Brain (1992), 115, 15-36
FUNCTIONAL NEUROANATOMY OF FACE AND OBJECT PROCESSING A POSITRON EMISSION TOMOGRAPHY STUDY by JUSTINE SERGENT, SHINSUKE OHTA and BRENNAN MACDONALD (From the Montreal Neurological Institute, McGill University, Montreal, Canada)
SUMMARY
INTRODUCTION
The human face holds a special place among visual objects. Any social animal must possess the capacity to differentiate and recognize members of its group and, in humans, the face is the most distinctive attribute for indexing identity reliably. For the face to assume such an indexing role, the processing organism must be endowed with structures and mechanisms adapted to performing the operations required by this function, and this implies a cognitive and structural architecture different from that required in the processing of other objects that are not called to play such a role. From a cognitive standpoint, there are several ways in which the processing of faces differs from that of other objects. For instance, most objects are processed at the basic category level (Rosch et al., 1976) regardless of the particularities of each object, whereas each instance of faces is treated as different within the category of faces and the differentiation of each face allows access to related stored biographical information. Correspondence to: Justine Sergent, Cognitive Neuroscience Laboratory, Montreal Neurological Institute, 3801 University Street, Montreal (Quebec) H3A 2B4, Canada. © Oxford University Press 1992
Downloaded from by guest on April 24, 2015
Studies of brain-damaged patients have revealed the existence of a selective impairment of face processing, prosopagnosia, resulting from lesions at different loci in the occipital and temporal lobes. The results of such studies have led to the identification of several cortical areas underlying the processing of faces, but it remains unclear what functional aspects of face processing are served by these areas and whether they are uniquely devoted to the processing of faces. The present study addresses these questions in a positron emission tomography (PET) study of regional cerebral blood flow in normal adults, using the "oxygen water bolus technique. The subjects participated in six tasks (with gratings, faces and objects), and the resulting level of cerebral activation was mapped on images of the subjects' cerebral structures obtained through magnetic resonance and was compared between tasks using the subtraction method. Compared with a fixation condition, regional cerebral blood flow (rCBF) changes were found in the striate and extrastriate cortex when subjects had to decide on the orientation of sine-wave gratings. A face-gender categorization resulted in activation changes in the right extrastriate cortex, and a face-identity condition produced additional activation of the fusiform gyrus and anterior temporal cortex of both hemispheres, and of the right parahippocampal gyrus and adjacent areas. Cerebral activation during an object-recognition task occurred essentially in the left occipito-temporal cortex and did not involve the right hemisphere regions specifically activated during the face-identity task. The results provide the first empirical evidence from normal subjects regarding the crucial role of the ventro-medial region of the right hemisphere in face recognition, and they offer new information about the dissociation between face and object processing.
16
J. SERGENT AND OTHERS
Downloaded from by guest on April 24, 2015
In addition, a large diversity of information about an individual can be derived on the sole basis of the physical attributes of a face, irrespective of its identity, and inferences about age, gender, race, emotion, the charm of the face, and so on, can be made that provide valuable information relevant to the bearer of the face. Another way faces differ from other objects, therefore, seems to be the variety of categories into which faces can be classified, with the same facial attributes being processed and combined differently depending on the information one wishes to access about an individual (Sergent, 1989). In contrast, for practical purposes, most objects belong to a single category and do not lend themselves to multiple categorizations. What makes face recognition a complex process is the large number of instances with which we are confronted, which requires the extraction, from a general configuration common to all faces, of the particularities that make each face unique. This implies refined perceptual mechanisms capable of detecting subtle differences among faces and of achieving a structural representation that uniquely defines each face (Sergent, 1989). Each representation must then be stored reliably, and this imposes added constraints on the processing organism given the many faces which we have to remember. Moreover, for a perceived face to be recognized as familiar, further operations must be implemented to make contact between perceived and stored facial information, and these operations are concerned with extracting the physiognomic invariants embedded in a given facial representation that may take diverse appearances resulting from the variety of possible formats, poses, angles