RESPONSE PROPERTIES OF THE HUMAN FUSIFORM FACE AREA

RESPONSE PROPERTIES OF THE HUMAN FUSIFORMFACE AREA

Frank TongHarvard University Cambridge and Massachusetts General Hospital Nuclear Magnetic Resonance Center Charlestown USA

Ken NakayamaHarvard University Cambridge USA

Morris MoscovitchUniversity of Toronto Canada

Oren WeinribHarvard University Cambridge USA

Nancy KanwisherMassachusetts Institute of Technology Cambridge and Massachusetts General Hospital Nuclear Magnetic Resource Centre

Charlestown USA

We used functional magnetic resonance imaging to study the response properties of the humanfusiform face area (FFA Kanwisher McDermott amp Chun 1997) to a variety of face-like stimuli inorder to clarify the functional role of this region FFA responses were found to be (1) equally strong forcat cartoon and human faces despite very different image properties (2) equally strong for entire humanfaces and faces with eyes occluded but weaker for eyes shown alone (3) equal for front and profile viewsof human heads but declining in strength as faces rotated away from view and (4) weakest for nonfaceobjects and houses These results indicate that generalisation of the FFA response across very differentface types cannot be explained in terms of a specific response to a salient facial feature such as the eyes ora more general response to heads Instead the FFA appears to be optimally tuned to the broad categoryof faces

INTRODUCTION

Numerous sources of evidence suggest that primatebrains have special-purpose neural machinery thatis selectively involved in the perception of facesPhysiological measurements especially single-unitrecordings in macaques and event-related poten-tials in humans provide some of the richest sourcesof evidence on the specificity of these systems

However these techniques do not allow us to quan-tify responses from specific regions of the humanbrain The goal of the present effort was to provide adetailed characterisation of the response propertiesof a region of human extrastriate cortex called thefusiform face area (Kanwisher McDermott ampChun 1997) We begin with a brief outline ofthe neurophysiological evidence for face-specificneural systems

COGNITIVE NEUROPSYCHOLOGY 2000 17 (123) 257ndash279

Oacute 2000 Psychology Press Ltdhttpwwwtandfcoukjournalspp02643294html

257

Requests for reprints should be addressed to Frank Tong Department of Psychology Harvard University 33 Kirkland St Cam-bridge MA 02138 USA (Email frankwjhharvardedu)

This work was supported by a Human Frontiers and NIMH grant 56073 to NK and an NSERC postgraduate scholarship to FTWe thank Bruce Rosen and many people at the MGH- NMR Center at Charlestown MA for technical assistance and Yuan-Sea Leefor research assistance

Early evidence that there may be specialisedneural regions for face perception came from casesof prosopagnosia the selective loss of face recogni-tion abilities in patients with focal brain damage(eg Bodamer 1947 Meadows 1974) Howeverthe first neurophysiological evidence of such spe-cialisation came from the discovery of face-selectivecells in the temporal cortex of macaques (GrossRoche-Miranda amp Bender 1972) Since this pio-neering work many studies have demonstrated thatldquoface cellsrdquo may be tuned to certain facial attributessuch as the identity (Yamane Kaji amp Kawano1988) expression (Hasselmo Rolls amp Baylis1989) viewpoint (Perrett et al 1991) or parts of aface (Perrett Rolls amp Caan 1982 Yamane et al1988)

Intracranial recordings from the human tempo-ral lobes and hippocampus (carried out forpresurgical planning purposes) have also revealedindividual neurons that respond selectively to facesparticular facial expressions or gender (Fried Mac-Donald amp Wilson 1997 Heit Smith amp Halgren1988 Ojemann Ojemann amp Lettich 1992)Evoked potentials recorded from strip electrodesimplanted on the surface of the human brain haverevealed distinct regions in the fusiform andinferotemporal gyri that produce face-specificN200 responses to faces but not to cars butterfliesor other control stimuli (Allison et al 1994) Fur-thermore electrical stimulation of these regionsfrequently produced a temporary inability to namefamous faces suggesting that these cortical regionsare not only engaged by but necessary for facerecognition

Although intracranial recordings provideimpressive evidence for anatomically restrictedresponses to faces with this technique it is difficultto collect enough data to clearly establish the face-specificity of the responses Further one can neverbe sure that the organisation of epileptic or dam-aged brains resembles that of the normal popula-tion Noninvasive scalp recordings from normalsubjects using event-related potentials (ERP) ormagnetoencephalography (MEG) can circumventthese problems and indeed face-selective responseshave been reported in several studies (Jeffreys

1989 1996 Sams Hietanen Hari Ilmoniemi ampLounasmaa 1997)

ERP and MEG studies provide excellent tem-poral resolution but incomplete information aboutthe anatomical source of the signal By contrastrecently developed neuroimaging techniques pro-vide a method of localising functional signals withhigh spatial precision and therefore provide a criti-cal new perspective A large number of studies havedemonstrated activation of human ventralextrastriate cortex during viewing of faces (egHaxby et al 1994 Puce Allison Asgari Gore ampMcCarthy 1996 Sergent Ohta amp MacDonald1992) Although these earlier studies did notattempt to establish the selectivity of theseresponses two recent functional magnetic reso-nance imaging (fMRI) studies have addressed thisquestion McCarthy Puce Gore and Allison(1997) found that a discrete region in the rightfusiform gyrus responded preferentially to faces ascompared to flowers or common objects Thesefindings agreed with other anatomical evidenceshowing that lesions in prosopagnosic patients(Meadows 1974) and intracranial face-specificresponses (measured preoperatively) in epilepticpatients (Allison et al 1994) frequently involve thefusiform gyrus Kanwisher et al (1997) demon-strated that this region called the fusiform face area(FFA) responded in a highly selective fashion tofaces as compared to objects houses scrambledfaces or human hands even when subjects per-formed a demanding matching task that requiredattention to both face and nonface stimuli

Kanwisher et alrsquos results suggest that the FFAresponse is unlikely to be explained in terms of dif-ferences in the low-level properties of the stimuli ageneral response to anything animate or human ora tendency for subjects to attend more to faces thannonfaces during passive viewing tasks Howeverthey tell us little about what aspects of a face areresponsible for activating the FFA For example isthe FFA specifically tuned to human faces or does itrespond more broadly such that any type of face orthe mere presence of a head will fully activate itDoes FFA activity reflect a response to the configu-ration of the face alone without detailed informa-

TONG ET AL

258 COGNITIVE NEUROPSYCHOLOGY 2000 17 (123)

tion about face parts or conversely are facialfeatures sufficient to activate the FFA withoutinformation about the configuration of the face Bystudying the response of the FFA to a variety offace-like stimuli we hoped to elucidate the func-tional role of this area and address whether the FFAis primarily involved in face perception or someother type of processing such as gaze perception(Heywood amp Cowey 1992) or head recognition(Sinha amp Poggio 1996)

To address these questions the present experi-ments incorporated several critical design featuresFor each experiment we first functionally localisedeach subjectrsquos FFA on independent localiser scansconducted in the same session (see GeneralMethods also Kanwisher et al 1997)Examples ofthe localised FFA of two subjects are shown innear-axial and near-coronal slices in Plate 3 of thecolour section situated between pages 160 and 161(please click here to see Plate 3) Such individuallocalisation was crucial because the FFA can varyconsiderably in its anatomical location and spatialextent across subjects (Kanwisher et al 1997) Byspecifying our region of interest in advance wecould objectively measure the magnitude of FFAresponses on experimental scans without having tocorrect for multiple comparisons across all scannedvoxels In each experiment some subjects viewedstimuli passively while others performed a ldquoone-backrdquo matching task which obligated them toattend to all stimuli regardless of inherent interestTo the extent that similar data were obtained in thetwo tasks we could infer that the results did notreflect confounding differences in the engagementof visual attention by different stimuli

