and different histologicpreparations that reveal regional cyto-architecture (Brodmann, 1909) andregional molecular content such asmyelination patterns (Smith, 1907;Mai et al., 1997), receptor bindingsites (Geyer et al., 1997), protein den-sities and mRNA distributions. Otherbrain maps have concentrated onfunction, quantified by positron emis-sion tomography (PET; Minoshima etal., 1994), functional MRI (Le Bihan,1996) or electrophysiology (Avoli etal., 1991; Palovcik et al., 1992). Addi-tional maps have been developed torepresent neuronal connectivity andcircuitry (Van Essen and Maunsell,1983), based on compilations of em-

Dr. Toga is Professor of Neurology, Di-rector of the Laboratory of Neuro Imag-ing, and Co-Director of the Division ofBrain Mapping at the UCLA School ofMedicine. He is also founder and editor-in-chief of the journal, NeuroImage. Dr.Thompson is Assistant Professor ofNeurology and member of the Labora-tory of Neuro Imaging. He is the recipi-ent of numerous awards including theDi Chiro Outstanding Scientific Paper(1998) and the Eiduson NeuroscienceAward. They are part of an interdiscipli-nary group focused on the imaging,mapping, and modeling of brain struc-ture and function in health and disease.*Correspondence to: Arthur W. Toga,Laboratory of Neuro Imaging, Depart-ment of Neurology, UCLA School ofMedicine, Rm. 4238, Reed NeurologicalResearch Center, 710 Westwood Plaza,Los Angeles CA 90095-1769. Fax: 310-206-5518. E-mail: [email protected]

THE ANATOMICAL RECORD (NEW ANAT.) 265:37–53, 2001

© 2001 Wiley-Liss, Inc.

pirical evidence (Brodmann, 1909;Berger, 1929; Penfield and Boldrey,1937).

Each of these brain maps has a dif-ferent spatial scale and resolution,emphasizes different functional orstructural characteristics, and none isinherently compatible with any other.Each strategy clearly has its placewithin a collective effort to map thebrain, but unless certain precautionsare taken enabling common registra-tion, these brain maps will remain asindividual and independent efforts,incomparable with each other.

Brain Atlases

To address these difficulties, brain at-lases provide a structural frameworkin which individual brain maps can beintegrated. Most brain atlases arebased on a single subject’s anatomy ina standardized 3D coordinate system,or stereotaxic space. The chosen dataset acts as a template on which otherbrain maps (such as functional im-ages) can be overlaid. The anatomicdata provides the additional detailnecessary to accurately localize acti-vation sites, as well as providing otherstructural perspectives such as che-moarchitecture. Digital mapping ofstructural and functional image datainto a common 3D coordinate space isa prerequisite for many types of brainimaging research, as it supplies aquantitative spatial reference systemin which brain data from multiplesubjects and modalities can be com-pared and correlated.

Given the fact that there is neither asingle (accepted) representative brainnor a simple method to construct an“average”, the construction of brainatlases to represent large human pop-ulations has become the focus of in-tense research. Deformable atlases,which can be adapted to reflect theanatomy of new subjects, and proba-bilistic atlases, which retain informa-tion on population variability, arepowerful approaches with a range ofclinical and research applications.These atlases can be used to supportpathology detection in individual sub-jects or groups. Single modality at-lases also ultimately may be insuffi-cient, because of the need to establishthe relationship between differentmeasurements of anatomy and physi-

ology. In response to these challenges,multi-modal atlases combine detailedstructural maps from multiple imag-ing sensors in the same 3D coordinatespace. Multi-modal atlases providethe best of all worlds, offering a real-istically complex representation ofbrain morphology and function in itsfull spatial and multidimensionalcomplexity.

Early Brain Atlases

Brain atlases were (are) built becauseaccurate localization of brain struc-ture and function in any modality maybe improved by correlation withhigher resolution anatomic dataplaced in an appropriate spatial coor-dinate system. Three-dimensional

neuroanatomic templates also havethe potential to provide important ref-erence information when planningstereotaxic surgical procedures, in-cluding radiosurgery and electrodeimplantations (Talairach and Szikla,1967; Kikinis et al., 1996). Most earlyatlases of the human brain, and otherspecies (Paxinos and Watson, 1986;Swanson, 1992), were derived fromone, or at best a few, individual postmortem specimens (Brodmann, 1909;Schaltenbrand and Bailey, 1959; Ta-lairach and Szikla, 1967; Schalten-brand and Wahren, 1977; Matsui andHirano, 1978; Talairach and Tour-noux, 1988; Ono et al., 1990). Suchatlases take the form of anatomicalreferences or represent a particularfeature of the brain (Van Buren and

Maccubin, 1962; Van Buren andBorke, 1972), such as a specific neu-rochemical distribution (Mansour etal., 1995) or the cellular architectureof the cerebral cortex (Brodmann,1909). None, to date, establishes a re-lationship between any two or morefeatures of the brain. Commonly usedhuman atlases include those of Ta-lairach and Tournoux (1988) and thethalamic and brainstem anatomicalmaps of Schaltenbrand and Wahren(1977). Due to individual variations inanatomy among normal subjects, pro-portional scaling systems are typicallyemployed to reference a given brain toan atlas brain (Talairach and Tour-noux, 1988). More sophisticated elas-tic or fluid transformations, involvinglocal matching, are described below.These approaches locally deform adigital atlas to reflect the anatomy ofnew subjects.

MRI-Based Atlases

Recent atlases based on magnetic res-onance image (MRI) data have the ad-vantage of intrinsic three-axis regis-tration and spatial coordinates(Damasio, 1995), but have relativelylow resolution and lack anatomic con-trast in important subregions. TheHarvard Brain Atlas, based on a 1 31 3 1.5 mm resolution 3D SPGR(spoiled gradient-recalled acquisition)scan of a 25-year-old, normal subject,was enhanced by anisotropic diffu-sion filtering (Kikinis et al., 1996; dif-fusion MRI described in Mori andBarker, 1999), before being seg-mented into 150 hand-labeled re-gions, which include white mattertracts as well as major neuroanatomicstructures. Nevertheless, many high-resolution MR atlases, with up to100–150 slices, a section thickness of2 mm, and 2562 pixel imaging planes(Evans et al., 1991; Lehmann et al.,1991) still result in resolutions lowerthan the complexity of many neuro-anatomic structures. A recent innova-tion in the collection of atlas qualityMR (Fig. 1) utilizes multiple scans(N 5 27) and a registered average of asingle individual to overcome the lackof contrast and relatively poor signalto noise (Holmes et al., 1998). Theresulting volumetric datasets canserve as the basis for generating de-

Design of appropriatemaps for human brain

data presentsconsiderable

challenges, becausethese systems mustcapture how brain

structure and functionvary across age andgender, in differentdisease states, etc.

38 THE ANATOMICAL RECORD (NEW ANAT.) REVIEW

tailed 3-dimensional models of ana-tomical structures, represented usinga variety of volumetric and graphicalsurface formats (Fig. 2). The modelscan then be viewed either interac-tively, or off-line, using volume ren-dering and ray-tracing algorithms. Se-lective transparency can be applied tothe models and animated fly-throughsgenerated to assist in visualizing thecomplex relationships among ana-tomical systems.

Cryosection Imaging

Several digital atlases have been de-veloped using photographic imagesof cryoplaned frozen specimens(Bohm et al., 1983; Greitz et al.,1991). Photographed material, al-though providing superior anatomicdetail, has limitations. For accurate

correlations, data must be placed ina plane equivalent to that of the im-age of interest. Digital imaging, how-ever, overcomes many limitations ofconventional film photography. Us-ing 1,0242, 24-bits/pixel digital colorcameras, spatial resolution can be ashigh as 100 microns/voxel for wholehuman head cadaver preparations,or higher for isolated brain regions(Toga et al., 1994). Cryosectioning inmicron increments permits data col-lection with high spatial resolutionin the axis orthogonal to the section-ing plane. Acquisition of images inseries directly from the consistentlypositioned cryoplaned blockfacealso avoids the need for serial imageregistration before reconstruction.Serial images can be reconstructedto a 3D anatomic volume that is

amenable to various resampling andpositioning schemes.

Multi-Modality Atlases

Characterizing a single subject withmultiple imaging devices clearly com-bines the strengths of each imagingmodality. In the Visible HumanProject (Spitzer et al., 1996; Spitzerand Whitlock, 1998), two (male andfemale) cadavers were cryoplaned andimaged at 1.0 mm intervals, and theentire bodies were also reconstructedvia 5,000 post mortem CT and MRIimages. The resulting digital datasets,consist of over 15 gigabytes of imagedata.

Although not an atlas per se, theVisible Human imagery has sufficientquality and accessibility to make it atest platform for developing methodsand standards (Spitzer et al., 1996).The data has served as the foundationfor developing related atlases of re-gions of the cerebral cortex (Druryand Van Essen, 1997), and high-qual-ity brain models and visualizations(Schiemann et al., 1996; Stewart et al.,1996). Using multi-modality datafrom a patient with a localized pathol-ogy, and more recently the Visible Hu-man data, Hohne and co-workers de-veloped a commercially availablebrain atlas designed for teaching neu-roanatomy (VOXEL-MAN; Hohne etal., 1990, 1992; Tiede et al., 1993;Pommert et al., 1994). Data from sin-gle subjects, premortem and postmor-tem, provides a unique view into therelationship between in vivo imagingand histologic assessment. Mega et al.(1997) scanned Alzheimer patients inthe terminal stages of their disease us-ing both MRI and PET. These datawere combined with post mortem 3Dhistological images showing the grossanatomy (Toga et al., 1994) and a Gal-lyas stain of neurofibrillary tangles.This multimodal, but single subject,atlas of Alzheimer disease relates theanatomic and histopathological un-derpinnings to in vivo metabolic andperfusion maps of this disease.

REFERENCE SYSTEMS

Matching a Brain to an Atlas

In existing atlases, spatial normaliza-tion systems are typically employed toreference a given brain with an atlas

Figure 1. High-resolution anatomy from MRI using registration and signal averaging. Usingadvances in image processing, atlas-quality MRI datasets can be generated with superiorcontrast and signal-to-noise than typically attainable using tomographic techniques. Twen-ty-seven volumetric MRI scans were acquired from the same individual, histogram-equal-ized, and mutually registered. Signal intensities then were averaged voxel-by-voxel acrossall images to produce a signal high-contrast MRI dataset. Individual nuclei of the basalganglia (a) and its striations (b), as well as deep white matter commissures (c) and thefoliations of the cerebellum (d) can be observed in detail usually attainable only withwhole-brain cryosection imagery (Toga et al., 1994). High-resolution MRI datasets areadvantageous as they can be digitally overlaid, using automated algorithms, onto lowerresolution structural or functional data to assist in image interpretation and analysis [datafrom Holmes et al. (1998)].