of view, expressions, distance or illumination. Additional processes must then be performed for a face to be identified, and these involve the reactivation of episodic and semantic (i.e. biographic) information related to that face and without which it would remain that of a stranger. In spite of its complexity, face recognition is an automatic, effortless and quasi-infallible process, and this attests to the high level of efficiency of the cerebral structures sustaining this function. A detailed specification of the anatomical architecture of face recognition has been difficult to achieve, however, and the only source of information bearing on this issue has so far come from the study of prosopagnosic patients. The correlation between, on the one hand, the behavioural impairments related to the inability to process faces and to associated deficits and, on the other hand, the location of the lesions, has offered the opportunity to identify cortical areas the damage of which produced prosopagnosia (Meadows, 1974; Damasio etal., 1982, 1990a). Yet the reliance on data from neurological patients to infer the structural and functional organization of the cerebral structures normally underlying face recognition is confronted with a series of theoretical and methodological difficulties. First, enquiring about the location of the lesions responsible for prosopagnosia and asking which cortical areas underlie the normal processes inherent in face recognition are distinct questions that may not have the same answers. For instance, a focal damage may have detrimental effects on the functions of distant intact areas and, within a damaged brain, local structural integrity does not guarantee normal functioning (Monakow, von, 1910; Sergent and Signoret, 1992). Secondly, there is such a large variability among patients in the aetiology and location of the damage associated with the occurrence of prosopagnosia that it has been difficult to achieve a comprehensive view of the biological architecture of face recognition. Thus, even though the ventro-medial occipito-temporal junction of the right hemisphere has been considered a critical area in the processing
PET STUDY OF FACE AND OBJECT PROCESSING
17
Downloaded from by guest on April 24, 2015
of faces, cases have been reported in whom the lesion did not invade this particular region (e.g. Damasio et al., 1982; Bruyer et al, 1983; Sergent and Poncet, 1990). Similarly, although it is widely held that bilateral damage underlies the occurrence of prosopagnosia (e.g. Meadows, 1974; Damasio, 1985), cases have been reported whose Magnetic Resonance Imaging (MRI) scans showed no sign of left-hemisphere involvement {see Michel et al., 1989, for a review). Thirdly, the problem is further complicated by the fact that prosopagnosia, albeit an essentially selective impairment, has so far never occurred in a pure form. Every case of prosopagnosia that has been thoroughly studied displayed associated deficits of a perceptual and/or mnesic nature {but see De Renzi, 1986). On the other hand, the occurrence of associated deficits could help refine the identification of the functional locus of prosopagnosia in each individual case (e.g. Sergent and Poncet, 1990), but two factors have so far prevented this possibility. It cannot as yet be established if the non-facial impairments accompanying prosopagnosia result from a common basis to the processing of faces and the other deficient functions or reflect adjacent cortical territories respectively dedicated to different functions but conjointly damaged. Moreover, even though the nature of the prosopagnosic and associated deficits appears to suggest perceptual impairments in some cases and memory impairments in others (c/He"caen, 1981; Sergent and Villemure, 1989), the informative value of such observations has not yet been fully exploited. It thus seems that the knowledge and understanding of the anatomical substrates of face recognition cannot be entirely achieved on the sole basis of data from prosopagnosic patients. The purpose of the present study is to contribute to a clarification of the anatomical architecture of face recognition and object recognition in the normal brain, using positron emission tomography (PET) measures of regional cerebral blood flow (rCBF) in normal adults performing cognitive tasks of face and object recognition. Positron emission tomography studies have opened the way to a better understanding of neurobiological substrates of psychological functions and, with short-lived isotopes such as I5 O, have led to a localization of cerebral areas activated during the performance of cognitive tasks (e.g. Posner et al., 1988). Although this approach is confronted with numerous technical, methodological and theoretical problems, these will be dealt with in the final discussion, assuming that the PET technique provides sufficiently significant findings not to engage in a critical evaluation of its merits at this point. The approach adopted in the present study is based on the technique developed by Raichle, Fox and Posner (e.g. Petersen et al., 1988; Posner et al., 1988), with "O as a radioactive tracer and successive and complementary tasks allowing, through