The most important methodological advance inthis study involved extending beyond simplepairwise comparisons to provide a richer descrip-tion of the response profile of the FFA across arange of visual stimuli It is well known that visualneurons tuned to simple or complex stimuli do notshow all-or-none responses but instead show agraded distribution of responses that peak aroundthe optimal stimulus By including four differentstimulus conditions within each experimental scanwe were able to quantify the magnitude of FFA

response to each stimulus class relative to others inthe same scan Each scan contained two referenceconditions (1) an optimal stimulus condition con-sisting of front-view human faces that are knownto produce strong FFA responses and (2) anonoptimal stimulus condition consisting of eithernonface objects or houses that produce weak FFAresponses typically less than half the magnitudefound for faces Each scan also contained two newstimulus conditions of interest which could then becompared to these optimal and nonoptimal refer-ence conditions Thus we could determine whethereach new stimulus category produced a weak FFAresponse no greater than those found for objects orhouses an intermediate level response or an opti-mal FFA response as strong as those found forfront-view human faces

GENERAL METHODS

Subjects

Eighteen healthy normal adults (eight women)ages 18ndash39 volunteered or participated for pay-ment in one to three of the following fMRI experi-ments All subjects had normal or corrected-to-normal vision and gave informed written consent toparticipate in the study Data from four subjectswere discarded because of artefacts caused by excesshead motion (two men) or because the FFA was notsuccessfully localised on independent scans (twomen)

Stimuli

For all experiments stimuli consisted of an equalnumber of grayscale images from each of four dif-ferent stimulus categories

Experimental Procedures

Each subject was run on (1) two or more functionallocaliser scans and (2) two or more scans from agiven experiment Each fMRI scan lasted a total

COGNITIVE NEUROPSYCHOLOGY 2000 17 (123) 259

PROPERTIES OF THE FUSIFORM FACE AREA

duration of 330 seconds and consisted of fourblocks of four consecutive 16-second stimulusepochs (one epoch for each stimulus condition)with a 16-second fixation baseline period occurringbefore each block (eg Fig 1B) Across the fourblocks each stimulus condition appeared once ineach serial position within a block Images from therelevant stimulus condition were serially presentedin random sequence during each epoch for subjectviewing During fixation baseline periods subjectsmaintained fixation on a central fixation point

In Experiments 1 and 2 images were centrallypresented at a rate of one image every 667msec(stimulus duration = 450msec interstimulus inter-val = 217msec) with a small spatial offset (10 ofthe image width) that alternated between a top-right and bottom-left position In Experiments 3and 4 images were centrally presented with no spa-tial offset at a rate of one image every 800msec(stimulus duration = 400msec interstimulus inter-val = 400msec)

Subjects either performed a passive viewing or aone-back matching task in each experiment Forthe passive viewing task subjects were simplyinstructed to attentively view the sequence ofimages In the one-back task subjects wereinstructed to press a button whenever they saw twoidentical pictures in a row Typically one or tworepetitions occurred in each epoch

MRI Scanning Procedures

Scanning was done on a 15 T scanner at theMGH-NMR Center in Charlestown MA using abilateral quadrature surface coil which provided ahigh signal to noise ratio in posterior brain regionsStandard echoplanar imaging procedureswere used(TR = 2sec TE = 70msec flip angle = 90deg 180degoffset = 25msec 165 imagesslice) Twelve 6mm or7mm thick near-coronal slices (parallel tobrainstem) covered the entire occipital lobe as wellas the posterior portion of the temporal lobesincluding the FFA A bite bar was used to minimisehead motion In all other respects imaging proce-dures were identical to those reported by Kanwisheret al (1997)

fMRI Data AnalysisFor each experiment each subjectrsquos FFA was iden-tified on separate functional localiser scans TheFFA was defined as the set of all contiguous voxelsin the fusiform gyrus that showed significant differ-ences in activity (P lt 0001 uncorrected incorpo-rating a 6-second delay to compensate forhaemodynamic lag) on a Kolmogorov-Smirnov testcomparing front-view human faces to either housesor familiar objects Because the localiser data wereused to independently define the FFA region ofinterest for experimental fMRI scans no correctionfor multiple voxel-wise comparisons was made

Each subjectrsquos predefined FFA was then used toextract the time course of the magnetic resonance(MR) signal intensity during the experimentalfMRI scans (averaging over all voxels identified inthe localiser scan) After assuming a 6-second delayin haemodynamic response FFA responses werethen measured in terms of the average percentagechange in MR signal during each stimulus epochcompared to fixation as a baseline

A repeated-measures analysis of variance wasperformed by pooling the mean FFA response foreach subject and stimulus condition with task (iepassive or one-back) as a between-subject variablePlanned comparisons were performed betweenstimulus conditions based on theoretical questionsof interest

Behavioural Data AnalysisThe percentage of correct responses on the one-back matching task was calculated for each subjectexperimental scan and stimulus condition in termsof the percentage of hits minus false alarms Table 1shows a summary of the mean performance acrosssubjects for Experiments 1ndash4 Discrimination per-formance was usually excellent (above 90) exceptfor stimuli that were perceived as highly visuallysimilar such as the schematic faces in Experiment1 the capped faces and eyes alone stimuli in Experi-ment 3 and the different head views in Experiment4 Because our main purpose was to study theresponse properties of the FFA and because wenever found an effect of task on FFA responses thebehavioural data were not analysed further

TONG ET AL


EXPERIMENT 1 HUMAN FACESCAT FACES SCHEMATIC FACESAND OBJECTS

This experiment tested whether the FFA respondsselectively to human faces or whether it generalisesto other faces such as cat or schematic faces Exam-ples of the stimuli are shown in Fig 1A and Appen-dix A

Although cat faces possess animate and expres-sive qualities and are readily perceived as faces theydiffer greatly from human faces in terms of theirlow-level features For example cat faces havehighly textured fur and are often distinguishedbased on the patterns of colours in the fur itselfRegarding facial features cats do not have lips orprominent eyebrows and their ears noses andmouths do not resemble those of human faces Theoverall shape and aspect ratio of cat faces also differfrom human faces Thus cat faces allowed us to testwhether the FFA is specifically tuned to humanfacial features or to more general aspects of a face

The use of cat face stimuli also allowed us to testwhether the FFA is involved in expert recognitionrather than face perception per se Some researchers

have suggested that the FFA response to humanfaces may instead reflect an effect of acquired exper-tise at identifying exemplars within a particular cat-egory (eg Gauthier Tarr Anderson Skudlarskiamp Gore 1997b) If such a view were correct onewould predict a much greater FFA response tohuman than cat faces given that people are far moreexperienced at recognising human faces Howeversingle-unit studies in monkeys have shown thatmost face cells generalise equally well across mon-key and human faces (eg Perrett et al 1982) Itwas therefore conceivable that the FFA might alsogeneralise to the faces of other species

Schematic faces allowed us to test whether abasic facial configuration would be sufficient toactivate the FFA Even newborn infants appear todifferentiate between simple schematic faces andscrambled faces as evidenced by their preference totrack a moving schematic face across a greater dis-tance (Goren Sarty amp Wu 1975 JohnsonDziurawiec Ellis amp Morton 1991) These resultssuggest that humans may have an innate preferencefor simple face-like visual patterns However sin-gle-unit recordings in monkeys have revealed thatface cells show either no response or a weak



Table 1 Summary of Behavioural Performance on one-back Matching Task Percentage ofHits - False Alarms

Stimulus ConditionmdashmdashmdashmdashmdashmdashmdashmdashmdashmdashmdashmdashmdashmdashmdashmdashmdashmdashmdashHuman Cat Schematic

Experiment 1 n faces faces faces Objects

Mean 4 92 90 72 95SE 78 16 48 30

Human Upright InvertedExperiment 2 n faces cartoon faces cartoon faces Objects

Mean 5 94 95 94 98SE 34 38 40 15

Entire Faces with EyesExperiment 3 n faces no eyes alone Houses

Mean 5 78 76 54 79SE 58 68 109 74

Experiment 4 n Front Profile Cheek Back

Mean 2 87 75 61 70SE 35 167 139 45

response to simple line drawings of a monkey face(Perrett et al 1982)

Unlike the schematic faces used in the abovestudies our schematic faces were defined by thearrangement of nonfacial features consisting ofsimple geometric shapes This allowed us to inves-tigate the extent to which facial configurationalone in absence of appropriate facial featureswould activate the FFA Front-view human facesand familiar (inanimate and animate) objects served

as optimal and nonoptimal reference conditionsrespectively

Method

SubjectsEight subjects (five women) participated in Experi-ment 1 four performed the passive viewing taskand four performed the one-back matching task