REVIEW THE ANATOMICAL RECORD (NEW ANAT.) 39

brain (Talairach and Szkilra, 1967;Talairach and Tournoux, 1988). Thisallows individual data to be superim-posed on the data in the atlas, in otherwords, to be transformed to match thespace occupied by the atlas. In theTalairach stereotaxic system, piece-wise affine transformations are ap-plied to 12 rectangular regions ofbrain, defined by vectors from the an-terior and posterior commissures tothe extrema of the cortex. These trans-formations re-position the anteriorcommissure of the subject’s scan atthe origin of the 3D coordinate space,vertically align the interhemisphericplane, and horizontally orient the lineconnecting the two commissures.Each point in the incoming brain im-age, after it is “warped” into the atlasspace, is labeled by an (x,y,z) addressreferable to the atlas brain. Although

originally developed to help interpretbrain stem and ventricular studies ac-quired using pneumoencephalogra-phy (Talairach and Szlikra, 1967), theTalairach stereotaxic system rapidlybecame an international standard forreporting functional activation sites inPET studies, allowing researchers tocompare and contrast results fromdifferent laboratories (Fox et al., 1985,1988; Friston et al., 1989, 1991).

The success of any brain atlas de-pends on how well the anatomies ofindividual subjects match the repre-sentation of anatomy in the atlas.Whereas stereotaxic methods providea common coordinate system forpooling activation data and multi-sub-ject comparisons, the accuracy andutility of the atlas is equally depen-dent on the anatomical template itself(Roland and Zilles, 1994). The Ta-

lairach templates are based on postmortem sections of the brain of a 60-year-old female subject that clearlydid not reflect the in vivo anatomy ofsubjects in activation studies. The at-las plates also have a variable sliceseparation (3 to 4 mm), and data fromorthogonal planes are inconsistent.

To address these limitations, a com-posite MRI dataset was constructedfrom several hundred young normalsubjects (239 males, 66 females; age:23.4 6 4.1 years) whose scans wereindividually mapped into the Ta-lairach system by linear transforma-tion, intensity normalized, and aver-aged on a voxel-by-voxel basis (Evanset al., 1992, 1994). Although the re-sulting average brain has regionswhere individual structures areblurred out due to spatial variabilityin the population (Evans et al., 1992;1994), the effect of anatomical vari-ability in different brain areas is illus-trated qualitatively by this map. Theaverage intensity template is part ofthe widely used Statistical ParametricMapping package (SPM; Friston et al.,1995). The development of automatedmethods to map new MRI and PETdata into a common space could beused to optimally align new MR datawith the template by maximizing ameasure of intensity similarity, suchas 3D cross-correlation (Collins et al.,1994a, 1995), ratio image uniformity(Woods et al., 1992), or mutual infor-mation (Viola and Wells, 1995; Wellset al., 1997). Any alignment transfor-mation defined for one modality, suchas MRI, can be identically applied toanother modality, such as PET, if aprevious cross-modality intrasubjectregistration has been performed(Woods et al., 1993). For the first timethen, PET data could be mapped intostereotaxic space via a correlated MRdataset (Woods et al., 1993; Evans etal., 1994). Registration algorithmstherefore made it feasible to automat-ically map data from a variety of mo-dalities into an atlas coordinate spacebased directly on the Talairach refer-ence system.

DEFORMABLE ATLASES

Cross-Subject AnatomicVariations

The use of spatial normalizationschemes based upon deep white mat-

Figure 2. Computational models of anatomy. Using a 3D coordinate system to representstructure locations, detailed models of anatomical substructures can be created. Thesemodels are visualized using a combination of 3D volume and surface rendering techniques.Structures such as the cortex can be represented using a connected mesh of triangulartiles, and their reflectance can be made to depend on their surface normals to give theimpression of depth. User-defined attributes such as selective transparency, color, andtextures may also be applied before visualization. This model shows different componentsof the limbic system, generated using both automated and manual techniques from thehigh-resolution MRI dataset shown in Figure 1. The resulting graphical models can bedatabased and indexed to anatomical nomenclature as components of a digital atlas(Toga et al., 1996; Thompson et al., 2000), and can serve as components of more complexscientific visualizations and animations (Thompson and Toga, 1997; Toga and Thompson,1998).


ter features (the AC and PC), such asoutlined above, has yet to completelyaccommodate the most variable brainstructure, the cortex. The cortex isalso the site of interest for most func-tional activation studies. Considerablenormal variation in sulcal geometryhas been found in primary motor, so-matosensory and auditory cortex(Missir et al., 1989; Rademacher et al.,1993), primary and association visualcortex (Stensaas et al., 1974), frontaland pre-frontal areas (Rajkowska andGoldman-Rakic, 1995), and lateralperisylvian cortex (Geschwind andLevitsky, 1968; Steinmetz et al., 1989,1990; Ono et al., 1990). More recent3D analyses of anatomic variability, inpost mortem, in vivo normal, and dis-eased populations, have found ahighly heterogeneous pattern of ana-tomic variation (Thompson et al.,1996b, 1998a).

In view of the complex structuralvariability between individuals, afixed brain atlas may fail to serve as afaithful representation of every brain(Roland and Zilles, 1994; Mazziotta etal., 1995). Because no two brains arethe same, this presents a challenge forattempts to create standardized at-lases. Even in the absence of any pa-thology, brain structures vary be-tween individuals not only in shapeand size, but also in their orientationsrelative to each other. Such normalvariations have also complicated thegoals of comparing functional and an-atomic data from many subjects(Rademacher et al., 1993; Roland andZilles, 1994).

Numerous studies have determinedhow severe the inter-subject varia-tions in anatomy are, even after trans-forming individual anatomic data intothe Talairach stereotaxic system.Clearly, direct averaging of digitalbrain maps, after transformation to acommon 3D coordinate space, is onlyvalid if homologous cortical regions indifferent subjects have been broughtinto register by the spatial normaliza-tion transformation. Extreme varia-tions in cortical patterns, observed innormal subjects and exacerbated indisease states by additional patho-logic change, suggest that caution isnecessary in selecting the transforma-tion system to support cross-subjectand cross-group comparisons of cor-

tically-derived events or functionalmaps. The most severe challenge oc-curs when the topology itself is under-going considerable change due to de-velopment or degeneration, forexample. Direct digital subtraction ofstereotaxic functional maps in studiesof disease states, such as dementia,may lead to spurious results: maps ofapparent significance may reflect dif-ferences that are anatomic, ratherthan functional, in character (Meltzerand Frost, 1994; Woods, 1996). Thesedifficulties have led to the suggestionthat direct reference to the sulci thatframe architectonic fields may presenta more reliable basis for functionalmapping than reference to a singlestandard or idealized brain (Stein-metz et al., 1990; Rademacher et al.,1993; Watson et al., 1993; Thompsonet al., 1996, 1998).

Atlas to Brain Transformations

Image warping algorithms, specifi-cally designed to handle 3D neuroana-tomic data (Christensen et al., 1993;1996; Collins et al., 1994a, 1995; Rab-bitt et al., 1995; Thirion, 1995; Bro-Nielsen and Gramkow, 1996; Davatzi-kos, 1996; Thompson and Toga, 1996)can be used to transfer all the infor-mation in a 3D digital brain atlas ontothe scan of any given subject, whilerespecting the intricate patterns ofstructural variation in their anatomy.These transformations must allowany segment of the atlas anatomy,however small, to grow, shrink, twistand even rotate, to produce a transfor-mation that represents and encodeslocal differences in topography fromone individual to another. Such de-formable atlases (Bajcsy and Kovacic,1989; Seitz et al., 1990; Evans et al.,

1991; Gee et al., 1993; Christensen etal., 1993; Miller et al., 1993; Sandorand Leahy, 1994; 1995; Rizzo et al.,1995) can be used to carry 3D maps offunctional and vascular territoriesinto the coordinate system of differentsubjects, as well as information on dif-ferent tissue types and the boundariesof cytoarchitectonic fields and theirneurochemical composition.

How Can One Brain BeDeformed to Match Another?

Any successful warping transform forcross-subject registration of braindata must be high-dimensional, to ac-commodate fine anatomic variations(Joshi et al., 1995; Christensen et al.,1996; Thompson and Toga, 1998).This warping is required to bring theatlas anatomy into structural corre-spondence with the target scan at avery local level (Fig. 3). Another diffi-culty arises from the fact that the to-pology and connectivity of the de-forming atlas have to be maintainedunder these complex transforms. Thisis hard or simply impossible toachieve in traditional image warpingmanipulations (Christensen et al.,1995a,b). Physical continuum modelsof the deformation address these dif-ficulties by considering the deformingatlas image to be embedded in a three-dimensional deformable medium,which can be either an elastic mate-rial or a viscous fluid. The medium issubjected to certain distributed inter-nal forces, which reconfigure the me-dium and eventually lead the image tomatch the target. These forces can bebased mathematically on the local in-tensity patterns in the datasets, withlocal forces designed to match imageregions of similar intensity.

SUBSYSTEM MAPS

The development of maps for anatom-ical subsystems, such as the somato-sensory or motor systems, is typicallydriven by the need to catalogue func-tional activation studies or to describepathways of connectivity. With the ad-vent of deformable maps that can in-corporate variability, however, indi-vidual structures such as specificnuclei or other distinguishable ana-tomic features are more frequentlythe subject of intensive mapping ef-

The success of any brainatlas depends on howwell the anatomies of

individual subjectsmatch the

representation ofanatomy in the atlas.


forts (Paxinos and Huang, 1995). Anumber of pathogenic processes haveselective neuroanatomic targets re-sulting in characteristic morphologi-cal changes. Careful population map-pings of these changes can produce aneuroanatomic signature that identi-fies the disease. Examples include thecorpus callosum (Fig. 4) in dementiaor schizophrenia (Thompson et al.,1998a; Narr et al., 2000), the cerebel-lum in autism (Courchesne, 1997) andthe cingulate gyrus in Alzheimer dis-ease (Mega and Cummings, 1996;Mega et al., 1998) among others.

