Use of Anatomical Parcellation to Catalogand Study Structure-Function Relationships

in the Human Brain

Natalie Tzourio,1 Laurent Petit,1 Emmanuel Mellet,1 Christophe Orssaud,1

Fabrice Crivello,1 Karim Benali,1 Georges Salamon,2 and Bernard Mazoyer1*

1Groupe d’Imagerie Neurofonctionnelle, UPRES-EA 2127 Universite de Caen et CEA-DSV-DRM,GIP Cyceron, Caen, France F-14074

2Service de Neuroradiologie, CHU La Timone, Marseille, France F-13000

r r

Abstract: We describe a functional neuroanatomy approach that combines structural (MRI) and functional(PET) data at the individual level. For each subject MRI dataset, sulci are first localized using hemispheresurface rendering and sections and stored. Using these landmarks, the subject brain volume is then dividedin 100 anatomical volumes of interest (AVOI). AVOI morphometric measurements are readily obtained aswell as functional parameters (CBF) after MRI-PET alignment. This approach allows structure-functionrelationship investigations both at the single case and at the intersubject average level; in addition,individual morphometric and functional parameters can be easily archieved in a database for furthermeta-analysis. This approach is applicable to all imaging modalities and is especially suited for a priorihypothesis testing and for the investigation of interindividual functional neuroanatomy variability. Hum.Brain Mapping 5:228–232, 1997. r 1997Wiley-Liss,Inc.

Key words: MRI; PET; anatomical volumes of interest; structure-function relationships; database; neuro-anatomy; brain activation

r r

INTRODUCTION

The existence of relationships between human brainstructures and functions has long been a matter ofphilosophical and religious debates that became opento scientific investigations only after Broca’s pioneer-ing work in the late 19th century. Nowadays, a largevariety of both structural and functional brain map-ping techniques have been made available, thanks tothe recent development of 3D neuroimaging methods,allowing the systematic investigation of these relation-

ships at different scales. At the macroscopic level,structural information is available either by means ofindividual MRI brain images or through the use of abrain atlas. However, the use of an atlas, such as theone designed by Talairach in 1988, is unlikely to beoptimal for studying structure-function relationshipsbecause it contains very poor structural information:anatomy is detailed only on a small number of sectionsof one hemisphere of a single individual, making themquite inaccurate at the cortical level. Another option isto use the individual gyral/sulcal anatomy to segre-gate the brain in anatomical volumes of interest.Gyral/sulcal anatomy is the usual reference neuroana-tomical space for various communities (neurophysiol-ogy, neuroanatomy, neurosurgery, neurology, neuropsy-chology) and constitutes a natural framework for the

*Correspondence to: Bernard Mazoyer, GIN, UPRES 2127,GIP Cyceron, BP 5229, 14074 Caen Cedex, France. E-mail:mazoyer@cyceron.frReceived for publication 3 April 1997; accepted 29 April 1997

r Human Brain Mapping 5:228–232(1997)r

r 1997Wiley-Liss,Inc.

registration of different brain images. In this paper, wepresent a brain parcellation method based on MRI andits application to the analysis of functional images. Asimilar approach has been designed by another group,albeit in a different context [Rademacher et al., 1992].

BRAIN PARCELLATION

For each individual, brain parcellation is achievedusing an MRI 3D dataset that allows a user to perform,by means of a dedicated software (Voxtool; GeneralElectric, Buc, France), brain segmentation and recon-struction of the external surfaces of both hemispheres,as well as sections in any incidence.