TONG ET AL


Fig 1 1A Overall FFA response (expressed in percentage MR signal change from fixation baseline) to human faces cat faces schematicfaces and objects averaged across eight subjects in Experiment 1 Example stimuli are shown 1B Time course showing mean FFA activitywhile subjects viewed sequences of human faces (F) cat faces (C) schematic faces (S) and objects (O) Note that FFA responses are delayedby approximately 6 seconds due to haemodynamic delay and that responses are strongest for human and cat faces in each of the four blocks

StimuliThe four stimulus conditions consisted of humanfaces cat faces schematic faces and objects In eachcondition 35 different images were used Figure1A shows an example of each of these stimuliAdditional examples can be seen in Appendix AFront-view human faces consisted of digitised col-lege ID photos of undergraduate men and womenCat faces were taken from a CD-ROM collectionof cats and kittens (Corel professional photos)Schematic faces were created by using differentrudimentary geometric shapes to compose the eyesnose mouth and external contour of each faceObjects included both easily recognisable inani-mate objects (eg camera iron tape cassette) andanimate objects (eg cow horse caterpillar) Allimages were cropped and presented within a squarewindow that subtended approximately 12deg times 12deg ofvisual angle

Results and Discussion

Figure 1A shows the overall mean FFA responseduring each of the four stimulus conditions(expressed in percentage MR signal change fromfixation baseline) The FFA responded moststrongly to human faces and cat faces much moreweakly to schematic faces and most weakly toobjects The consistency of these response differ-ences can be seen in the average time course of FFAactivity plotted in Fig 1B In each of the fourblocks the response to human and cat faces wasalways greater than the response to schematic facesor objects

An ANOVA across subjects revealed a highlysignificant difference among stimulus conditions[F(318) = 549 P lt 00001] There was no signifi-cant effect of task [F(16) lt 1] or interactionbetween task and stimulus [F(318) lt 1] (A sum-mary of mean FFA responses during passive view-ing versus one-back matching for all experimentscan be found in Table 2) Planned comparisonsrevealed significantly greater FFA activity forschematic faces than objects [t(7) = 366 P lt 01]and for cat faces than schematic faces [t(7) = 49P lt 002] However no significant difference was

found between human and cat faces [t(7) = 10P = 35]

The fact that the FFA responded as strongly tocat faces as human faces is quite striking given thatthe FFA was localised on independent scans as theregion in the fusiform gyrus that responded signifi-cantly more to human faces than to objects orhouses These results clearly indicate that the FFAcan generalise to faces with very different low-levelfeatures even those of another species Given thatour subjects were far more experienced at discrimi-nating and identifying human faces than cat faces itseems unlikely that the response of the FFA can beadequately explained in terms of visual expertisealone

The much weaker response to schematic facesindicates that the presence of facial configurationalone in absence of appropriate facial features isnot sufficient for fully activating the FFA Theweakness of this response does not seem attribut-able to the impoverished nature of the schematicfaces given that simplistic two-tone Mooney faceshave been shown to strongly activate the FFA(Kanwisher Tong amp Nakayama 1998) whereascomplex objects do not However our results dosuggest that facial configuration alone may lead to apartial or weak FFA response as evidenced by theslightly greater response to schematic faces thanfamiliar objects The present results agree with thefinding that face cells in monkeys generally showlittle or no response to simple schematic faces(Perrett et al 1982)

EXPERIMENT 2 HUMAN FACESUPRIGHT CARTOON FACESINVERTED CARTOON FACES ANDOBJECTS

Experiment 2 tested whether the FFA would gen-eralise to familiar upright and inverted cartoonfaces that were loosely based on either animals (egMickey Mouse) or fictitious characters (eg Ernieand Bert see Fig 2A and Appendix B for examplesof the stimuli) Unlike realistic faces cartoon facesare drawn in a highly schematic and exaggeratedfashion The facial features of cartoon faces often



deviate from normal faces in colour size shape andplacement Despite this fact cartoons are readilyperceived as animate faces perhaps much more sothan the schematic faces in Experiment 1

There is some neuropsychological evidence sug-gesting that human faces and cartoon faces access acommon face recognition system MoscovitchWinocur and Behrmann (1997) reported an objectagnosic patient CK who showed spared recogni-tion for human faces as well as familiar cartoonfaces CK could identify both human and cartoonfaces as well as normal controls but was much moreseverely impaired than normals when the samefaces were shown upside-down Moscovitch et alsuggested that CK had an intact face recognitionsystem that could only operate on upright faces butan impaired object recognition system that wouldnormally be used for identifying objects and alsocontribute to the identification of inverted facesConsistent with this distinction between uprightand inverted face processing Farah Wilson Drainand Tanaka (1995) reported a prosopagnosicpatient who showed the opposite deficit ofimpaired recognition for upright faces but normalrecognition for inverted faces

This experiment used the same cartoon facesthat were tested on patient CK to see if upright car-toon faces would activate the FFA as strongly asupright human faces We also presented the car-toon faces upside-down to see if this would result inweaker FFA activity and if so whether FFA activ-ity would drop to the level found for objects Giventhe neuropsychological and cognitive evidence thatface-specific processing is severely disrupted forinverted faces (Farah et al 1995 Moscovitch et al1997 Tanaka amp Farah 1993 Tong amp Nakayamain press Yin 1971 Young Hellawell amp Hay1987) one might expect that inverted faces shouldfail to access face-specific mechanisms Howeversingle-unit recordings in monkeys have revealedmixed results Whereas most face cells in the lateralinferotemporal cortex respond more to uprightthan inverted faces (Tanaka Saito Fukada ampMoriya 1991) cells in the superior temporal sulcusrespond equally strongly to both types of faces(Perrett et al 1982 1985)

In previous fMRI studies of human observerswe have found that FFA responses are only slightlystronger for upright than inverted grayscale humanfaces even though face discrimination performance

TONG ET AL


Table 2 Summary of Mean FFA Response as a Function of Stimulus Condition and Taskfor All Experiments

Stimulus ConditionndashndashndashndashndashndashndashndashndashndashndashndashndashndashndashndashndashndashndashndashndashndashndashndashndashndashndashndashndashndashndashndashndashndashndashndashndashHuman Cat Schematic


Passive 4 18 18 10 071-back 4 15 14 08 06


Passive 4 17 18 14 071-back 5 18 17 14 08


Passive 3 22 19 14 101-back 5 16 15 12 05


Passive 4 18 18 13 081-back 2 18 19 12 12

is much poorer for the inverted faces (Kanwisher etal 1998) By contrast two-tone Mooney faceswhich are difficult to perceive as faces at all afterinversion were found to yield a larger and moreconsistent FFA inversion effect

Given that the subjects in the present experi-ment readily saw the inverted cartoons as faces wepredicted that the FFA inversion effect for cartoonfaces would be rather small In fact even after inver-sion the cartoons often appeared easy to discrimi-nate and identify because they contained highlydistinctive features (eg the ears of Mickey Mouseor Bugs Bunny) We were therefore curiouswhether distinctive cartoon faces would still yieldan FFA inversion effect

Method

SubjectsNine subjects (seven women) participated inExperiment 2 four performed the passive viewingtask and five performed the one-back matchingtask

StimuliStimulus conditions consisted of upright cartoonfaces (eg Mickey Mouse Big Bird Ernie andBert) inverted cartoon faces (ie the same cartoonsshown upside-down) human faces (college IDphotos) and objects In each condition 34 differ-ent images were used All images were cropped andpresented within a square window that subtendedapproximately 12deg times 12deg of visual angle See Fig 2Aand Appendix B for examples


FFA responses were strongest for upright humanfaces and upright cartoon faces slightly weaker forinverted cartoon faces and weakest for objects (seeFig 2A) The consistency of these activation differ-ences including the cartoon inversion effect couldbe seen in the average fMRI time course (see Fig2B) as well as in the individual subject data