Maps of the Corpus CallosumNeuroimaging studies of the corpuscallosum, for example, are easier tounderstand if its elaborate internal or-ganization is considered. The corpuscallosum connects the cortical sur-faces of the two brain hemispheres,and there is a topographically specificorganization of callosal fibers in rela-tion to the cortical regions they con-nect. Tract-tracing studies using an-terograde or retrograde labels such asbiocytin or rhodamine-labeled latexmicrospheres (Innocenti, 1994) haveestablished the topographic distribu-

tion of callosal connections at the cor-tex in several species. A massive peri-natal loss of callosal axons, lastingfrom the 35th gestational week to theend of the first post-natal month(Clarke et al., 1989; La Mantia andRakic, 1990) is thought to lead to arestricted pattern of adult callosalconnections (Innocenti, 1994). In theadult callosum, the genu (or anteriorthird) connects pre-frontal cortices,the midbody (middle third) connectsmotor, somatosensory and auditorycortices, and the splenium (posteriorfifth) carries temporal, parietal, and

Figure 3. Transforming one anatomical dataset to match another. 3D image deformation, or warping algorithms (Toga, 1998; Thompsonet al., 2000) have powerful applications in the analysis of anatomical and functional maps. They apply a complex pattern of dilation andcontraction (top left) to a 3D dataset, which reshapes the anatomy to match the anatomy in another dataset. They can correct fordistortions due to post mortem anatomical change or histologic staining, reshaping cryosection or histologic images to match their in vivoconfiguration. Atlases can also be deformed to equate their anatomy with individual datasets. Anatomically-based warping algorithms(Thompson and Toga, 1996) use models of anatomical surfaces and curves (top right and bottom panels) to constrain the transformation,allowing key functional boundaries (here the calcarine and parieto-occipital sulci) to be matched in both datasets. By contrast with globalregistration approaches, the amount of deformation applied typically varies across the brain (see magnitude of deformation, top right).


occipital (visual) fibers. Perisylvian fi-bers from superior temporal and pa-rietal cortex relay information fromcritical language and association ar-eas, and cross mainly in the isthmus.To a certain degree, callosal fibertypes are also organized topographi-cally. Fast-conducting, large diameter(.3 mm sensorimotor fibers are con-

centrated in the posterior midbodyand splenium, whereas thinner, morelightly myelinated fibers are found atthe genu. These fibers at the genu of-fer a lower conduction velocity, con-necting pre-frontal regions implicatedin longer-term planning and organiza-tion of behavior (Aboitiz et al., 1992).Nonetheless, the idea of a sharply-de-

fined cortical map at the callosum hasbeen mitigated by recent anterogradetracer studies in humans (Di Virgilioand Clarke, 1997). These suggest thatheterotopic connections (i.e., betweennon-equivalent cortical areas in eachbrain hemisphere) are numerous andwidespread, even in the genu and sple-nium where callosal axons are mosthighly segregated.

To avoid making arbitrary defini-tions, Denenberg et al. (1991) per-formed a factor analysis to determinea “natural” partition of the callosum.Thickness measurements were ob-tained from a population of 104 nor-mal adults (by connecting 100 equallyspaced points on the inner and outercallosal boundaries; Fig. 4f), and thesemeasures were used to determine

seven regions with consistent varia-tions (7 factors). Ultimately, the parti-tioning scheme chosen depends onthe application objectives and thescale of the expected structural effects(Bookstein, 1997). Many of the con-flicts among different callosal map-ping studies may derive from the useof different anatomical mapping andpartitioning techniques (see Bishopand Wahlsten, 1997; reviewed inThompson et al., 2000a).

MULTIPLE MODALITIES ANDDIMENSIONS

As noted earlier, due to pronouncedanatomic variability between individ-ual human brains, any atlas or clinicaldiagnostic system based on a single

Figure 4. Common partitioning schemes for regional analysis of the corpus callosum. In viewof the topographically-specific relation between callosal regions and the cortical regionsthey connect, several partitioning approaches have been devised to allow separateanalysis of different callosal sectors. Vertical partitions subdivide the callosum based onfractions of its maximal anterior-posterior length. Vertical fifths (a) (e.g., Duara et al., 1991;Larsen et al., 1992) or equal angular sectors, relative to the callosal centroid (labeled CG inb), are commonly used. Variants of the radial partition have used rays emanating from themidpoint of the line joining the inferior rostrum and inferior splenium (Weis et al., 1993). Avariant of the widely-used Witelson partition (c; Witelson, 1989), further subdivides theanterior third (A) into rostrum, genu, and rostral midbody, using an additional vertical line(Rajapakse et al., 1996). Angular rays from the callosal centroid (CG) can be used toproduce a partition with 100 equiangular elements (e) (Rajapakse et al., 1996). The centerof these elements (e) can also be used to derive a curvilinear reference line (d) (Clarke etal., 1989). By using a set of nodes to partitioning this line into 5 (as in d) or 30 (Clarke et al.,1989) equal segments, a slightly different set of sectors can be defined based on theshortest line through each node connecting inner and outer callosal boundaries. Theseshortest lines form the basis for maximal splenial width (SW) and minimal body width (BW)measurements (Clarke and Zaidel, 1994), and derived bulbosity measures (Clarke et al.,1989). Alternatively, sets of 100 nodes equally spaced along inner and outer boundaries (f)can be joined directly (Denenberg et al., 1991), before a factor analysis to determine anatural partition. Stievenart et al. (1997) partition the callosum by defining rays (g) normal toa series of equidistant nodes on the ventral callosal boundary. Finally, approaches basedon deformable templates (h) measure the deformation required to match a digital tem-plate of the callosum with callosal outlines from individual subjects. Atlases also can bepassively carried along in the deformation to reveal patterns of shape difference or shapechange over time (Thompson et al., 2000). A simulated deformation is shown in h. Statisticsof these deformation maps can be used to localize and test the significance of patterns ofshape differences (Davatzikos et al., 1996; Bookstein, 1997; Thompson and Toga, 1998)(schematics shown here are adapted from diagrams created by Clarke et al., 1989;Witelson, 1989; Clarke and Zaidel, 1994; Rajapakse et al., 1996; Stievenart et al., 1997).

With the advent ofdeformable maps that

can incorporatevariability, individual

structures such asspecific nuclei or other

distinguishableanatomic features are

more frequently thesubject of intensive

mapping efforts.


subject’s anatomy cannot succeedfully. A deformable brain atlas coun-teracts some of the limitations of afixed atlas by using mathematicallyflexible transformations, but its suc-cess remains based on the premisethat brains resemble a prototypicaltemplate of anatomy, and can be pro-duced by continuously deforming it.

To realize the quantitative potentialof digital atlases, data from single sub-jects must be extendable to popula-tions (Mazziotta et al., 1995, 2000).Atlasing considerations suggest that astatistical confidence limit, ratherthan an absolute representation ofneuroanatomy, may be more appro-priate for representing particular sub-populations.

Population-Based AtlasingApproaches

Methods to create probabilistic brainrepresentations currently fall intothree major categories, each differingslightly in its conceptual foundations.The three methods are: the density-based, label-based, and deformation-based approaches. Benefits of each ap-proach are outlined below.

Density-based approaches. Initialapproaches to population-based atlas-ing concentrated on generating ‘aver-age’ representations of anatomy by in-tensity averaging of multiple MRIscans (Evans et al., 1992; Andreasen etal., 1994). A large number of MRIscans are each linearly transformedinto stereotaxic space, intensity-nor-malized and averaged on a voxel-by-voxel basis, producing an average in-tensity MRI dataset. The averagebrains that result have large areas, es-pecially at the cortex, where individ-ual structures are blurred out due tospatial variability in the population.Although this blurring limits theirusefulness as a quantitative tool, thetemplates can be used as targets forthe automated registration and map-ping of MR and co-registered func-tional data into stereotaxic space(Evans et al., 1994).

Label-based approaches. In label-based approaches (Evans et al., 1994;also known as SPAM approaches,short for ‘statistical/probabilistic

anatomy maps’), large ensembles ofbrain data are manually labeled, or‘segmented,’ into sub-volumes, aftermapping individual datasets into ste-reotaxic space. A probability map isthen constructed for each segmentedstructure, by determining the propor-tion of subjects assigned a given ana-tomic label at each voxel position instereotaxic space (Evans et al., 1994;Otaky et al., 1995; Paus et al., 1996).The prior information that theseprobability maps provide on the loca-tion of various tissue classes in stereo-taxic space has been useful in design-ing automated tissue classifiers andapproaches to correct radio-frequencyand intensity inhomogeneities in MRscans (Zijdenbos and Dawant, 1994).In our laboratory, we have also usedSPAM probabilistic maps to constrainthe search space for significant activa-tions in PET and SPECT imaging ex-periments (Dinov et al., 1998; Mega etal., 1998). Statistical data on anatomiclabels and tissue types normally foundat given positions in stereotaxic spaceprovide a vital independent source ofinformation to guide and informmathematical algorithms that analyzeneuroanatomic data in stereotaxicspace.

Deformation-based approaches. Asnoted earlier, when applied to two dif-ferent 3D brain scans, a non-linearregistration or warping algorithm cal-culates a deformation map thatmatches up brain structures in onescan with their counterparts in theother. The deformation map indicates3-dimensional patterns of anatomicdifferences between the two subjects(Bookstein, 1989). In probabilistic at-lases based on deformation maps(Thompson and Toga, 1997, 1998a),statistical properties of these defor-mation maps are encoded locally todetermine the magnitude and direc-tional biases of anatomic variation.Encoding of local variation can thenbe used to assess the severity of struc-tural variants outside of the normalrange, which may be a sign of disease(Amit et al., 1991; Callini et al., 1994b;Thompson et al., 1997). A major goalin designing this type of pathology de-tection system is to recognize thatboth the magnitude and local direc-tional biases of structural variability

in the brain may be different at everysingle anatomic point (Thompson etal., 1996). In contrast to the intensityaveraging of other current approaches(Evans et al., 1992; Andreasen et al.,1994), an anisotropic random vectorfield framework is introduced to en-code directional biases in anatomicvariability and map out abnormalitiesin new subjects (Thompson et al.,1997; Grenander and Miller, 1998).

The three major approaches forprobabilistic atlas construction differonly in the attribute whose statisticaldistribution is modeled and analyzed.Random vector fields (i.e., vector dis-tributions of deformation vectors ateach point in space) are analyzed inapproaches based on deformationmaps, while random scalar fields areused to model MR intensity statisticsin the density-based approach, and tomodel the incidence of binary labelsin space in the label-based approach.

Encoding Brain Variation

Realistically complex mathematicalstrategies are needed to encode com-prehensive information on structuralvariability in human populations. Par-ticularly relevant is 3-dimensional sta-tistical information on group-specificpatterns of variation, and how thesepatterns are altered in disease. This in-formation can be encoded so that it canbe exploited by expert diagnostic sys-tems, whose goal is to detect subtle ordiffuse structural alterations in disease.Strategies for detecting structuralanomalies can leverage information indatabased anatomic data by invokingencoded knowledge on the variations ingeometry and location of neuroana-tomic regions and critical functional in-terfaces, especially at the cortex.

Atlases of Cortical Patterns

The random vector field approach is ageneral strategy to construct popula-tion-based atlases of the brain(Thompson and Toga, 1997). Briefly,given a 3D MR image of a new subject,a high-resolution parametric surfacerepresentation of the cerebral cortexis automatically extracted. The algo-rithm then calculates a set of high-dimensional volumetric maps, elasti-cally deforming this surface intostructural correspondence with other


cortical surfaces, selected one by onefrom an anatomic image database.The family of volumetric warps soconstructed encodes statistical prop-erties of local anatomical variationacross the cortical surface. Special-ized strategies elastically deform thesulcal patterns of different subjectsinto structural correspondence, in away that matches large networks of

gyral and sulcal landmarks with theircounterparts in the target brain.