the subtraction method, the comparison of the localization and level of cerebral activation as a function of the specific processing demands inherent in the experimental tasks. Some characteristics of the present approach differ from earlier studies. It consists in the first PET study on face identification, comparing the recognition of faces with that of common objects. In an attempt to identify the cerebral activation specific to face recognition, this particular condition was compared with a condition requiring the processing of another property of faces than identity, namely the gender of faces. Given that gender categorization is typically preserved in prosopagnosic patients (e.g. Tranel et al., 1988; Sergent and Villemure, 1989), a comparison of gender discrimination and face recognition will provide an 'isolation' of the structures specifically involved in the process of identifying faces, rather than performing a comparison of this condition
18
J. SERGENT AND OTHERS
with a rest condition or a passive-looking condition (e.g. Posner et ai, 1988) which would result in the activation of cerebral areas not uniquely engaged in the processes of face recognition. In addition, one of the difficulties inherent in the interpretation of PET results lies in the localization of the foci of activation in the cerebral structures. In the present study, functional data from PET scans are superimposed on structural data from MRI scans obtained from each subject. This procedure (Evans et al., 1988) offers the opportunity to identify the cerebral areas activated during the performance of a given task and directly to compare the localization of foci of activation in the normal brain with the site of lesions in the brains of prosopagnosic patients. This study is thus concerned with clarifying the anatomical substrates of face and object processing, and it addresses a series of specific questions. On what cortical structures does the processing of faces rely? Does the processing of different properties of faces (e.g. gender and identity) engage different cortical structures? Does the processing of faces and objects rely on common anatomical areas? Do findings from a PET study on normal subjects concur with the anatomical and radiological evidence obtained from prosopagnosic patients? What is the respective role of the cerebral hemispheres in the processing of faces and of objects? What is the relation between hemisphere processing contribution as inferred in divided visual-field studies, and the actual involvement of the cerebral hemispheres as indicated by PET measures of cerebral activation?
Subjects Seven male adults, aged between 22 and 31 yrs, participated in the experiment. They were in good health, under no medication, and had no neurological or psychiatric disorders. They were right handed, as assessed with Bryden's (1982) questionnaire, with no left handers among their close relatives, and they had normal visual acuity and contrast sensitivity. They were fully informed of the risks associated with exposure to radioactive material, in accordance with the regulations of the Medical Research Council of Canada and the Control Board of Atomic Energy of Canada, and they read and signed consent forms for their participation. They were remunerated $100. Procedures and equipments The experiment comprised four phases, described in the order they took place. Preparatory phase. This phase consisted in introducing the subject to the experiment, explaining the risks, obtaining his formal consent, testing his handedness and vision, and preparing him to the experimental tasks in order to lessen potential anxiety and to reduce hesitations during the experiment proper. Each experimental task was explained and run, in conditions identical to those prevailing in the PET study, except that different stimuli were used. The subjects were also requested to fill in a questionnaire containing a list of 250 names of famous persons, and to indicate whether or not the name was familiar, and, if so, whether they could image a representation of the corresponding face and how clear this representation was. The faces used in one condition of the PET study were selected from this list for each subject, using those faces that subjects could represent clearly. PET experiment. The PET study took place the day following the preparatory phase and consisted in measuring rCBF in six different conditions run at 15 min intervals, with a different order of conditions for each subject. The stimuli were presented on a high-resolution Mitsubishi monitor, driven by a PS/2 IBM computer, in total darkness. The computer also controlled the duration of stimulus presentation (1 s), the interstimulus interval (3 s) and recorded the speed and accuracy of the subject's responses. The monitor was located above the subject's abdomen, at 85 cm from his eyes, and was oriented obliquely so that the screen was perpendicular to the subject's line of gaze. The stimuli appeared in the centre of the screen, and their size (in degrees of visual angle: 8° in height and 6° in width) and luminance were constant across conditions. In each condition, 40 stimuli were presented, except for a 'fixation' condition.