An ANOVA revealed a highly significant dif-ference among stimulus conditions [F(321) = 429P lt 00001] but no significant effect of task

[F(17) lt 1] or interaction between task and stimu-lus [F(321) lt 1] Planned comparisons revealednegligible differences in FFA response to uprightcartoon faces and human faces [t(8) = 056P = 59] This indicates that the FFA response gen-eralises equally well to upright cartoon faces as tohuman or cat faces (Experiment 1) despite theirvery different low-level image properties Theseresults are consistent with the preserved face recog-nition found in object agnosic patient CK whocould recognise both upright human and cartoonfaces equally well (Moscovitch et al 1997)

We found a small but significant decrease inFFA activity for inverted cartoon faces relative toupright cartoon faces [t(8) = ndash 674 P lt 0001]despite negligible differences in behavioural perfor-mance on the one-back matching task (invertedcartoons = 94 correct vs upright cartoons = 95correct) Compared to the small FFA inversioneffect (upright cartoons ndash inverted cartoons= 17 ndash 14 = 03) the FFA response toinverted cartoon faces was much greater than theresponse found for objects [14 vs 07 respec-tively t(8) = 470 P lt 002] in fact almost twice inmagnitude The weak FFA inversion effect foundhere contrasts with the severe recognition deficitthat CK showed for inverted cartoon faces(Moscovitch et al 1997) but agrees well with therather weak FFA inversion effects previously foundfor grayscale human faces (Kanwisher et al 1998)Our results suggest that inverted faces can accessface-specific mechanisms such as the FFA to a con-siderable extent though not quite as effectively asupright faces

EXPERIMENT 3 ENTIRE FACESFACES WITHOUT EY ES EYESALONE AND HOUSES

Experiment 3 investigated the extent to which theFFA response could be attributed to the presence ofthe eyes This was done by comparing the FFAresponse for entire human faces faces with eyesoccluded eyes alone and houses (see Fig 3A andAppendix C for examples of the stimuli)



One possible explanation of the generalisedFFA response found across human cat and car-toon faces is that it reflects a more basic response toa salient facial feature such as the eyes Perrett et al(1985) found that for many ldquoface cellsrdquo in superiortemporal sulcus selective responses for a particularface view (front or profile) could be better attributedto the gaze direction of the eyes Furthermorelesions of the superior temporal sulcus have beenfound to impair the discrimination of gaze direc-tion but have little effect on face recognition perfor-

mance suggesting that this region of monkey cor-tex may be specifically involved in gaze perception(Heywood amp Cowey 1992) ERP studies inhumans have also revealed an N170 potential overthe posterior temporal scalp that is greater for theeyes than for the nose mouth or entire face(Bentin Allison Puce Perez amp McCarthy 1996)Such an ERP may reflect an underlying gaze detec-tion mechanism However such a mechanism hasyet to be localised in the human brain By usingfMRI we were able to test whether the FFA is

TONG ET AL


Fig 2 Example stimuli overall FFA response (panel A) and time course of FFA activity (panel B) averaged across nine subjects inExperiment 2 during visual presentation of human faces (F) upright cartoon faces (U) inverted cartoon faces (I) and objects (O) Notethat FFA responses are strongest for upright human and cartoon faces somewhat weaker for inverted cartoon faces and weakest for objectsin all four blocks

selectively involved in gaze perception or involvedin more global aspects of face perception

Method

SubjectsEight subjects (four women) participated in Exper-iment 3 three performed the passive viewing taskand five performed the one-back matching task

StimuliFront-view digital images of 38 men wearing ablack ski hat were taken under controlled laboratoryconditions Stimuli consisted of 19 images of entirefaces a different set of 19 faces from which the eyeswere digitally removed (and used for the eyes alonestimuli) 19 images of eyes alone and 19 houses Allimages were cropped and presented within a squarewindow that subtended approximately 16deg times 16deg ofvisual angle except for the eyes alone stimuli whichsubtended approximately 6deg times 18deg See Fig 3A andAppendix C for examples


FFA responses were strongest for entire faces andfaces without eyes weaker for eyes alone and weak-est for houses (see Fig 3A) The consistency ofthese activation differences can be seen in the aver-age fMRI time course for each of the four blocks(see Fig 3B)

An ANOVA revealed a highly significant dif-ference among stimulus conditions [F(318) = 285P lt 00001] but no significant effect of task[F(16) = 10 P = 30] or interaction between taskand stimulus [F(318) lt 1] Planned comparisonsrevealed a negligible difference in FFA response toentire faces and faces without eyes [t(7) = 13 P= 23] but a significantly greater response to faceswithout eyes than eyes alone [t(7) = 330 P lt 02]These results indicate that the eyes are neither nec-essary nor sufficient to fully activate the FFA Itfurther suggests that the FFA is not selectivelyinvolved in gaze detection (cf Bentin et al 1996Heywood amp Cowey 1992) but instead is likely tobe involved in global aspects of face perception It

should also be noted that we found no other brainregions in our posterior slices that showed consis-tently greater responses to the presence of the eyesbased on comparisons between eyes versus housesor entire faces versus faces without eyes

The FFA response to eyes alone was muchgreater than the response to houses [t(7) = 63 P lt0005] even though the house stimuli were muchlarger and more visually complex Apparently asmall isolated facial feature such as the eyes can stillevoke a fairly large FFA response However whenthis feature was added to a face without eyes to forma complete face no increase in FFA response wasobserved This suggests that the FFA does notcombine information from various facial features ina strictly linear fashion (see also Perrett et al 19821985)

EXPERIMENT 4 FRONT PROFILECHEEK AND BACK OF HEADVIEWS

Experiment 4 investigated whether the FFAresponse is specific to faces or whether it reflects amore general response to a broader class of stimulisuch as different views of heads We tested four dif-ferent head views front view (0deg rotation) profileview (90deg) cheek view (135deg) and back view (180deg)Examples of the stimuli are shown in Fig 4A andAppendix D

A possible explanation of the generalised FFAresponse found across cat cartoon and humanfaces is that the FFA is more generally tuned toheads For example Sinha and Poggio (1996) havesuggested that head recognition rather than facerecognition is often used to identify individuals Bycontrast Perrett et al (1991) have suggested thatthe reason why most ldquoface cellsrdquo in the macaquesuperior temporal sulcus show specificity for vari-ous head views (including back of head and cheekviews) is that these neurons are involved in codingthe direction of another personrsquos attention ratherthan facial identity

We reasoned that if the FFA is generally tunedto heads then one would predict equally elevatedresponses across all head views However if the



FFA is more specifically tuned to faces one wouldpredict equally strong activity for front and profileviews in which the internal features of the faceremain visible but declining activity thereafter asthe face becomes progressively hidden from view

Method

SubjectsFive subjects (two women) participated in Experi-ment 4 Three subjects performed only the passive

viewing task one subject performed only the one-back task and one subject performed both tasks ondifferent scans in the same session

StimuliImages of 13 different men were taken from 4 dif-ferent viewpoints which served as the 4 differentstimulus conditions front view (0deg rotation) profileview (90deg) cheek view (135deg) and back view (180deg)All images were cropped and presented within a

TONG ET AL


Fig 3 Example stimuli overall FFA response (panel A) and time course of FFA activity (panel B) averaged across eight subjects inExperiment 3 during visual presentation of entire human faces (F) faces with no eyes (N) eyes alone (E) and houses (H) Note that FFAresponses are strongest for entire faces and faces with no eyes weaker for eyes alone and weakest for houses in all four blocks

square window that subtended approximately12deg times 12deg of visual angle See Fig 4A and AppendixD for examples of the stimuli


The FFA responded most strongly to front andprofile views more weakly to cheek views and mostweakly to back of head views (see Fig 4A) Theconsistency of these activation differences can beseen in the average fMRI time course for each of thefour blocks (see Fig 4B)