Cortical morphology is notoriouslycomplex, and presents unique chal-lenges in anatomic modeling investi-gations. In response to these chal-lenges, much research has beendevoted to developing cortical param-eterization and flattening algorithms.These methods optimally transform

maps of cortical features onto a sim-pler, non-convoluted surface such as a2D plane (Van Essen and Maunsell,1980; Schwartz and Merker, 1986;Carman et al., 1995; Drury et al.,1996), an ellipsoid (Dale and Sereno,1993; Sereno et al., 1996) or a sphere(Davatzikos, 1996; Thompson et al.,1996a, 1997, 1998a).

Cortical parameterization offerssubstantial advantages for visualizingcortical topography, and provides po-tentially clearer interpretations ofhow architectonic fields and func-tional loci are related in the cortex(Van Essen et al., 1997). Parameter-ization and flattening techniques havealso been applied to cortical modelsderived from the Visible Human data-sets (Spitzer et al., 1996), and the re-sulting templates have served as thefoundation for developing related at-lases of cortical regions (Drury andVan Essen, 1997). These cortical at-lases serve as a structural frameworkupon which architectonic, functional,and electrophysiological data, can becompared and integrated. Parametricmodels of the cortex also make com-parisons of cortical anatomy moretractable in disease states (Thompsonet al., 1997) and even across species(Van Essen et al., 1997). They can alsohelp overcome problems caused bywide cross-subject variations in corti-cal geometry, by supporting non-lin-ear registration of cortically-derivedfunctional data and histologic brainmaps localized at the cortex (Davatzi-kos, 1996; Thompson and Toga, 1996;Mega et al., 1997).

Warping the Cerebral Cortex

Despite the advantages provided bytransformations that simplify its ge-ometry, the cortical surface presentssignificant challenges for all brainmapping and registration algorithmsthat strive to match the anatomy ofone subject’s cortex with another. Theneed to make comparative measure-ments at the cortex across subjects re-quires a surface-to-surface warp thatnot only matches overall cortical ge-ometry, but also enforces point-to-point correspondence to a higher de-gree. Specialized approaches havebeen developed to match cortical re-gions, so that networks of sulci andgyri are individually matched (Thomp-

Figure 5. Cortical flattening and averaging. Average cortical surfaces can be generatedfor a subject population by matching gyral patterns across individuals. a: A flattenedrepresentation of the cortex (cf., Van Essen et al., 1997; Thompson et al., 2000) for the lefthemisphere of one subject, with the average cortical pattern for the group overlaid(colored lines). b: Shows the result of transforming the individual’s sulcal pattern into theaverage configuration for the group, using an image warping algorithm (Thompson et al.,2000). The 3D cortical regions that map to these average locations are then recovered ineach individual subject, as follows. A color code (c) (Thompson and Toga, 1997) represent-ing 3D cortical point locations in this subject is convected along with the flow that drives thesulcal pattern into the average configuration for the group (d). Once this is done in allsubjects, points on each individual’s cortex are recovered that have the same relativelocation to the primary folding pattern in all subjects. Averaging of these correspondingpoints results in a crisp average cortex (f); this average is based on N 5 9 Alzheimerpatients). Note that a smooth cortex is produced (e) if the color images are averagedwithout warping, as gyral landmarks do not constructively interfere in the averaging pro-cess. The transformation fields that map individuals onto the group average can also bestored and used to measure regional anatomic variability (see Thompson et al., 2000 for adetailed review).


son and Toga, 1996, 1998; Fig. 5). Dif-ferences in the serial organization ofcortical gyri prevent exact gyrus-by-gyrus matching of one cortex with an-other. Some cortical areas are partic-ularly subject to variations in theincidence and topology of accessorygyri, and one subject may have two orthree gyri where one gyrus is found inanother subject. This feature is espe-cially notable in studies of paracingu-late and temporo-parietal regions, inparticular the planum temporale andposterior perisylvian areas that form acritical part of the language represen-tation of the left hemisphere (Ono etal., 1990; Leonard, 1996; Paus et al.,1996). Because the assumption thatbrains are topologically equivalentbreaks down once the functional unitsof the cortex are finely subdivided, animportant intermediate goal has beento identify and match a comprehen-sive network of sulcal and gyral ele-ments that are consistent in their in-cidence and topology across subjects(Ono et al., 1990; Rademacher et al.,1993; Thompson et al., 1996, 1997).

In surface-based approaches, 3Ddeformable models (Cohen and Co-hen, 1992; MacDonald et al., 1993; Da-vatzikos, 1996; Thompson and Toga,1996) are used to automatically ex-tract parametric representations ofeach subject’s cortex, on which corre-sponding networks of anatomical

curves are identified. The transforma-tion relating these networks is ex-pressed as a vector flow field in theparameter space of the cortex that in-directly specifies a correspondencefield in 3D driving one cortical surfaceinto the shape of another (Drury et al.,1996; Davatzikos, 1996; Thompsonand Toga, 1996; see Fig. 5).

An advantage of this approach isthat an average model of the cortexcan be generated for a population thatretains the major gyral and sulcallandmarks (Fig. 5f). First, a set of col-or-coded 2D images is generated sothat each point in the 2D image in-dexes a point on the cortical surface(Fig. 5c). These images essentiallystore information on the morphologyof the cortex. These images, however,can then be warped so that corre-sponding gyri are aligned across sub-jects. If the warped images (Fig. 5d)are averaged pixel-by-pixel, the result-ing color-coded image can be con-verted back into a cortical model withthe average geometry of the group,which still retains gyral features (Fig.5f; Thompson et al., 2000b).

This cortical averaging approachcan therefore retain features of thecortex that would in general bewashed away by averaging brain im-ages directly. By retaining an averagegeometry for the cortical pattern,other features of cortical organization

emerge after group averaging, thatmay not be apparent in an individual,due to the large cross-subject varia-tions in cortical patterning. The aver-age profile of cortical asymmetry in apopulation can be computed, for ex-ample, by averaging geometric mod-els of the sulcal pattern, and compar-ing the results in both brainhemispheres (Fig. 6; Thompson et al.,2000). The ability to compare thesecortical models across groups andacross time can then be used to testhypotheses about the loss of corticalasymmetry in disease (Narr et al.,2000), its progression during develop-ment (Sowell et al., 2000), and modu-lation by disease processes (Thomp-son et al., 1998a).

Disease States

Cortical structure is severely affectedin disease states such as Alzheimerdisease, Pick disease and other de-mentias, by tumor growth, and incases of epilepsy, cortical dysplasias,and schizophrenia. Cortical matchingapproaches can be exploited by algo-rithms that detect these alterations. Inone approach (Thompson et al.,1997), a probability space of randomtransformations, based on the theoryof anisotropic Gaussian randomfields, encodes information on com-plex variations in gyral and sulcal to-pography from one individual to an-other. Confidence limits in stereotaxicspace are determined, for cortical sur-face points in a new subject’s brain,and color-coded probability maps arecreated to highlight and quantify re-gional patterns of deformity in theanatomy of new subjects.

POPULATION SPECIFICITY

Coordinate Systems

Atlasing of developmental brain datapresents unique challenges. Imposi-tion of standardized coordinate sys-tems is difficult, and their relationshipto anatomic nomenclature is hard todefine, when potentially drastic mor-phological differences exist amongdata sets. In Yoon et al. (1997), a pho-tographic atlas of the human embryowas created, based on detailed obser-vations in utero from the 4th to the7th week after ovulation (CarnegieStages 10–18). In Chong et al. (1997),

Figure 6. Mapping brain asymmetry. By averaging geometric models of sulcal landmarks ineach brain hemisphere, the average pattern of structural asymmetry can be computed fora population. Here the magnitude of structural asymmetry clearly apparent (N 5 20 normalelderly subjects) in perisylvian cortices. The average right (R) hemisphere cortical pattern istorqued forward relative to the same cortical elements in the left brain hemisphere (L),consistent with in vivo and post mortem volumetric studies. By better defining the param-eters of allowable normal variations, the resulting information can be leveraged to distin-guish normal from abnormal anatomical variants (Thompson et al., 2000a).


26 normal formalin-fixed fetal speci-mens with a gestational age of 9 to 24weeks were examined with high-reso-lution MRI using a conventional clin-ical magnet and pulse sequences, andMR findings were correlated with his-tologic atlas data. Although templatesof normal development helped toidentify expected developmental fea-tures, it was noted that direct correla-tion of fetal MR images with anatomicatlases might result in a mistaken di-agnosis of delayed development, be-cause of a time lag in the detection ofstructures on MR images.

Current atlases of fetal development(O’Rahilly and Muller, 1987; England,1990) use collections of labeled datafrom multiple imaging modalities tocharacterize specific developmentalstages. The first comprehensive MRIatlas of pediatric cranial anatomy(Salamon et al., 1990) incorporates180 MRI scans acquired parallel to theorbito-meatal anatomical plane, and360 explanatory diagrams depictingfunctional neuroanatomy from birththrough 16 years of age. In this collec-tion, 3D horizontal and sagittal im-ages facilitate identification of sulciand gyri. Stereotaxic coordinate sys-tems, however, were not applied tothe atlas data due to difficulties in us-ing them to reference embryonic andpediatric data. In the spirit of the de-formable atlas methods described ear-lier, extreme deformations could beimposed to fit all stages of develop-ment into a standardized atlas, butthis would hardly meet the primaryrequirement of atlasing, which is toprovide a natural coordinate frame-work in which to localize and classifystructures present in developing brains.

Alternatively, different atlases andcoordinate systems for several dis-crete stages of development might beused. Numerous anatomic features,due to their emergence and disappear-ance during development, could beused to place individual brains into anappropriate atlas in the set. Warpingapproaches could then be applied tothe atlas coordinate systems as a basisto compare and quantitate develop-ment (Toga et al., 1996b; Thompson etal., 1998b). Temporal interpolationbetween atlases in the set could beused to generate additional anatomictemplates, representing brains at any

stage of maturity in between thosestages represented in the initial inven-tory.

DYNAMIC MAPS

In many ways, static representationsof brain structure are ill-suited to an-alyzing dynamic processes of braindevelopment and disease. The intenseinterest in brain development and dis-ease mandates the design of mathe-matical systems to track anatomicalchanges over time and map dynamicpatterns of growth or degeneration.

Temporal Maps of BrainStructure

Current structural brain imaging in-vestigations typically focus on theanalysis of 3-dimensional models ofbrain structure, derived from volu-

metric images acquired at a singletime-point from each subject in thestudy. Serial scanning of human sub-jects, when combined with a powerfulset of warping and analysis algo-rithms, however, can enable diseaseand growth processes to be tracked intheir full spatial and temporal com-plexity.