Sulcus identification

Sulci identification is based both on the specificmorphological properties of each sulcus and on theirconstant topological relationships. The possibility ofusing multiple incidences, as given by the Voxtoolsoftware, is essential for accurate anatomical analysisbecause it allows use of multiple features for sulcusrecognition. For example, the rolandic sulcus can beidentified 1) on a medial sagittal view as the firstsulcus anterior to the cingulate sulcus [Steinmetz et al.,1990], 2) on lower axial slices as the sulcus thatintersects neither the superior frontal sulcus nor theintraparietal sulcus while presenting a specific curva-ture, known as the rolandic genu, that corresponds tothe hand projection [Rumeau et al., 1994], 3) on lateralsagittal incidences, using a recognizable pattern madeof the ascending branch of the sylvian fissure, followedby the lower part of the precentral gyrus, and then bythe rolando sulcus [Ebeling et al., 1989]. When neces-sary, white matter morphological features can also beused for sulcus identification [Iwasaki et al., 1991].Once identified, the whole sulcus course is then markedusing an interactive cursor that appears simulta-neously on surface views and sections, and stored as awireframe that is further used to mark the sulcusposition on the series of axial MRI slices where itappears. Twenty sulci per hemisphere can be identifiedthis way (see sulci list in Table I), and projected onto acortical surface rendering (see Fig. 1).

Anatomical volumes of interest definition

In our approach, anatomical volumes of interest(AVOI) may be gyri or gray nuclei, but also pieces ofgyri when anatomical landmarks exist that allow theirsubdivision. For example, the inferior frontal gyruscan be divided in pars opercularis, pars triangularis andpars orbitaris thanks to the ascending and horizontal

TABLE I. List and abbreviated names of sulciand anatomical volumes of interest used

for brain parcellation

Sulci for parcellationAnatomical volumes of

interest

Calloso-marginal: cmParacentral: paracSubparietal: spParieto-occipital: poCalcarine: calcCentral: rolPrecentral: precSylvian: SySuperior frontal: f1Middle frontal: f2Sylvian, ascending: SyvascSylvian, horizontal: SyhorPostcentral: postcIntraparietal: ipsJensenSuperior temporal: t1Middle temporal: t2Anterior occipital: antoccLateral occipital: latoccMeynert: Meyn

Precentral gyrus: PrecSuperior frontal gyrus (BA 6,

BA 8, per se, orbitaris): F1Middle frontal gyrus (BA 6,

BA 8, per se, orbitaris): F2Inferior frontal gyrus: F3

(opercularis: F3op, triangu-laris: F3t, orbitaris: F3O)

Dorsolateral prefrontal cortexMedian frontal gyrus (orbi-

taris, per se): MFSupplementary motor area:

SMA

Postcentral gyrus: PostcSuperior parietal gyrus: P1Inferior parietal gyrus: P2Angularis gyrus: AGSupramarginalis gyrus: SMGParacentral gyrus: ParacPrecuneusIntraparietal sulcus cortex

Superior temporal gyrus: T1(Heschl gyrus, Planum tem-porale, lateral, temporalpole)

Middle temporal gyrus: T2Inferior temporal gyrus: T3Parahippocampal region

Calcarine cortex: CalcCuneusLateral occipital gyrus (supe-

rior, inferior): SO, IOFusiform and lingual gyri: FL

ThalamusLentiform nucleusCaudate nucleus

Cerebellum (vermis, hemi-spheric cortex)

When available, subdivisions of gyri are listed in parentheses. BA,Brodmann’s area.

r Functional Anatomy UsingBrain Parcellation r

r 229 r

Figure 1.Typical parcellation scheme of the brain MRI of one subject. Lateral and median surface views of asegmented right hemisphere with the sulci (top) used for the parcellation of the hemisphere involumes of interest (bottom).

rT

zourioetal.

r

r230

r

branches of the sylvian fissure. To subdivide thesuperior and middle frontal gyri, for which no anatomi-cal landmarks are available, we used the AC-PCreference system and the Talairach atlas to delineateAVOIs corresponding to Brodmann’s areas 6 and 8.Although it is clear that these are neither anatomicalnor cytoarchitectonic regions, we defined them inorder to increase the parcellation of the frontal lobe.This was done in particular for the study of theoculomotor system functional anatomy, because fron-tal eye field localization was supposed to be in Brod-mann’s area 8. In the same vein, the VAC plane is usedto separate the lateral part of the superior temporalgyrus from the temporal pole.