An ANOVA revealed a highly significant dif-ference among stimulus conditions [F(312) = 129P lt 0005] but no significant effect of task[F(14) lt 1] or interaction between task and stimu-lus [F(312) lt 1] Planned comparisons revealednegligible FFA response differences between frontand profile views [t(5) = ndash008 P = 94] but signif-icantly greater activity for profile than cheek views[t(5) = 401 P lt 02] and for cheek views than backof head views [t(5) = 286 P lt 05] Although anobject control condition was not included in thisexperiment the magnitude of FFA response tobacks of heads appeared comparable to or onlyslightly greater than the response typically foundfor non-face objects

FFA activity declined as the head rotatedbeyond profile view and the internal features of theface became progressively more occluded Thisstrongly suggests that the FFA responds specifi-cally to faces and not more generally to headsThese results also rule out the possibility that thegeneralised FFA response found for cat and cartoonfaces reflects a response to heads or animate formsin general All subjects clearly recognised that theywere viewing the back of human heads and somesubjects who were familiar with the persons showneven reported that they could identify people fromthe back of their heads Despite the fact that suchviews were readily perceived as animate and weresometimes even recognised as belonging to a cer-tain individual the FFA failed to show muchresponse to these views

GENERAL DISCUSSION

The experiments reported here show that the FFAresponse generalises equally well across cat facescartoon faces front-view human faces (with orwithout eyes) and profile-view human faces eventhough the low-level visual features that composethese stimuli differ greatly Such similar responsesto very heterogeneous stimuli cannot be explainedin terms of a more specific response to a salientfacial feature such as the eyes or a more generalresponse to heads or animate objects Instead ourresults strongly suggest that the FFA is optimallytuned to the broad stimulus category of faces

Whereas cat cartoon and human faces evokedthe strongest FFA responses stimuli that conveyedonly certain perceptual characteristics of a face (egsimple schematic faces inverted cartoon faces eyesalone and cheek views of heads) evoked intermedi-ate responses By contrast nonface stimuli such asobjects houses and backs of heads consistentlyyielded the weakest responses These results have anumber of implications regarding the functionalrole of the FFA and the underlying processes that itmay be performing We address these issues below

Is the FFA Selectively Involved in GazePerception

There is some evidence that superior temporalsulcus lesions in monkeys may selectively impair theability to perceive the direction of anotherrsquos gazewithout affecting face recognition performance(Heywood amp Cowey 1992) ERP recordings inhumans have also revealed an N170 potential overthe lateral occipital regions that is greater for theeyes than for the nose mouth or entire face (Bentinet al 1996) These studies suggest that there mayexist neural regions dedicated to the more specifictask of gaze perception

However our results in Experiment 3 revealedno evidence of neural regions that respond selec-tively to gaze Outside of the FFA we failed to findpreferential responses to the eyes Note howeverthat our brain slices extended from the occipitalpole to the posterior temporal lobes therefore it



remains possible that one might find gaze-selectiveresponses in more anterior regions

Within the FFA itself fMRI responses wereequally strong for entire faces and faces withouteyes but significantly weaker for eyes shown aloneThus the eyes are neither necessary nor sufficientfor evoking an optimal FFA response This sug-gests that FFA is not selectively involved in gazedetection but instead is involved in more globalaspects of face perception

Does FFA Activity Reflect the Processing ofFacial Features Facial Configuration orBoth

Although the processing of individual facial fea-tures contributes to face recognition cognitivestudies have generally emphasised the importanceof configural or holistic processing for effective facerecognition (Rhodes Brennan amp Carey 1987Tanaka amp Farah 1993 Young et al 1987) How-

TONG ET AL


Fig 4 Example stimuli overall FFA response (panel A) and time course of FFA activity (panel B) averaged across six subjects inExperiment 4 during visual presentation of front (F) profile (P) cheek (C) and back (B) views of human heads Note that FFA responsesare strongest for front and profile views weaker for cheek views and weakest for back views in all four blocks

ever recent investigation of object agnosic patientCK who appears to have a selectively spared facerecognition system suggests that both configuraland featural information are encoded within a uni-tary system (Moscovitch et al 1997) Our studylikewise suggests that both global configurationand local facial features serve to activate the FFA

In Experiment 1 we found that schematic facesdefined by simple geometric shapes activated theFFA only slightly more than nonface objects Theweak response to schematic faces indicates that thepresence of facial configuration alone in theabsence of appropriate facial features is insufficientfor evoking an optimal FFA response but can leadto a partial response

In Experiment 2 subjects viewed upright andinverted cartoon faces that provided the same low-level feature information However FFA responseswere significantly weaker for the inverted cartoonfaces (see also Kanwisher et al 1998) most proba-bly because configural processing was disrupted forthese faces It is well known that face inversionseverely disrupts configural or holistic processingbut generally has little effect on piecemeal process-ing of individual facial features (Tanaka amp Farah1993Tanaka amp Sengco 1997Young et al 1987)

These results suggest that neither configuralinformation alone nor feature information alone issufficient for evoking an optimal FFA responseInstead optimal FFA activation appears to requireboth appropriate facial features and appropriatefacial configuration These response propertiesmayreflect the fact that different FFA neurons are acti-vated by different aspects of the face Single-unitrecordings in monkeys have revealed that face cellsmay be tuned to a particular facial feature (eg eyesnose mouth chin) combination of features (egmouth and chin) or the overall configuration of aface (Perrett et al 1982 Yamane et al 1988) IfFFA neurons possess similar variations in tuningone would predict that the FFA population wouldbe most fully activated by a complete set of facialfeatures presented in a proper configuration Bycontrast one would predict that fewer FFA neu-rons would respond to a face with inappropriatefeatures (eg schematic faces) inappropriate con-

figuration (eg inverted faces) or to an isolatedfacial feature (eg eyes alone) as was found here

Why Does the FFA Respond to InvertedCartoon Faces

Although we found a significant inversion effect forcartoon faces the response to inverted cartoonsremained almost twice as strong as the response tononface objects Perhaps one reason why invertedcartoon faces and inverted grayscale human faces(Kanwisher et al 1998) evoke considerable FFAactivity is that many facial features such as the eyesremain highly salient even after configural process-ing is disrupted by inversion As Experiment 3revealed the eyes alone can evoke a fairly large FFAresponse

Another possible explanation which does notexclude the preceding one is that the FFA is largelyactivated by successful acts of face perception ratherthan successful acts of face recognition Inversiongenerally leads to rather small decreases in FFAactivity regardless of whether recognition perfor-mance is severely impaired as in the case ofgrayscale human faces (Kanwisher et al 1998) orunimpaired as was found for cartoon faces Evenwhen these faces are inverted they are still readilyperceived as faces By contrast two-tone Mooneyfaces which are difficult to perceive as faces onceinverted yield stronger FFA inversion effects(Kanwisher et al 1998) This suggests that theFFA may be primarily activated by the perceptionof a face

Supporting this view Tong NakayamaVaughan and Kanwisher (1998) recently foundthat when subjects continuously viewed a bistablefacehouse display under conditions of binocularrivalry the FFA showed a sharp increase in activitywhenever subjects reported a perceptual switchfrom house to face By contrast the FFA showed asharp decrease in activity during perceptualswitches from face to house FFA activity maytherefore be tightly related to the phenomenalexperience of seeing a face even when it is the sameface that is repeatedly perceived



Might the FFA be Primarily Involved inSubordinate-level Categorisation or ExpertRecognition

Are there alternative accounts that might explainthe apparent face-specificity of the FFA GauthierAnderson Tarr Skudlarski and Gore (1997a)have suggested that the FFA may reflect subordi-nate-level categorisation within a homogenousvisual category rather than face-specific processingHowever such an account seems unlikely giventhat Kanwisher et al (1997) previously showed thatFFA responses were almost four times greaterwhen subjects discriminated between individualfaces than between individual hands Here wefound that FFA responses were strongest for catfaces upright cartoons and human faces interme-diate for schematic faces inverted cartoons eyesalone and cheek views and weakest for houses andbacks of heads irrespective of whether subjects per-formed subordinate-level discriminations in theone-back matching task or a simple passive viewingtask (see Table 2 for summary) These results sug-gest that the FFA response is quite automatic andfar more dependent upon stimulus attributes thanupon the difficulty of the visual discriminationsinvolved (see Table 1) or the task at hand