One of the most promising applica-tions of warping algorithms is theiruse as a virtual sensor (Gee et al.,1993), creating exquisitely detailedmaps of anatomic differences. Mapsof anatomical change can also be gen-erated by warping scans acquiredfrom the same subject over time(Thompson et al., 1998b). Serial scan-ning of human subjects (Fox et al.,1996; Freeborough et al., 1996;Thompson et al., 1998b) or experi-mental animals (Jacobs and Fraser,

1994) in a dynamic state of disease ordevelopment offers the potential tocreate 4-dimensional models of brainstructure. These models incorporatedynamic descriptors of how the brainchanges during maturation or dis-ease. For a range of patient popula-tions, 4D models of the brain can bebased on imaging and modeling its3-dimensional structure at a sequenceof time-points. In a changing mor-phology, warping algorithms enableone to model structural changes thatoccur over prolonged periods, such asdevelopmental, aging or disease pro-cesses, as well as structural changesthat occur more rapidly, as in recov-ery after trauma or tumor growth. A4-dimensional approach can providecritical information on local patternsand rates of tissue growth, atrophy,shearing and dilation that occur in thedynamically changing architecture ofthe brain (Toga et al., 1996b; Thomp-son et al., 1998b).

Mapping Growth Patterns inFour Dimensions

In our initial human studies (Thomp-son et al., 1998b; Thompson andToga, 1998b), we developed several al-gorithms to create 4-dimensionalquantitative maps of growth patternsin the developing human brain. Time-series of high-resolution pediatricMRI scans were analyzed. The result-ing tensor maps of growth providedspatially-detailed information on localgrowth patterns, quantifying rates oftissue maturation, atrophy, shearingand dilation in the dynamicallychanging brain architecture.

3D (2562 3 124 resolution) T1-weighted fast SPGR (spoiled GRASS)MRI volumes were acquired fromyoung normal subjects (mean age:8.663.1 yrs.) at intervals ranging from2 weeks to 4 years. Pairs of scans wereselected to determine patterns ofstructural change across the intervalbetween the two scans. These scanpairs were pre-processed with a radio-frequency bias field correction algo-rithm, and rigidly registered using au-tomated image registration software(Woods et al., 1993). Registered scanswere then histogram-matched and apreliminary map of differences in MRsignal intensities between the twoscans was constructed. Although dif-

The intense interest inbrain development anddisease mandates the

design of mathematicalsystems to track

anatomical changesover time and mapdynamic patterns of

growth or degeneration.


ference maps help to determinewhether structural change has oc-curred in dementia (Freeborough etal., 1996), these maps do not localizechange, nor do they provide 3-dimen-sional measures of dilation, contrac-tion or shearing of anatomic regions.To address this, parametric meshmodels (Thompson et al., 1996a,1997, 1998a) were created to repre-sent a comprehensive set of deep sul-cal, callosal, caudate and ventricularsurfaces at each time-point. Parame-terized cortical surface models werealso automatically extracted fromeach of the mutually registered histo-gram-matched scans. Surface modelsbased on manually-digitized datawere averaged across multiple trials(N 5 6) to minimize error. The defor-mation field required to match the

surface anatomy of one scan with theother was extended to the full volumeusing a continuum-mechanical modelbased on the Cauchy-Navier operatorof linear elasticity (Thompson andToga, 1998b). Deformation processesrecovered by the warping algorithmwere then analyzed using vector fieldoperators to produce a variety of ten-sor maps. These maps were designedto reflect the magnitude and principaldirections of dilation or contraction,the rate of strain, and the local curl,divergence and gradient of flow fieldsrepresenting the growth processes re-covered by the transformation.

Tensor Maps of Growth

In contrast to the near-zero maps ofchange recovered at short time inter-

vals (2 weeks), tensor maps of growthspanning large time intervals (4 years)showed complex and heterogeneouspatterns of change. In one subjectscanned at ages 7, 9 and 11, compar-ative stability of lobar and thalamicanatomy, and negligible changes atthe cortex, were accompanied by pro-nounced focal growth of the callosalisthmus (up to 80%; Fig. 7a,b), ven-tricular enlargement and loss of cau-date tissue (Fig. 7c). Similar tensormapping approaches can reveal thecomplex dynamics of tissue loss in in-dividuals or groups of patients withAlzheimer disease (Fig. 7d,e; Thomp-son et al., 2000f).

In the near future, these brain map-ping techniques will provide the abil-ity to map growth and degeneration intheir full spatial and temporal com-

Figure 7. Mapping dynamic growth and regressive processes in the brain. Tensor maps (Thompson et al., 2000) can be used to reveal thepatterns of growth in the developing brain. Local growth rates can also be mapped. An image warping algorithm, based on the principlesof continuum mechanics, is used to reconfigure a subject’s anatomy from its shape at an initial time-point into its shape at a latertime-point. Eighty percent growth at the callosal isthmus is observed across a 4-year time-span, (a) in a girl imaged with MRI at ages 8 and12, and (b) in another girl imaged at ages 9 and 13. Tissue loss (blue colors, c) is also observed in a young boy between the ages of 7 and11, revealing the complex profiles of growth and regressive processes occurring at this comparatively late period of brain development.Tensor maps also reveal the dynamic complexity of tissue loss in a female patient with Alzheimer disease imaged at the beginning and endof a 19-month interval with high-resolution MRI (d). The patient, aged 74.5 years at first scan, exhibits faster tissue loss rates (e) in thehippocampal head (10% per year) than in the fornix. These maps can elucidate the dynamics of degeneration and therapeutic responsein an individual or a population (Thompson et al., 2000).


plexity. Despite logistic and technicalchallenges, these mapping ap-proaches hold tremendous promisefor representing, analyzing and un-derstanding the extremely complexdynamic processes that affect regionalanatomy in the healthy and diseasedbrain. (See Marshall and Marshall,1994, for a detailed review.)

SPECIES

An immense variety of brain mapsand atlases have been created to rep-resent non-human primates, and a va-riety of other species. Despite the ad-vent of modern neuroimaging, ourknowledge of the molecular, cellular,systems and functional organizationof the human brain remains rudimen-tary compared with the vast numberof neuroscientific studies conductedin other species. The systematic elec-trophysiologic mapping of the neocor-tex in a wide range of mammals hasrevealed how cortical specializationvaries across species (Manzoni, 1997).Many specialized stereotaxic systems,atlases, and parcellation schemeshave also been developed for brainmapping in non-human species (e.g.,Shantha et al., 1968; Szabo andCowan, 1984; Paxinos and Watson,1986; Felleman and Van Essen, 1991;Martin and Bowden, 1996). In parallelwith the development of interna-tional, electronic registries for humanbrain mapping data (e.g., BrainMap;Fox et al., 1994), similar systems areunder active development for the ar-chival and meta-analysis of brainmapping data obtained in other spe-cies.

Multi-Modality Maps

Neuroimaging investigations in non-human primates have also includedmulti-modality studies. In our labora-tory (Cannestra et al., 1998), a 3Dmulti-modality computerized map ofthe Nemestrina monkey brain was cre-ated by combining pre-mortem CT,PET, and MRI with a reconstructedpost mortem volume of high-resolu-tion full-color cryosection images,captured at 50 micron increments.The resulting data sets were reposi-tioned into the Horsley and Clark(1908) stereotaxic coordinate system,and labeled 3D surface models were

reconstructed to represent nuclei,tracts, and other neuroanatomicalfeatures.

In the future, statistical character-ization of the relationships amongcortical regions in human and non-human primate populations will beinvaluable in investigating how archi-tectural patterns vary across individu-als and species. Because neuronalconnectivity patterns have beenmapped extensively in several species,brain mapping tools that supportcross-species comparisons of brainmaps are likely to help in elucidatingthe complex functional geography ofthe human cortex.

CONCLUSION

As we have seen, the use of brainmaps is as varied as their construc-tion. Their utility results from theircapacity to measure, visualize, com-pare and summarize brain images.Maps can take on many forms, fromdescriptions of structure to functionof the whole brain to maps of groupsor populations. Individual systems ofthe brain can be mapped as canchanges over time, as in developmentor degeneration. Maps enable com-parison across individuals, modalitiesor states. Differences between speciescan be catalogued. But in most cases,the value added by brain maps is theunique and critical ability to integrateinformation from multiple sources.Although dependent upon appropri-ate coordinate systems, deformationmethods and visualization strategies,accurate and representative brainmaps hold the most promise for help-ing to create a comprehensive under-standing of brain in health and dis-ease.

ACKNOWLEDGMENTSThis work was generously supported byresearch grants from the National Cen-ter for Research Resources (RR13642),National Library of Medicine (LM/MH05639), NINDS (NS38753), theNational Science Foundation (BIR 93-22434), and by a Human Brain Projectgrant known as the International Con-sortium for Brain Mapping, which isfunded jointly by NIMH and NIDA(P20 MH/DA52176).

LITERATURE CITED

Aboitiz F, Scheibel AB, Fisher RS, ZaidelE. 1992. Fiber composition of the humancorpus callosum. Brain Res 598:143–153.

Allen LS, Richey MF, Chai YM, Gorski RA.1991. Sex differences in the corpus cal-losum of the living human being. J Neu-rosci 11: 933–942.

Amit Y, Grenander U, Piccioni M. 1991.Structural image restoration through de-formable templates. J Am Stat Assoc 86:376–386.

Andreasen NC, Arndt S, Swayze V, CizadloT, Flaum M, O’Leary D, Ehrhardt JC,Yuh WTC. 1994. Thalamic abnormalitiesin schizophrenia visualized throughmagnetic resonance image averaging.Science 266:294–298.

Avoli M, Hwa GC, Kostopoulos G, Oliver A,Villemure JG. 1991. Electrophysiologicalanalysis of human neocortex in vitro: ex-perimental techniques and methodolog-ical approaches. Can J Neurol Sci 18:636–639.

Bajcsy R, Kovacic S. 1989. Multi-resolu-tion elastic matching, computer vision,graphics and image processing. CompGraphics and Applications 46:1–21.

Berger H. 1929. Uber das elektrenkephalo-gramm des menschen. Arch PsychiatrNervenkr 87:527–580.

Bishop KM, Wahlsten D. 1997. Sex differ-ences in the human corpus callosum:myth of reality? Neurosci Biobehav Rev21:581–601.

Bohm C, Greitz T, Kingsley D, BerggrenBM, Olsson L. 1983. Adjustable comput-erized brain atlas for transmission andemission tomography. Am J Neuroradiol4:731–733.

Bookstein F. 1989. Principal warps: thin-plate splines and the decomposition ofdeformations. IEEE Trans Pattern Anal-ysis Machine Intelligence 11:567–585.

Bookstein FL. 1997. Landmark methodsfor forms without landmarks: morpho-metrics of group differences in outlineshape. Med Image Anal 1:225–243.