Downloaded from by guest on April 24, 2015
METHODS
PET STUDY OF FACE AND OBJECT PROCESSING
19
Downloaded from by guest on April 24, 2015
The subjects were in a supine position, the head firmly held in a customized frame containing rapidly hardening self-inflating foam-like material and located in the centre of the tomograph scanner. Ears were occluded with wax, and an intravenous catheter was placed into the left brachial vein for injection of H215O (40 mCi per injection) which served as CBF tracer. In each experimental condition, stimulus presentation started conjointly with the injection of the radioactive solution, and the recording of the emission of positrons began 15 s later and lasted for 60 s. Stimulus presentation went on for one more minute after the end of scanning in order to obtain additional data about the speed and accuracy of the subjects' responses. In four conditions, the subjects responded by pressing, with the index or middle finger of the right hand, one of two buttons of the computer 'mouse' which was placed on their abdomen. The study comprised three control conditions and three experimental conditions: (1) fixation: the subject looked at a fixation point located in the centre of the screen, and was required to concentrate on this point. The illuminated area of the screen was of the same size and average luminance as the faces and objects presented in the other conditions; (2) passive face-looking: the stimuli were black-and-white faces of individuals unfamiliar to the subjects, half of each gender, which the subject was requested to look at passively; (3) gratings: the stimuli were sine-wave gratings, with spatial frequency varying between 0.2 and 16 cycles per degree of visual angle, and contrast being held constant at 0.5. Half the gratings were oriented horizontally and the other half vertically, and the subject's task was to press one button for one orientation and the other button for the other orientation; (4) gender discrimination: the stimuli were blackand-white faces of individuals unfamiliar to the subjects, half of each gender, which the subject had to categorize as male or female by pressing one of the two buttons; (5) face identity: the stimuli were blackand-white faces of famous persons, drawn from the list presented to the subject the day before, and they were therefore all known by the subjects. Half the faces depicted actors and the other half depicted persons from other professional categories (politicians, sportsmen, newsmen, singers). The subject's task was to decide whether or not the face was that of an actor, and to press one of the two buttons to indicate the response; (6) object recognition: the stimuli were black-and-white photographs of common objects, half of which were 'living' or natural (cat, tree, fish) and the other half 'non-living' or man-made (car, fork, table). The subject had to press one button for one category and the other button for the other category. It must be noted that the face- and object-recognition tasks are, strictly speaking, categorization tasks, as the specific name or identity of the object or the face does not have to be accessed. However, such a categorization cannot be performed unless the stimulus is actually recognized (e.g. Sergent, 1985). In all conditions, and particularly in conditions 5 and 6, the subject was specifically instructed not to name the faces and objects, neither explicitly or implicitly, nor to attempt to retrieve their names. However, after the preparatory phase, the PET experiment, and the subsequent divided visual-field study, all the subjects commented that they had been unable to refrain from thinking of the names of the famous faces and of the objects. Divided visual-field study. This phase took place between 2 and 4 wks after the PET study. It consisted in a divided visual-field experiment using the same conditions and stimuli as in the PET study, without PET scanning and with two main differences. Only the conditions 3 to 6 were tested (i.e. gratings, gender discrimination, face identity, and object recognition). The stimuli were presented twice, once in each visual field at an eccentricity of 5° of visual angle, resulting in the presentation of 80 stimuli in each condition. To ensure as much similarity of stimulus presentation in this and the PET experiments, the same monitor, computer, stimulus duration and interstimulus interval were used in the two studies. Because of the lateral presentation, the size of the stimuli was reduced by half; however, to keep the actual visual angle of the stimuli the same in the two studies, the subjects viewed the stimuli from half the distance used in the PET experiment (43 cm). In addition, the stimuli were presented for 1 s as in the PET study. This made it necessary to control for eye fixation in the centre of the screen in order to ensure that the laterally presented stimuli would effectively be projected to the contralateral hemisphere. This was achieved on-line with a Gulf and Western (Model 1994B) Eye-View-Monitor system using comeal reflection to determine the locus of fixation. Before the experiment, central fixation was established for each subject, and the digital coordinates of central eye position were entered in the computer. During the experiment, eye position was sampled every 16.7 ms. Stimulus presentation required five consecutive samples of eye position to be within 0.5° of visual angle of the central fixation coordinates; any departure of more than 0.5° would delay stimulus presentation and, during the 1 s of exposure, would interrupt the presentation of the stimulus, in which case the trial was discarded. This procedure guaranteed that initial stimulation was projected to the contralateral hemisphere. Subjects were very accurate in maintaining central fixation, and less than 1 % of the trials had to be discarded because of eye deviation from the centre.