More recently Gauthier et al (1997b) have sug-gested that the FFA may instead reflect subordi-nate-level categorisation within a visual categoryfor which one has acquired visual expertise Wethink that such a modified account is also unlikelygiven the very strong FFA responses found here forcat faces Although all of our subjects were far moreexperienced at recognising and individuatinghuman faces than cat faces the FFA showedequally strong activations to both face types Anexpertise account would also have difficultyexplaining why letters which represent anotherhighly overlearned visual category activate differ-ent brain regions than faces (Puce et al 1996)These findings suggest that the FFA response can-not be adequately explained in terms of the degreeof visual expertise that one has for a particular cate-gory Although some other modified subordinate-level classification account may still be viable we

believe that the simplest account requiring the few-est ad hoc assumptions is that the FFA is primarilyengaged in face perception

Possible Roles of the FFA in the DetectionPerception and Recognition of Faces

At present there is little evidence to suggest thatthe FFA is involved in recognition memory forindividual faces Neuroimaging studies have shownthat the FFA responds equally well to familiar ver-sus novel faces suggesting that the FFA might notbe the site of memory storage for faces (ClarkMaisog amp Haxby 1998Haxby et al 1996)Thesestudies do not preclude the possibility that the FFAis involved in recognition memory for faces butthey do not support such involvement

A plausible alternative is that the FFA plays arole in the detection of faces By face detection wemean the ability to discriminate between faces andnonfaces The present study revealed that the FFAis optimally activated by cat cartoon and humanfaces but is weakly activated by nonface stimuliThis suggests that the FFA may not be specialisedfor recognising human faces but instead may serve amore basic function in detecting any type of face Inother studies we have found that face inversionmost severely impairs the FFA response when itimpairs the subjectrsquos ability to detect the presence ofa face (Kanwisher et al 1998) and that the FFAautomatically responds when the same face isrepeatedly perceived in a bistable facehouse display(Tong et al 1998) These results suggest that theFFA response may reflect the conscious detectionof a face

Why might people develop a specialised mecha-nism for face detection A face detection mecha-nism could signal the presence of humans or otheranimals in the immediate environment and thusserve as an alerting mechanism providing impor-tant information for adaptive behaviour and sur-vival Face detection may also provide the necessaryinput for subsequent face recognition processesFor many computer vision algorithms a face mustfirst be located in a scene before face identification

TONG ET AL


can proceed (eg FaceIt 1998 Turk amp Pentland1991) and surprisingly face detection in complexstatic scenes can sometimes prove as challenging asface identification itself (Leung Burl amp Perona1995)

The FFA may also play a role in face perceptionperhaps by representing the underlying shape orstructure of a face Inversion of grayscale humanfaces impairs both the FFA response (Kanwisher etal 1998) and the perception of facial configuration(Tanaka amp Sengco 1997 Young et al 1987) eventhough such inverted images are readily detected asfaces Thus FFA activity may not only reflect facedetection but may also reflect the perception of afacersquos shape and configuration

One reason why face detection and face percep-tion may rely on a common mechanism is computa-tional efficiency A common representationalscheme such as a multidimensional ldquoface spacerdquo(eg Rhodes et al 1987 Valentine 1991) couldeffectively serve both functions Face detectionwould involve determining whether a particularitem sufficiently resembles the central tendency offaces whereas face perception would involve thecomplementary process of describing how a facedeviates from this central tendency Interestinglywhen subjects must distinguish between intact andscrambled faces in a speeded-response task facesthat are highly atypical and recognisable are actuallymore difficult to classify as faces (Valentine 1991)This suggests that face detection like face percep-tion may involve coding a face in terms of its devia-tion from the central tendency of faces

According to this framework the FFA may playan important role in face detection and face percep-tion and may ultimately serve as a gateway to thehigher-level areas where memories for individualfaces are stored FFA activity presumably reflectsthe activation of face-selective units that signal thepresence of a face These units may also provide astructural code for each face perhaps by represent-ing the local features and global configurationSuch structural information would provide the nec-essary input to higher-level recognition areas thatmatch the incoming face input to stored facememories

Concluding Remarks

We have shown that the FFA responds optimally toa variety of faces (humans cats and cartoons) andless strongly to stimuli that convey only certain per-ceptual characteristics of faces (eg schematic facesinverted cartoons eyes alone cheek views ofheads) The response profile of the FFA cannot beadequately explained in terms of low-level imageproperties given that upright and inverted faceswith common image properties evoke differentresponses and given that cat cartoon and humanfaces with very different image properties evokecomparable responses Instead the FFA responseprofile may be better understood in terms ofhigher-order representations for the general cate-gory of faces One possibility is that activity in theFFA is tightly linked to the phenomenal experienceof detecting or perceiving a face (Tong et al 1998)

Regarding the functional role of the humanfusiform face area our results suggest that the FFAis not primarily involved in gaze perception headdetection subordinate-level categorisation orexpert recognition Instead the FFA appears to bestrongly involved in face detection and face percep-tion and may play a role in processing the local fea-tures and global configuration of faces

REFERENCES

Allison T Ginter H McCarthy G Nobre ACPuce A Luby M amp Spencer DD (1994) Facerecognition in human extrastriate cortex Journal ofNeurophysiology 71 821ndash825

Bentin S Allison T Puce A Perez E amp McCarthyG (1996) Electrophysiological studies of face per-ceptions in humans Journal of Cognitive Neuroscience8 551ndash565

Bodamer J (1947) Die Prosopagnosie Archiv fuumlrPsychiatrie und Nervenkrankheiten 179 6ndash53

Clark VP Maisog JM amp Haxby JV (1998) fMRIstudy of face perception and memory using randomstimulus sequences Journal of Neurophysiology 793257ndash3265

FaceIt (1998) [Computer software] Jersey City NJVisionics Corporation



Farah MJ Wilson KD Drain HM amp TanakaJR (1995) The inverted face inversion effect inprosopagnosia Evidence for mandatory face-specificperceptual mechanisms Vision Research 35 2089ndash2093

Fried I MacDonald KA amp Wilson CL (1997)Single neuron activity in human hippocampus andamygdala during recognition of faces and objectsNeuron 18 753ndash765

Gauthier I Anderson AW Tarr MJ SkudlarskiPamp Gore JC (1997a) Levels of categorization invisual recognition studied using function magneticresonance imaging Current Biology 7 645ndash651

Gauthier I Tarr MJ Anderson AW SkudlarskiPamp Gore JC (1997b) Expertise training with novelobjects can recruit the fusiform face area Society forNeuroscience Abstracts 23(2) 2229

Goren CC Sarty M amp Wu PYK (1975) Visualfollowing and pattern discrimination of face-likestimuli by newborn infants Pediatrics 56 544ndash549

Gross CG Roche-Miranda GE Bender DB(1972) Visual properties of neurons in the infero-temporal cortex of the macaque Journal of Neuro-physiology 35 96ndash111

Hasselmo ME Rolls ET amp Baylis GC(1989) The role of expression and identity in theface-selective responses of neurons in the temporalvisual cortex of the monkey Behavioral BrainResearch 32 203ndash218

Haxby JV Horwitz B Ungerleider LG MaisogJM Pietrini P amp Grady CL (1994) The func-tional organization of human extrastriate cortex APET-rCBF study of selective attention to faces andlocations Journal of Neuroscience 14 6336ndash6353

Haxby JV Ungerleider LG Horwitz B MaisogJM Rapoport SI amp Grady CL (1996) Faceencoding and recognition in the human brain Pro-ceedings of the National Academy of Sciences USA 93922ndash927

Heit G Smith ME amp Halgren E (1988) Neuralencoding of individual words and faces by the humanhippocampus and amygdala Nature 333 773ndash775

Heywood CA amp Cowey A (1992) The role of thelsquoface-cellrsquo area in the discrimination and recognitionof faces by monkeys Philosophical Transactions of theRoyal Society of London Biological Sciences 335(1273)31ndash37

Jeffreys DA (1989) A face-responsive potentialrecorded from the human scalp Experimental BrainResearch 78 193ndash202

Jeffreys DA (1996) Evoked potential studies of faceand object processing Visual Cognition 3 1ndash38