Brodmann K. 1909. Vergleichende lokal-isationslehre der grosshirnrinde in ihrenprinzipien dargestellt auf grund des zel-lenbaues, Barth, Leipzig. In: Some pa-pers on the cerebral cortex, translatedas: On the comparative localization ofthe cortex, p 201–230.

Broit C. 1981. Optimal registration of de-formed images. PhD Dissertation, Uni-versity of Pennsylvania, USA.

Bro-Nielsen M, Gramkow C. 1996. Fastfluid registration of medical images. In:Hohne KH, Kikinis R, editors. Visualiza-tion in biomedical computing, Ham-burg, Germany. Lecture notes in com-puter science. Berlin: Springer-Verlag. p267–276.

Cannestra AF, Santori EM, Holmes CJ,Toga AW. 1998. A 3D multi-modalitybrain map of the Nemestrina monkey.Brain Res Bull 43:141–148.

Cao J, Worsley KJ. 2001. The geometry ofthe Hoteling’s T-squared random field


with applications to the detection ofshape changes. Ann Stat (in press).

Carman GJ, Drury HA, Van Essen DC.1995. Computational methods for recon-structing and unfolding the cerebral cor-tex. Cereb Cortex 5:506–517.

Chong BW, Babcook CJ, Pang D, Ellis WG.1997. A magnetic resonance template fornormal cerebellar development in thehuman fetus. Neurosurgery 41:924–928.

Christensen GE, Rabbitt RD, Miller MI.1993. A deformable neuroanatomy text-book based on viscous fluid mechanics.27th Annual Conference on InformationSciences and Systems, p 211–216.

Christensen GE, Miller MI, Marsh JL, Van-nier MW. 1995a. Automatic analysis ofmedical images using a deformable text-book. Proc Comp Assist Radiol. Berlin:Springer Verlag. p 152–157.

Christensen GE, Rabbitt RD, Miller MI,Joshi SC, Grenander U, Coogan TA, VanEssen DC. 1995b. Topological propertiesof smooth anatomic maps. In: Bizais Y,Barillot C, Di Paola R, editors. Informa-tion processing in medical imaging. Klu-wer Academic Press, p 101–112.

Christensen GE, Rabbitt RD, Miller MI.1996. Deformable templates using largedeformation kinematics. IEEE TransImage Processing 5:1435–1447.

Clarke JM, Zaidel E. 1994. Anatomical-be-havioral relationships: corpus callosummorphometry and hemispheric special-ization. Behav Brain Res 64:185–202.

Clarke S, Kraftsik R, Van der Loos H, In-nocenti GM. 1989. Forms and measuresof adult and developing human corpuscallosum. J Neuropathol Exp Neurol280:213–230.

Cohen LD, Cohen I. 1992. Deformablemodels for 3D medical images using fi-nite elements and balloons. Proceedingsof the IEEE Computer Society Confer-ence on Computer Vision and PatternRecognition (Cat. No. 92CH3168-2), LosAlamitos, CA. IEEE Computer SocietyPress. p 592–598.

Collins DL, Neelin P, Peters TM, Evans AC.1994a. Automatic 3D intersubject regis-tration of MR volumetric data into stan-dardized Talairach space. J Comp AssistTomogr 18:192–205.

Collins DL, Peters TM, Evans AC. 1994b.An automated 3D non-linear image de-formation procedure for determinationof gross morphometric variability in thehuman brain. Proc VisualizationBiomed Comp 3:180–190.

Collins DL, Holmes CJ, Peters TM, EvansAC. 1995. Automatic 3D model-basedneuroanatomical segmentation. HumBrain Mapp 3:190–208.

Collins DL, Le Goualher G, Venugopal R,Caramanos A, Evans AC, Barillot C.1996. Cortical constraints for non-linearcortical registration. In: Hohne KH, Kiki-nis R, editors. Visualization in biomedicalcomputing, Hamburg, Germany, Sept.1996. Lecture notes in computer science.Berlin: Springer-Verlag. p 307–316.

Courchesne E. 1997. Brainstem, cerebellarand limbic neuroanatomical abnormali-ties in autism. Curr Opin Neurobiol7:269–278.

Dale AM, Sereno MI. 1993. Improved lo-calization of critical activity by combin-ing EEG and MEG with MRI corticalsurface reconstruction—A linear ap-proach. J Cogn Neurosci 5:162–176.

Damasio H. 1995. Human brain anatomyin computerized images. Oxford, NewYork: Oxford University Press.

Davatzikos C. 1996. Spatial normalization of3D brain images using deformable mod-els. J Comp Assist Tomogr 20:656–665.

Davatzikos C, Vaillant M, Resnick SM,Prince JL, Letovsky S, Bryan RN. 1996. Acomputerized approach for morphologi-cal analysis of the corpus callosum.J Comp Assist Tomogr 20:88–97.

Denenberg VH, Kertesz A, Cowell PE. 1991.A factor analysis of the human’s corpuscallosum. Brain Res 548:126–132.

Dinov ID, Thompson PM, Woods RP, MegaMS, Holmes CJ, Sumners D, Saxena S,Toga AW. 2000. Probabilistic sub-vol-ume partitioning techniques for deter-mining the statistically significant re-gions of activation in stereotaxicfunctional data. J Computer Assisted To-mography 24:128–138.

DiVirgilio G, Clarke S. 1997. Direct inter-hemispheric visual input to human speechareas. Hum Brain Mapp 5:347–354.

Drury HA, Van Essen DC, Joshi SC, MillerMI. 1996. Analysis and comparison ofareal partitioning schemes using two-di-mensional fluid deformations. PosterPresentation, 2nd International Confer-ence on Functional Mapping of the Hu-man Brain, Boston, MA, June 17–211996. NeuroImage 3:S130.

Drury HA, Van Essen DC. 1997. Analysis offunctional specialization in human cere-bral cortex using the visible man surfacebased atlas. Hum Brain Mapp 5:233–237.

Duara R, Kushch A, Gross-Glenn K, BarkerWW, Jallad B, Pascal S, Loewenstein DA,Sheldon J, Rabin M, Levin B. 1991. Neu-roanatomic differences between dyslexicand normal readers on magnetic reso-nance imaging scans. Arch Neurol 48:410–416.

Duvernoy HM. 1991. The human brain.New York: Springer-Verlag.

England MA. 1990. Colour atlas of life be-fore birth: normal fetal development.Year Book Medical.

Evans AC, Collins DL, Milner B. 1992. AnMRI-based stereotactic brain atlas from300 young normal subjects. Proceedingsof the 22nd Symposium of the Societyfor Neuroscience, Anaheim, CA. p 408.

Evans AC, Collins DL, Neelin P, Mac-Donald D, Kamber M, Marrett TS. 1994.Three-dimensional correlative imaging:applications in human brain mapping.In: Thatcher RW, Hallett M, Zeffiro T,John ER, Huerta M, editors. Functionalneuroimaging: technical foundations. p145–162. San Diego: Academic Press.

Evans AC, Dai W, Collins DL, Neelin P,Marrett S. 1991. Warping of a comput-erized 3D atlas to match brain image

volumes for quantitative neuroanatomi-cal and functional analysis. SPIE MedImaging 1445:236–247.

Felleman DJ, Van Essen DC. 1991. Distrib-uted hierarchical processing in the pri-mate cerebral cortex. Cereb Cortex 1:1–47.

Fox NC, Freeborough PA, Rossor MN.1996. Visualization and quantification ofrates of cerebral atrophy in Alzheimerdisease. Lancet 348:94–97.

Fox PT, Mikiten S, Davis G, Lancaster JL.1994. BrainMap: a database of humanfunctional brain mapping. In: ThatcherRW, Hallett M, Zeffiro T, John ER,Huerta M, editors. Functional neuroim-aging: technical foundations. San Diego:Academic Press. p 95–106.

Fox PT, Mintun MA, Reiman EM, RaichleME. 1988. Enhanced detection of focalbrain responses using inter-subject aver-aging and change distribution analysisof subtracted PET images. J Cereb BloodFlow Metab 8:642–653.

Fox PT, Perlmutter JS, Raichle M. 1985. Astereotactic method of localization forpositron emission tomography. J CompAssist Tomogr 9:141–153.

Freeborough PA, Woods RP, Fox NC. 1996.Accurate registration of serial 3D MRbrain images and its application to visu-alizing change in neurodegenerative dis-orders. J Comput Assist Tomogr20:1012–1022.

Friston KJ, Frith CD, Liddle PF, Frackow-iak RSJ. 1991. Plastic transformation ofPET images. J Comp Assist Tomogr9:141–153.

Friston KJ, Holmes AP, Worsley KJ, PolineJP, Frith CD, Frackowiak RSJ. 1995. Sta-tistical parametric maps in functionalimaging: a general linear approach.Hum Brain Mapp 2:189–210.

Friston KJ, Passingham RE, Nutt JG,Heather JD, Sawle GV, Frackowiak RSJ.1989. Localization in PET images: directfitting of the intercommissural (AC-PC)line. J Cereb Blood Flow Metab 9:690–695.

Gee JC, Reivich M, Bajcsy R. 1993. Elasti-cally deforming an atlas to match ana-tomical brain images. J Comp Assist To-mogr 17:225–236.

Geschwind N, Levitsky W. 1968. Humanbrain: left-right asymmetries in temporalspeech region. Science 161:186.

Geyer S, Schleicher A, Zilles K. 1997. Thesomatosensory cortex of man: cytoarchi-tecture and regional distributions of re-ceptor binding sites. NeuroImage 6:27–45.

Greitz T, Bohm C, Holte S, Eriksson L.1991. A computerized brain atlas: con-struction, anatomical content and appli-cation. J Comp Assist Tomogr 15:26–38.

Grenander U, Miller MI. 1998. Computa-tional anatomy: an emerging discipline.Technical report. Dept. of Mathematics,Brown University.

Hardy TL. 1994. Computerized atlas forfunctional stereotaxis, robotics and ra-diosurgery. SPIE 2359:447–456.

Hohne KH, Bomans M, Pommert A, Ri-emer M, Schiers C, Tiede U, Wiebecke G.1990. 3D visualization of tomographic


volume data using the generalized voxelmodel. Visual Comput 6:28–36.

Hohne KH, Bomans M, Riemer M, SchubertR, Tiede U, Lierse W. 1992. A 3D anatom-ical atlas based on a volume model. IEEEComput Graphics Appl 12:72–78.

Holmes CJ, Hoge R, Collins L, Woods R,Toga AW, Evans AC. 1998. Enhance-ment of MR images using registrationfor signal averaging. J Comput Assist To-mogr 22:324–333.

Horsley V, Clarke RW. 1908. The structureand functions of the cerebellum exam-ined by a new method. Brain 31:45–124.

Innocenti GM. 1994. Some new trends inthe study of the corpus callosum. BehavBrain Res 64:1–8.