Two-dimensional regions of interest correspondingto the intersection of each AVOI with the MRI axialslices aligned to the PET coordinates are delineatedusing an interactive cursor; 3D AVOI are thus definedas the logical union of their pieces on the different MRIaxial slices. Fifty AVOIs are usually defined per hemi-sphere (see AVOI list in Table I and Fig. 1), which canbe further combined to obtain larger regions, such aslobes. Structural parameters (surface, volume, MRIsignal intensity) are eventually computed for eachAVOI.

Brain parcellation applied to functional images

Functional parameters, such as tracer concentration,can be computed for the AVOI set after alignment ofthe functional and MRI images using either custom orstandardized algorithms such as AIR. For PET-H2O15

activation experiments, each AVOI NrCBF during eachcondition is estimated as the ratio (in percent) of theradioactivity concentration in the AVOI to that of thewhole brain.

FUNCTIONAL NEUROANATOMY USINGBRAIN PARCELLATION

Intersubject averaging

Statistical analysis of AVOI functional parametersmeasured in a sample of subjects can be readilyconducted using standard techniques such as ANOVA,correlation, and path analysis, making it an alternativeto the widely used stereotactic averaging approach[Fox et al., 1985]. The method has, however, somespecific features. First, it is especially suited for testingspecific hypotheses made on the involvement of agiven AVOI in a given cognitive function. As anexample, this approach has allowed us to demonstratethat the location of frontal eye fields in humans is the

precentral gyrus [Petit et al., 1993] and not Brodmann’sarea 8 as previously hypothesized on the basis ofanimal studies. Second, the AVOI method is also usefulfor testing the existence of functional asymmetries: in aseries of experiments on speech processing we werethus able to demonstrate that NrCBF leftward asymme-tries in lateral temporal AVOIs depend on the speechstimulus linguistic content [Mazoyer et al., 1993].Finally, this method also offers the opportunity to lookat correlation between structural and functional param-eters in the same AVOIs.

Single case analysis and interindividual functionalneuroanatomy variability

The brain parcellation approach is specially tailoredfor analyzing structure-function relationships at thesingle case level. When combined to powerful detec-tion algorithm [Poline and Mazoyer, 1994], it providesan anatomical description of activated areas in eachindividual. Constant relationships of activated areaswith nearby sulci and gyri can thus be studied, as wellas their variabilities between subjects. For example, werecently demonstrated in single cases that an activa-tion focus was lying in the deepness of the intrapari-etal sulcus when the subjects performed a previouslylearned saccade sequence [Petit et al., 1996]. Similarly,the pars opercularis of the inferior frontal gyrus (IFG)was shown to be the frontal area activated by silentverb generation in right-handers [Crivello et al., 1995].Interestingly, the activation was located on the left sidein all but one subject, whose activation was lying in theexact symmetric region, namely the right IFG parsopercularis.

Database

The brain parcellation approach offers a convenientframework for building a database of both structuraland functional parameters measured across subjectsperforming various cognitive tasks. In this kind ofdatabase, as opposed to ones based on literature,individual raw data, namely AVOI structural param-eters and radioactive concentrations measured duringeach PET acquisition, are available as one non-imagingdata, such as age, gender, and psychometric measure-ments. Our group is currently building such a database[Benali et al., 1996] with the prospect of runningmeta-analyses across protocols.

DISCUSSION

In our approach, individual neuroanatomical dataare mandatory. Ten years ago, this requirement would

r Functional Anatomy UsingBrain Parcellation r

r 231 r

have severely limited its applicability. Nowadays,acquisition of brain MRI data prior to functionalstudies is common, in particular as fMRI develops. Themethod also requires extensive human observer inter-vention for sulcus identification and for volume ofinterest delineation. These limitations will be partlyovercome, thanks to the advent of automated algo-rithms for both tasks. A more intrinsic limitation of theproposed approach arises when one analyzes intersub-ject averaged data, in that both activation and deactiva-tion may coexist within the same AVOI. Note that thiskind of limitation also exists for stereotactically aver-aged data because of the large applied filter. A possibleway of circumventing this limitation is to complementthe intersubject analysis by single case detection. Fi-nally, one must acknowledge that the resolution atwhich structure-function relationships can be de-scribed using this method is limited to gyrus size (i.e.one to several centimeters).