Johnson MH Dziurawiec S Ellis H amp MortonJ (1991) Newbornsrsquo preferential tracking of face-like stimuli and its subsequent decline Cognition 401ndash19

Kanwisher N McDermott J amp Chun MM (1997)The fusiform face area A module in humanextrastriate cortex specialized for face perceptionJournal of Neuroscience 17(11) 4302ndash 4311

Kanwisher N Tong F amp Nakayama K (1998) Theeffect of face inversion on the human fusiform facearea Cognition 68 B1ndashB11

Leung TK Burl MC amp Perona P (1995June) Finding faces in cluttered scenes using randomgraph matching International Conference on Com-puter Vision Cambridge MA

McCarthy G Puce A Gore JC amp Allison T(1997) Face-specific processing in the humanfusiform gyrus Journal of Cognitive Neuroscience 9(5)604ndash609

Meadows JC (1974) The anatomical basis ofprosopagnosia Journal of Neurology Neurosurgery andPsychiatry 37 489ndash501

Moscovitch M Winocur G amp Behrmann M(1997) What is special about face recognitionNineteen experiments on a person with visual objectagnosia and dyslexia but normal face recognitionJournal of Cognitive Neuroscience 9(5) 555ndash604

Ojemann JG Ojemann GA Lettich E (1992)Neuronal activity related to faces and matching inhuman right nondominant temporal cortex Brain115 1ndash13

Perrett DI Oram MW Harries MH Bevan RHietanen JK Benson PJ amp Thomas S (1991)Viewer-centred and object-centred coding of heads inthe macaque temporal cortex Experimental BrainResearch 86 159ndash173

Perrett DI Rolls ET amp Caan W (1982) Visualneurones responsive to faces in the monkey temporalcortex Experimental Brain Research 47 329ndash342

Perrett DI Smith PAJ Potter DD Mistlin AJHead AS Milner AD amp Jeeves MA (1985)Visual cells in the temporal cortex sensitive to faceview and gaze direction Proceedings of the Royal Societyof London B 223 293ndash317

Puce A Allison T Asgari M Gore JC amp McCar-thy G (1996) Differential sensitivity of humanvisual cortex to faces letterstrings and textures Afunctional magnetic resonance imaging study Journalof Neuroscience 16 5205ndash5215

TONG ET AL


Rhodes G Brennan S amp Carey S (1987) Identifi-cation and ratings of caricatures Implications formental representations of faces Cognitive Psychology19(4) 473ndash497

Sams M Hietanen JK Hari R Ilmoniemi RJLounasmaa OV (1997) Face-specific responsesfrom the human inferior occipito-temporal cortexNeuroscience 77 49ndash55

Sergent J Ohta S MacDonald B (1992) Functionalneuroanatomy of face and object processing a posi-tron emission tomography study Brain 115 15ndash36

Sinha P amp Poggio T (1996) I think I know thatface Nature 384 404

Tanaka JW amp Farah MJ (1993) Parts and wholesin face recognition Quarterly Journal of ExperimentalPsychology 46A(2) 225ndash245

Tanaka JW amp Sengco JA (1997) Features andtheir configuration in face recognition Memory andCognition 25(5) 583ndash592

Tanaka K Saito H-A Fukada Y amp Moriya M(1991) Coding visual images of objects in theinferotemporal cortex of the macaque monkey Jour-nal of Neurophysiology 66 170ndash189

Tong F amp Nakayama K (in press) Robust represen-tations for faces Evidence from visual search Journalof Experimental Psychology Human Perception andPerformance

Tong F Nakayama K Vaughan JT amp KanwisherN (1998) Binocular rivalry and visual awareness inhuman extrastriate cortex Neuron 21 753ndash759

Turk M amp Pentland A (1991) Eigenfaces for recog-nition Journal of Cognitive Neuroscience 3 71ndash86

Valentine T (1991) A unified account of the effects ofdistinctiveness inversion and race in face recogni-tion Quarterly Journal of Experimental Psychology43A 161ndash204

Yamane S Kaji S amp Kawano K (1988) What facialfeatures activate face neurons in the inferotemporalcortex of the monkey Experimental Brain Research73 209ndash214

Yin RK (1969) Looking at upside-down faces Jour-nal of Experimental Psychology 81 141ndash 145

Young AW Hellawell D amp Hay DC (1987)Configural information in face perception Perception16 747ndash759



TONG ET AL


APPENDIX A

Examples of stimuli in Experiment 1



APPENDIX B


TONG ET AL


APPENDIX C




APPENDIX D


Plate 3 (Tong et al) Axial and coronal slices showing the fusiform face area (FFA) in two subjects Arrows point to the FFA whichincluded all contiguous voxels in the fusiform gyrus that showed significant differences in activity to faces minus objects or faces minus houses(P lt 0001 uncorrected on a Kolmogorov-Smirnov test see adjacent colour scale) on functional localiser scans Note that the right hemisphereis shown on the left side in all images and vice versa

Early evidence that there may be specialisedneural regions for face perception came from casesof prosopagnosia the selective loss of face recogni-tion abilities in patients with focal brain damage(eg Bodamer 1947 Meadows 1974) Howeverthe first neurophysiological evidence of such spe-cialisation came from the discovery of face-selectivecells in the temporal cortex of macaques (GrossRoche-Miranda amp Bender 1972) Since this pio-neering work many studies have demonstrated thatldquoface cellsrdquo may be tuned to certain facial attributessuch as the identity (Yamane Kaji amp Kawano1988) expression (Hasselmo Rolls amp Baylis1989) viewpoint (Perrett et al 1991) or parts of aface (Perrett Rolls amp Caan 1982 Yamane et al1988)

Intracranial recordings from the human tempo-ral lobes and hippocampus (carried out forpresurgical planning purposes) have also revealedindividual neurons that respond selectively to facesparticular facial expressions or gender (Fried Mac-Donald amp Wilson 1997 Heit Smith amp Halgren1988 Ojemann Ojemann amp Lettich 1992)Evoked potentials recorded from strip electrodesimplanted on the surface of the human brain haverevealed distinct regions in the fusiform andinferotemporal gyri that produce face-specificN200 responses to faces but not to cars butterfliesor other control stimuli (Allison et al 1994) Fur-thermore electrical stimulation of these regionsfrequently produced a temporary inability to namefamous faces suggesting that these cortical regionsare not only engaged by but necessary for facerecognition

Although intracranial recordings provideimpressive evidence for anatomically restrictedresponses to faces with this technique it is difficultto collect enough data to clearly establish the face-specificity of the responses Further one can neverbe sure that the organisation of epileptic or dam-aged brains resembles that of the normal popula-tion Noninvasive scalp recordings from normalsubjects using event-related potentials (ERP) ormagnetoencephalography (MEG) can circumventthese problems and indeed face-selective responseshave been reported in several studies (Jeffreys

1989 1996 Sams Hietanen Hari Ilmoniemi ampLounasmaa 1997)

ERP and MEG studies provide excellent tem-poral resolution but incomplete information aboutthe anatomical source of the signal By contrastrecently developed neuroimaging techniques pro-vide a method of localising functional signals withhigh spatial precision and therefore provide a criti-cal new perspective A large number of studies havedemonstrated activation of human ventralextrastriate cortex during viewing of faces (egHaxby et al 1994 Puce Allison Asgari Gore ampMcCarthy 1996 Sergent Ohta amp MacDonald1992) Although these earlier studies did notattempt to establish the selectivity of theseresponses two recent functional magnetic reso-nance imaging (fMRI) studies have addressed thisquestion McCarthy Puce Gore and Allison(1997) found that a discrete region in the rightfusiform gyrus responded preferentially to faces ascompared to flowers or common objects Thesefindings agreed with other anatomical evidenceshowing that lesions in prosopagnosic patients(Meadows 1974) and intracranial face-specificresponses (measured preoperatively) in epilepticpatients (Allison et al 1994) frequently involve thefusiform gyrus Kanwisher et al (1997) demon-strated that this region called the fusiform face area(FFA) responded in a highly selective fashion tofaces as compared to objects houses scrambledfaces or human hands even when subjects per-formed a demanding matching task that requiredattention to both face and nonface stimuli