Jacobs RE, Fraser SE. 1994. Magnetic res-onance microscopy of embryonic celllineages and movements. Science 263:681–684.

Joshi SC, Miller MI, Christensen GE, Ban-erjee A, Coogan TA, Grenander U. 1995.Hierarchical brain mapping via a gener-alized Dirichlet solution for mappingbrain manifolds. Vision Geometry IV,Proceedings of the SPIE Conference onOptical Science, Engineering and Instru-mentation, San Diego, CA, Aug. 1995.2573:278–289.

Kikinis R, Shenton ME, Iosifescu DV, Mc-Carley RW, Saiviroonporn P, HokamaHH, Robatino A, Metcalf D, Wible CG,Portas CM, Donnino R, Jolesz F. 1996. Adigital brain atlas for surgical planning,model-driven segmentation, and teach-ing. IEEE Trans on Visualization andComp Graphics, Sept 1996. 2:232–241.

LaMantia AS, Rakic P. 1990. Axon overpro-duction and elimination in the corpuscallosum of the developing Rhesus mon-key. J Neurosci 10:2156–2175.

Larsen JP, Høien T, Odegaard H. 1992.Magnetic resonance imaging of the cor-pus callosum in developmental dyslexia.Cognit Neuropsychol 9:123–134.

Le Bihan D. 1996. Functional MRI of thebrain: principles, applications and limi-tations. Neuroradiology 23:1–5.

Lehmann ED, Hawkes D, Hill D, Bird C,Robinson G, Colchester A, Maisley M.1991. Computer-aided interpretation ofSPECT images of the brain using anMRI-derived neuroanatomic atlas. MedInformatics 16:151–166.

Leonard CM. 1996. Structural variation inthe developing and mature cerebral cor-tex: noise or signal? In: Thatcher RW,Reid Lyon G, Rumsey J, Krasnegor N,editors. Developmental neuroimaging:mapping the development of brain andbehavior. San Diego: Academic Press. p207–231.

MacDonald D, Avis D, Evans AC. 1993. Au-tomatic parameterization of human cor-tical surfaces. Annual Symp Info Proc inMed Imag (IPMI).

Mai J, Assheuer J, Paxinos G. 1997. Atlas ofthe human brain. San Diego: AcademicPress.

Mansour A, Fox CA, Burke S, Akil H,Watson SJ. 1995. Immunohistochemical

localization of the cloned mu opioid re-ceptor in the rat CNS. J Chem Neuroanat8:283–305.

Manzoni T. 1997. The callosal connectionsof the hierarchically organized somato-sensory areas of primates. J NeurosurgSci 41:1–22.

Marshall LL, Marshall MM. 1994. Stereo-taxic atlases and related references(1991; revised, 1994), David Kopf Instru-ments, Inc., available at: http://www-ca-jal.ucsd.edu/Pages/AtlasDocs/Kopf_At-las_Directory.html

Martin RF, Bowden DM. 1996. A stereotaxictemplate atlas of the macaque brain fordigital imaging and quantitative neuro-anatomy. NeuroImage 4:119–150.

Matsui T, Hirano A. 1978. An atlas of thehuman brain for computerized tomogra-phy, Igako-Shoin.

Mazziotta JC, Toga AW, Evans AC, Fox P,Lancaster J. 1995. A probabilistic atlas ofthe human brain: theory and rationale forits development. NeuroImage 2:89–101.

Mazziotta JC, Toga AW, Evans AC, Fox PT,Lancaster J, Zilles K, Woods RP, Paus T,Simpson G, Pike B, Holmes CJ, CollinsDL, Thompson PM, MacDonald D,Schormann T, Amunts K, Palomero-Gal-lagher N, Parsons L, Narr KL, Kabani N,Le Goualher G, Boomsma D, Cannon T,Kawashima R, Mazoyer B. 2001. A prob-abilistic atlas and reference system forthe human brain [Invited Paper]. J RoyalSoc (in press).

Mega MS, Chen S, Thompson PM, WoodsRP, Karaca TJ, Tiwari A, Vinters H, SmallGW, Toga AW. 1997. Mapping pathologyto metabolism: co-registration of stainedwhole brain sections to PET in Alzheimerdisease. NeuroImage 5:147–153.

Mega MS, Cummings JL. 1996. The cingu-late and cingulate syndromes. In: Cum-mings JL, Trimble MR, editors. Contem-porary behavioral neurology. Boston:Butterworth Heinemann.

Mega MS, Dinov ID, Lee L, Woods RP,Thompson PM, Holmes CJ, Back CL,Collins DL, Evans AC, Toga AW. 1998.Dissecting neural networks underlyingthe retrieval deficit from the amnesticmemory disorder using [99mTc]-HM-PAO-SPECT. Proc Am Behav NeurolSoc, Feb 1998.

Mega MS, Thompson PM, Cummings JL,Back CL, Xu LQ, Zohoori S, Goldkorn A,Moussai J, Fairbanks L, Small GW, TogaAW. 1998. Sulcal variability in the Alz-heimer brain: correlations with cogni-tion. Neurology 50:145–151.

Meltzer CC, Frost JJ. 1994. Partial volumecorrection in emission-computed to-mography: focus on Alzheimer disease.In: Thatcher RW, Hallett M, Zeffiro T,John ER, Huerta M, editors. Functionalneuroimaging: technical foundations.San Diego: Academic Press. p 163–170.

Miller MI, Christensen GE, Amit Y,Grenander U. 1993. Mathematical text-book of deformable neuroanatomies. ProcNatl Acad Sci USA 90:11944–11948.

Minoshima S, Koeppe RA, Frey KA, Ishi-hara M, Kuhl DE. 1994. StereotacticPET atlas of the human brain: aid for

visual interpretation of functional brainimages. J Nucl Med 35:949–954.

Missir O, Dutheil-Desclercs C, Meder JF,Musolino A, Fredy D. 1989. Central sul-cus patterns at MRI. J Neuroradiol 16:133–144.

Mori S, Barker PB. 1999. Diffusion mag-netic resonance imaging: its principleand applications. Anat Rec (New Anat)257:102–109.

Narr KL, Thompson PM, Sharma T, Mous-sai J, Cannestra AF, Toga AW. 2000.Mapping corpus callosum morphology inschizophrenia. Cereb Cortex 10:40–49.

Ono M, Kubik S, Abernathey CD. 1990.Atlas of the cerebral sulci. Stuttgart: Thi-eme.

Oppenheim JS, Skerry JE, Tramo MJ, Gaz-zaniga MS. 1989. Magnetic resonanceimaging morphology of the corpus callo-sum in monozygotic twins. Ann Neurol26:100–104.

O’Rahilly R, Muller F. 1987. Developmen-tal stages in human embryos. CarnegieInstitute of Washington, Publication637, Washington, DC.

Otaky N, Paus T, D’Avirro D, Gutmans D,MacDonald D, Caramanos Z, TomaiouloF, Evans AC. 1995. Volumetric analysisof the human cingulate, paracingulate,and superior rostral sulci. Soc NeurosciAbstr 21:154.

Palovcik RA, Reid SA, Principe JC, Albu-querque A. 1992. 3D computer anima-tion of electrophysiological responses.J Neurosci Meth 41:1–9.

Paus T, Tomaioulo F, Otaky N, MacDonaldD, Petrides M, Atlas J, Morris R, EvansAC. 1996. Human cingulate and paracin-gulate sulci: Pattern, variability, asym-metry and probabilistic map. Cereb Cor-tex 6:207–214.

Paxinos G, Watson C. 1986. The rat brainin stereotaxic coordinates. San Diego:Academic Press.

Paxinos G, Huang XF. 1995. The humanbrainstem: a cyto- and chemoarchitec-tonic study. San Diego: Academic Press.

Penfield W, Boldrey E. 1937. Somatic mo-tor and sensory representation in the ce-rebral cortex of man as studied by elec-trical stimulation. Brain 60:389–443.

Pommert A, Schubert R, Riemer M, Schi-emann T, Tiede U, Hohne KH. 1994.Symbolic modeling of human anatomyfor visualization and simulation. IEEEVis Biomed Comp 2359:412–423.

Rabbitt RD, Weiss JA, Christensen GE,Miller MI. 1995. Mapping of hyperelasticdeformable templates using the finite el-ement method. Proc SPIE 2573:252–265.

Rademacher J, Caviness VS Jr, SteinmetzH, Galaburda AM. 1993. Topographicalvariation of the human primary cortices:Implications for neuroimaging, brainmapping and neurobiology. Cereb Cor-tex 3:313–329.

Rajapakse JC, Giedd JN, Rumsey JM, Vai-tuzis AC, Hamburger SD, Rapoport JL.1996. Regional MRI measurements of thecorpus callosum: a methodological and de-velopmental study. Brain Dev 18:379–388.

Rajkowska G, Goldman-Rakic P. 1995. Cy-toarchitectonic definition of pre-frontal


areas in the normal human cortex: II.Variability in locations of areas 9 and 46and relationship to the Talairach coordi-nate system. Cereb Cortex 5:323–337.

Roland PE, Zilles K. 1994. Brain atlases—Anew research tool. Trends Neurosci 17:458–467.

Rizzo G, Gilardi MC, Prinster A, Grassi F,Scotti G, Cerutti S, Fazio F. 1995. Anelastic computerized brain atlas for theanalysis of clinical PET/SPET data. EurJ Nucl Med 22:1313–18.

Salamon G, Raynaud C, Regis J, RumeauC. 1990. Magnetic resonance imaging ofthe pediatric brain: an anatomical atlas.New York: Lippincott-Raven Publishers.

Sandor SR, Leahy RM. 1994. Matching de-formable atlas models to pre-processedmagnetic resonance brain images. ProcIEEE Conf on Image Processing 3:686–690.

Sandor SR, Leahy RM. 1995. Toward au-tomated labeling of the cerebral cortexusing a deformable atlas. In: Bizais Y,Barillot C, Di Paola R, editors. Info Pro-ceedings in Medical Imaging, June 1995.p 127–138.

Schaltenbrand G, Bailey P. 1959. Introduc-tion to stereotaxis with an atlas of thehuman brain. New York, Stuttgart: Thi-eme.

Schaltenbrand G, Wahren W. 1977. Atlasfor stereotaxy of the human brain, 2nded. New York, Stuttgart: Thieme.

Schiemann T, Nuthmann J, Tiede U,Hohne KH. 1996. Segmentation of thevisible human for high-quality volume-based visualization. Vis Biomed Comp4:13–22.

Schwartz EL, Merker B. 1986. Computer-aided neuroanatomy: differential geom-etry of cortical surfaces and an optimalflattening algorithm. IEEE Comp Graph-ics and Applications 6:36–44.

Seitz RJ, Bohm C, Greitz T, Roland PE,Eriksson L, Blomqvist G, Rosenqvist G,Nordell B. 1990. Accuracy and precisionof the computerized brain atlas programfor localization and quantification inpositron emission tomography. J CerebBlood Flow Metab 10:443–457.