The method is a first attempt to establish ‘‘true’’structure-function relationships over the entire brainvolume, combining structural and functional imagesboth at the individual and at the population levels. It isapplicable to all imaging modalities and can accommo-date ‘‘abnormal’’ brains whereas other approachesmay have troubles: this would be the case whendealing with brains bearing lesions or with infantbrains. Applied to the analysis of functional brainimages, the brain parcellation approach nicely offersthe opportunity to conduct specific statistical analysesthat would be hard to perform otherwise, such astesting a priori hypothesis about the involvement of aspecific AVOI or the existence of an activation asymme-try in an AVOI, or for studying interindividual func-tional variability. As such, it should thus be viewed asa complementary rather than a competitive approachwhen compared to stereotactic averaging methods.

REFERENCES

Benali K, Tzourio N, Petit L, Mellet E, Mazoyer BM (1996): BIRD: Abrain imaging relational database. NeuroImage 3:S112.

Crivello F, Tzourio N, Poline JB, Woods RP, Mazziotta JC, Mazoyer B(1995): Intersubject variability in functional neuroanatomy ofsilent verb generation: Assessment by a new activation detectionalgorithm based on amplitude and size information. NeuroIm-age 2:253–263.

Ebeling U, Steimetz H, Huang Y, Kahn T (1989): Topography andidentification of the inferior precentral sulcus in MR imaging.Am J Neuroradial 10:937–942.

Fox PT, Perlmutter JSA, Raichle ME (1985): A stereotactic method ofanatomical localization for positron emission tomography. JComput Assist Tomogr 9:141–153.

Iwasaki S, Nakagawa H, Fukusumi A, Kichikawa K, Kitamura K,Otsuji H, Uchida H, Ohishi H, Yaguchi K, Sumie H, Kuru Y(1991): Identification of pre- and postcentral gyri on CT and MRimages on the basis of the medullar pattern of cerebral whitematter. Radiology 179:207–213.

Mazoyer B, Dehaene S, Tzourio N, Frale G, Murayama N, Cohen L,Solomon G, Synota A, Mechler J (1993): The cortical representa-tion of speech. J Cogn Neurosci 5:467–479.

Petit L, Orssaud C, Tzourio N, Salamon G, Mazoyer B, Berthoz A(1993): PET study of voluntary saccadic eye movements inhumans: Basal ganglia-thalamocortical system and cingulatecortex involvement. J Neurophysiol 69:1009–1016.

Petit L, Orssaud C, Tzourio N, Crivello F, Berthoz A, Mazoyer B(1996): Functional anatomy of a pre-learned sequence of horizon-tal saccades in humans. J Neurosci 16:3714–3726.

Poline JB, Mazoyer BM (1994): Analysis of individual brain activa-tion maps using hierarchical description and multiscale detec-tion. IEEE Trans Med Imag 13:702–710.

Rademacher J, Galaburda AM, Kennedy DN, Filipek PA, Caviness JrVS (1992): Human cerebral cortex: Localization, parcellation, andmorphometry with magnetic resonance imaging. J Cogn Neuro-sci 4:352–374.

Rumeau C, Tzourio N, Murayama N, Peretti-Vitton P, Levrier O,Joliot M, Mazoyer B, Salamon G (1994): Location of handsensorimotor cortex: Magnetic resonance and functional correla-tion AJNR 14:915–925.

Steinmetz H, Furst G, Freund HJ (1990): Variation of left perisylvianand calcarine anatomical landmarks with stereotaxic propor-tional coordinates. Am J Neuroradial 11:1123–1130.

Talairach J, Tournoux P (1988): Co-Planar Stereotaxic Atlas of theHuman Brain. Stuttgart: Thieme-Verlag.

r Tzourio et al. r

r 232 r