Kanwisher et alrsquos results suggest that the FFAresponse is unlikely to be explained in terms of dif-ferences in the low-level properties of the stimuli ageneral response to anything animate or human ora tendency for subjects to attend more to faces thannonfaces during passive viewing tasks Howeverthey tell us little about what aspects of a face areresponsible for activating the FFA For example isthe FFA specifically tuned to human faces or does itrespond more broadly such that any type of face orthe mere presence of a head will fully activate itDoes FFA activity reflect a response to the configu-ration of the face alone without detailed informa-

TONG ET AL






GENERAL METHODS

Subjects


Stimuli















TONG ET AL













Mean 4 92 90 72 95SE 78 16 48 30


Mean 5 94 95 94 98SE 34 38 40 15


Mean 5 78 76 54 79SE 58 68 109 74


Mean 2 87 75 61 70SE 35 167 139 45




Method


TONG ET AL


















TONG ET AL





Passive 4 18 18 10 071-back 4 15 14 08 06


Passive 4 17 18 14 071-back 5 18 17 14 08


Passive 3 22 19 14 101-back 5 16 15 12 05


Passive 4 18 18 13 081-back 2 18 19 12 12



Method














TONG ET AL




Method















Method




TONG ET AL








GENERAL DISCUSSION












TONG ET AL






















TONG ET AL






Concluding Remarks



REFERENCES

































TONG ET AL


















TONG ET AL


APPENDIX A




APPENDIX B


TONG ET AL


APPENDIX C




APPENDIX D







GENERAL METHODS

Subjects


Stimuli















TONG ET AL













Mean 4 92 90 72 95SE 78 16 48 30


Mean 5 94 95 94 98SE 34 38 40 15


Mean 5 78 76 54 79SE 58 68 109 74


Mean 2 87 75 61 70SE 35 167 139 45




Method


TONG ET AL


















TONG ET AL





Passive 4 18 18 10 071-back 4 15 14 08 06


Passive 4 17 18 14 071-back 5 18 17 14 08


Passive 3 22 19 14 101-back 5 16 15 12 05


Passive 4 18 18 13 081-back 2 18 19 12 12



Method














TONG ET AL




Method















Method




TONG ET AL








GENERAL DISCUSSION












TONG ET AL






















TONG ET AL






Concluding Remarks



REFERENCES

































TONG ET AL


















TONG ET AL


APPENDIX A




APPENDIX B


TONG ET AL


APPENDIX C




APPENDIX D












TONG ET AL













Mean 4 92 90 72 95SE 78 16 48 30


Mean 5 94 95 94 98SE 34 38 40 15


Mean 5 78 76 54 79SE 58 68 109 74


Mean 2 87 75 61 70SE 35 167 139 45




Method


TONG ET AL


















TONG ET AL





Passive 4 18 18 10 071-back 4 15 14 08 06


Passive 4 17 18 14 071-back 5 18 17 14 08


Passive 3 22 19 14 101-back 5 16 15 12 05


Passive 4 18 18 13 081-back 2 18 19 12 12



Method














TONG ET AL




Method















Method




TONG ET AL








GENERAL DISCUSSION












TONG ET AL






















TONG ET AL






Concluding Remarks



REFERENCES

































TONG ET AL


















TONG ET AL


APPENDIX A




APPENDIX B


TONG ET AL


APPENDIX C




APPENDIX D














Mean 4 92 90 72 95SE 78 16 48 30


Mean 5 94 95 94 98SE 34 38 40 15


Mean 5 78 76 54 79SE 58 68 109 74


Mean 2 87 75 61 70SE 35 167 139 45




Method


TONG ET AL


















TONG ET AL





Passive 4 18 18 10 071-back 4 15 14 08 06


Passive 4 17 18 14 071-back 5 18 17 14 08


Passive 3 22 19 14 101-back 5 16 15 12 05


Passive 4 18 18 13 081-back 2 18 19 12 12



Method














TONG ET AL




Method















Method




TONG ET AL








GENERAL DISCUSSION












TONG ET AL






















TONG ET AL






Concluding Remarks



REFERENCES

































TONG ET AL


















TONG ET AL


APPENDIX A




APPENDIX B


TONG ET AL


APPENDIX C




APPENDIX D






Method


TONG ET AL


















TONG ET AL





Passive 4 18 18 10 071-back 4 15 14 08 06


Passive 4 17 18 14 071-back 5 18 17 14 08


Passive 3 22 19 14 101-back 5 16 15 12 05


Passive 4 18 18 13 081-back 2 18 19 12 12



Method














TONG ET AL




Method















Method




TONG ET AL








GENERAL DISCUSSION












TONG ET AL






















TONG ET AL






Concluding Remarks



REFERENCES

































TONG ET AL


















TONG ET AL


APPENDIX A




APPENDIX B


TONG ET AL


APPENDIX C




APPENDIX D


















TONG ET AL





Passive 4 18 18 10 071-back 4 15 14 08 06


Passive 4 17 18 14 071-back 5 18 17 14 08


Passive 3 22 19 14 101-back 5 16 15 12 05


Passive 4 18 18 13 081-back 2 18 19 12 12



Method














TONG ET AL




Method















Method




TONG ET AL








GENERAL DISCUSSION












TONG ET AL






















TONG ET AL






Concluding Remarks



REFERENCES

































TONG ET AL


















TONG ET AL


APPENDIX A




APPENDIX B


TONG ET AL


APPENDIX C




APPENDIX D





Method














TONG ET AL




Method















Method




TONG ET AL








GENERAL DISCUSSION












TONG ET AL






















TONG ET AL






Concluding Remarks



REFERENCES

































TONG ET AL


















TONG ET AL


APPENDIX A




APPENDIX B


TONG ET AL


APPENDIX C




APPENDIX D





TONG ET AL




Method















Method




TONG ET AL








GENERAL DISCUSSION












TONG ET AL






















TONG ET AL






Concluding Remarks



REFERENCES

































TONG ET AL


















TONG ET AL


APPENDIX A




APPENDIX B


TONG ET AL


APPENDIX C




APPENDIX D




Method















Method




TONG ET AL








GENERAL DISCUSSION












TONG ET AL






















TONG ET AL






Concluding Remarks



REFERENCES

































TONG ET AL


















TONG ET AL


APPENDIX A




APPENDIX B


TONG ET AL


APPENDIX C




APPENDIX D




Method




TONG ET AL








GENERAL DISCUSSION












TONG ET AL






















TONG ET AL






Concluding Remarks



REFERENCES

































TONG ET AL


















TONG ET AL


APPENDIX A




APPENDIX B


TONG ET AL


APPENDIX C




APPENDIX D








GENERAL DISCUSSION












TONG ET AL






















TONG ET AL






Concluding Remarks



REFERENCES

































TONG ET AL


















TONG ET AL


APPENDIX A




APPENDIX B


TONG ET AL


APPENDIX C




APPENDIX D







TONG ET AL






















TONG ET AL






Concluding Remarks



REFERENCES

































TONG ET AL


















TONG ET AL


APPENDIX A




APPENDIX B


TONG ET AL


APPENDIX C




APPENDIX D






















TONG ET AL






Concluding Remarks



REFERENCES

































TONG ET AL


















TONG ET AL


APPENDIX A




APPENDIX B


TONG ET AL


APPENDIX C




APPENDIX D







Concluding Remarks



REFERENCES

































TONG ET AL


















TONG ET AL


APPENDIX A




APPENDIX B


TONG ET AL


APPENDIX C




APPENDIX D




























TONG ET AL


















TONG ET AL


APPENDIX A




APPENDIX B


TONG ET AL


APPENDIX C




APPENDIX D



















TONG ET AL


APPENDIX A




APPENDIX B


TONG ET AL


APPENDIX C




APPENDIX D





APPENDIX B


TONG ET AL


APPENDIX C




APPENDIX D



TONG ET AL


APPENDIX C




APPENDIX D





APPENDIX D



RESPONSE PROPERTIES OF THE HUMAN FUSIFORM FACE AREA

Documents

Transcript of RESPONSE PROPERTIES OF THE HUMAN FUSIFORM FACE AREA