Sereno MI, Dale AM, Liu A, Tootell RBH.1996. A surface-based coordinate systemfor a canonical cortex. Proceedings ofthe 2nd International Conference on Hu-man Brain Mapping, Boston, MA. Neu-roImage 3:S252.

Shantha TR, Manocha SL, Bourne GH.1968. A stereotaxic atlas of the java mon-key brain (Macaca irus). Basel: Karger.

Smith GE. 1907. A new topographical sur-vey of the human cerebral cortex, beingan account of the distribution of the an-atomically distinct cortical areas andtheir relationship to the cerebral sulci. JAnat 41:237–254.

Sowell ER, Thompson PM, Rex DE, Korn-sand DS, Jernigan TL, Toga AW. 2001.Mapping sulcal pattern asymmetry andcortical gray matter thickness in vivo:maturation of posterior perisylvian cor-tices. (in press).

Spitzer V, Ackerman MJ, Scherzinger AL,Whitlock D. 1996. The visible humanmale: a technical report (http://www.

nlm.nih.gov/extramural_research.dir/visible_human.html). J Am Med InformAssoc 3:118–130.

Spitzer VM, Whitlock DG. 1998. The visi-ble human dataset: the anatomical plat-form for human simulation. Anat Rec(New Anat) 253:49–57.

Steinmetz H, Furst G, Freund H-J. 1989.Cerebral cortical localization: applica-tion and validation of the proportionalgrid system in MR imaging. J Comp As-sist Tomogr 13:10–19.

Steinmetz H, Furst G, Freund H-J. 1990.Variation of perisylvian and calcarineanatomic landmarks within stereotaxicproportional coordinates. Am J Neurora-diol 11:1123–1130.

Stensaas SS, Eddington DK, Dobelle WH.1974. The topography and variability ofthe primary visual cortex in man. J Neu-rosurg 40:747–755.

Stewart JE, Broaddus WC, Johnson JH.1996. Rebuilding the visible man. VisBiomed Comp 4:81–86.

Stievenart JL, Iba-Zizen MT, Tourbah A,Lopez A, Thibierge M, Abanou A, Ca-banis EA. 1997. Minimal surface: a use-ful paradigm to describe the deeper partof the corpus callosum? Brain Res Bull44:117–124.

Swanson LW. 1992. Brain maps: structureof the rat brain. Amsterdam; New York:Elsevier.

Szabo J, Cowan WM. 1984. A stereotaxicatlas of the brain of the cynomolgusmonkey (Macaca fascicularis). J CompNeurol 222:265–300.

Talairach J, Szikla G. 1967. Atlasd’anatomie stereotaxique du telenceph-ale: etudes anatomo-radiologiques. Par-is: Masson & Cie.

Talairach J, Tournoux P. 1988. Co-planarstereotaxic atlas of the human brain.New York: Thieme.

Thirion J-P. 1995. Fast non-rigid matchingof medical images. INRIA Internal Report2547, Projet Epidaure, INRIA, France.

Thompson PM, Giedd JN, Blanton RE, Lind-shield C, Woods RP, MacDonald D, EvansAC, Toga AW. 1998b. Growth patterns inthe developing human brain detected us-ing continuum-mechanical tensor mapsand serial MRI, 5th International Confer-ence on Human Brain Mapping, Mon-treal, Canada.

Thompson PM, Giedd JN, Woods RP, Mac-Donald D, Evans AC, Toga AW. 2000d.Growth patterns in the developing braindetected by using continuum-mechani-cal tensor maps. Nature 404:190–193.

Thompson PM, MacDonald D, Mega MS,Holmes CJ, Evans AC, Toga AW. 1997.Detection and mapping of abnormalbrain structure with a probabilistic atlasof cortical surfaces. J Comp Assist To-mogr 21:567–581.

Thompson PM, Mega MS, Cummings JL,Toga AW. 2000f. Detecting dynamic (4D)profiles of degenerative rates in Alzhei-mer disease patients, using tensor map-ping and a population-based brain atlas.Proc Soc Neurosci.

Thompson PM, Mega MS, Toga AW.2000c. Disease-specific, probabilistic

brain atlases. Proceedings of the IEEEInternational Conference on Mathemat-ical Methods in Biomedical Image Anal-ysis (MMBIA2000), June 2000.

Thompson PM, Mega MS, Woods RP,Blanton RE, Moussai J, Zoumalan CI,Aron J, Cummings JL, Toga AW. 2000e.Cortical change in Alzheimer disease de-tected with a disease-specific popula-tion-based brain atlas. Cereb Cortex 11:1–16.

Thompson PM, Moussai J, Khan AA, Zo-hoori S, Goldkorn A, Mega MS, SmallGW, Cummings JL, Toga AW. 1998a.Cortical variability and asymmetry innormal aging and Alzheimer disease.Cereb Cortex 8:492–509.

Thompson PM, Narr KL, Blanton RE,Toga AW. 2000a. Mapping structural al-terations of the corpus callosum duringbrain development and degeneration, In:Iacoboni M, Zaidel E, editors. The cor-pus callosum. Kluwer Academic Press(in press).

Thompson PM, Schwartz C, Lin RT, KhanAA, Toga AW. 1996b. 3D statistical anal-ysis of sulcal variability in the humanbrain. J Neurosci 16:4261–4274.

Thompson PM, Schwartz C, Toga AW.1996a. High-resolution random mesh al-gorithms for creating a probabilistic 3Dsurface atlas of the human brain. Neuro-Image 3:19–34.

Thompson PM, Toga AW. 1996. A surface-based technique for warping 3-dimen-sional images of the brain. IEEE TransMed Imaging 15:1–16.

Thompson PM, Toga AW. 1997. Detection,visualization and animation of abnormalanatomic structure with a deformableprobabilistic brain atlas based on ran-dom vector field transformations. MedImage Anal 1:271–294.

Thompson PM, Toga AW. 1998a. Anatom-ically-driven strategies for high-dimen-sional brain image warping and pathol-ogy detection. In: Toga AW, editor. Brainwarping. San Diego: Academic Press. p311–336.

Thompson PM, Toga AW. 1998b. Mathe-matical/computational strategies for cre-ating a probabilistic atlas of the humanbrain, workshop on Statistics of BrainMapping, Centre de Recherches Math-ematiques, McGill University, Canada,June 13–14, 1998.

Thompson PM, Woods RP, Mega MS, TogaAW. 2000b. Mathematical/computationalchallenges in creating population-basedbrain atlases. Hum Brain Mapp 9:81–92.

Thurfjell L, Bohm C, Greitz T, Eriksson L.1993. Transformations and algorithmsin a computerized brain atlas. IEEETrans Nucl Sci 40:1167–1191.

Tiede U, Bomans M, Hohne KH, PommertA, Riemer M, Schiemann T, Schubert R,Lierse W. 1993. A computerized 3D atlasof the human skull and brain. Am J Neu-roradiol 14:551–559.

Toga AW, Ambach K, Quinn B, Hutchin M,Burton JS. 1994. Postmortem anatomyfrom cryosectioned whole human brain.J Neurosci Meth 54:239–252.


Toga AW, Mazziotta JC. 1996. Brain map-ping: the methods. San Diego: AcademicPress.

Toga AW, Thompson PM, Holmes CJ,Payne BA. 1996a. Informatics and Com-putational Neuroanatomy, Proceedingsof the American Medical Informatics As-sociation, Section S75-Brain Mappingand Image Databases, Washington, DC,Oct. 26–30, 1996. p 299–303.

Toga AW, Thompson PM, Payne BA.1996b. Modeling morphometric changesof the brain during development. In:Thatcher RW, Lyon GR, Rumsey J, Kras-negor N, editors. Developmental neuro-imaging: mapping the development ofbrain and behavior. San Diego: Aca-demic Press.

Toga AW, Thompson PM. 1998. An intro-duction to brain warping. In: Toga WA,editor. Brain warping. San Diego: Aca-demic Press.

Van Buren JM, Borke RC. 1972. Variationsand connections of the human thalamus,vol. 1,2. New York: Springer.

Van Buren JM, Maccubin D. 1962. An out-line atlas of human basal ganglia andestimation of anatomic variants. J Neu-rosurg 19:811–839.

Van Essen DC, Drury HA, Joshi SC, MillerMI. 1997. Comparisons between humanand macaque using shape-based defor-mation algorithms applied to corticalflat maps, 3rd International Conferenceon Functional Mapping of the HumanBrain, Copenhagen, May 19–23 1997.NeuroImage 5:S41.

Van Essen DC, Maunsell JHR. 1980. Two-dimensional maps of the cerebral cortex.J Comp Neurol 191:255–281.

Van Essen DC, Maunsell JHR. 1983. Hier-archical organization and functionalstreams in the visual cortex. Trends Neu-rol Sci 6:370–375.

Viola PA, Wells WM. 1995. Alignment bymaximization of mutual information,5th IEEE Int Conf on Computer Vision,16–23, Cambridge, MA.

Watson JDG, Myers R, Frackowiak RSJ,Hajnal JV, Woods RP, Mazziotta JC,Shipp S, Zeki S. 1993. Area V5 of thehuman brain: evidence form a combinedstudy using positron emission tomogra-phy and magnetic resonance imaging.Cereb Cortex 3:79–94.

Weis S, Kimbacher M, Wegner E. 1993.Morphometric analysis of the corpuscallosum using MR: correlation of mea-

surements with aging in healthy individ-uals. Am J Neuroradiol 14:637–645.

Wells WM, Viola P, Atsumi H, Nakajima S,Kikinis R. 1997. Multi-modal volumeregistration by maximization of mutualinformation. Med Image Anal 1:35–51.

Witelson SF. 1989. Hand and sex differencesin the isthmus and genu of the humancorpus callosum: a postmortem morpho-logical study. Brain 112:799–835.

Woods RP. 1996. Modeling for intergroupcomparisons of imaging data. NeuroIm-age 4:84–94.

Woods RP, Cherry SR, Mazziotta JC. 1992.Rapid automated algorithm for aligningand reslicing PET images. J Comp AssistTomogr 16:620–633.

Woods RP, Mazziotta JC, Cherry SR. 1993.MRI-PET registration with automatedalgorithm. J Comp Assist Tomogr17:536–546.

Yoon H, Shin YS, Lee KC, Park HW. 1997.Morphological characteristics of the de-veloping human brain during the embry-onic period. Yonsei Med J 38:26–32.

Zijdenbos AP, Dawant BM. 1994. Brain seg-mentation and white matter lesion detec-tion in MR images. Crit Rev Biomed Eng22:401–465.


Documents

Maps of the Brainusers.loni.ucla.edu/~thompson/PDF/TT_NewAnat2001.pdfREVIEW Maps of the Brain ARTHUR W. TOGA* AND PAUL M. THOMPSON We review recent developments in brain mapping and