20
J. SERGENT AND OTHERS
MR1 study. This phase took place immediately after the lateraJity study. It consisted in a MRI scan of each subject's brain, with a Philips Gyroscan (1.5 Telsa), using the same head-holder and the same orientation of the head as for the PET scan.
RESULTS
Two main sets of data were collected in this investigation, one in the PET study, and the other in the divided visual-field study, and they will be discussed in turn. PET study The PET study comprised three control and three experimental conditions. However, the passive face-looking condition turned out to be inadequate as a control and, compared with the grating condition, resulted in no significant change in activation. This can be attributed to the variety of information conveyed by faces, which implies that faces lend themselves to a diversity of cognitive operations, and the subjects thus had the possibility to consider different properties of the faces during this condition. This suggests that a passive-looking condition, often used in PET studies (e.g. Posner et al., 1988), may not always serve as an adequate control given the impossibility to determine the cognitive operations actually performed by the subjects in the absence of specific instructions. The results of the other four conditions (gratings and the three experimental conditions) proved reliable as response accuracy was controlled and subjects did perform the requested tasks. In each condition, each subject's response accuracy was equal or superior to 95%, except in the face-identity condition in which accuracy dropped to 85% in two subjects.
Downloaded from by guest on April 24, 2015
Scanning and analyses PET scanning was performed by measuring rCBF using the intravenous l5oxygen water bolus technique (Raichle et al., 1983) and a Scanditronix PC-2048 PET scanner with in-plane and spatial resolutions of 5 — 6 mm full width at half maximum. Fifteen slices with a centre-to-centre distance of 6.8 mm were simultaneously imaged, which could be further divided by bilinear interpolation to produce 80 transaxial slices. A rotating ^germanium source was used to correct the images for attenuation of the gamma rays in the skull and brain tissue. The image planes were chosen parallel to the subject's glabella-inion line which was aligned with the tomograph's laser reference line and transcribed onto the head-holder as a marker for the correlation of PET and MR imaging. The MR images were obtained at the same 15 planes as in the PET study. Slice registration was ensured using a thin CuSO4-filled (5 mm) catheter attached to the side of the customized head-holder, and visible in an MR image, coincident with the reference line identifying the PET scan reference plane (Evans et al., 1989). The PET activation functional images were mapped onto the MR structural images using a PIXAR threedimensional computer and landmark-matching software (Evans et al., 1989). Interactive three-dimensional image software was used to establish an orthogonal coordinate frame based on the anterior commissureposterior commissure (AC-PC) line as identified in the MR image volume. These coordinates were used to apply a linear resampling of each matched pair of MRI and PETdatasets into a standardized stereotactic coordinate system (TaJairach and Tournoux, 1988). PET images were then normalized for global CBF and the difference between control and experimental conditions determined for each subject. The mean state-dependent change image volume was obtained by averaging across subjects (Fox et al., 1985) and then converted to a r-statistic by dividing the mean state-dependent change by the mean standard deviation in normalized CBF for all intra-cerebral voxels and by multiplying this quotient by the square root of N (number of subjects). Anatomical and functional images were merged to allow direct localization on the MR images of /-statistic peaks identified by an automatic peak-detection algorithm (Mintun et al., 1989) and for the anatomical correlation of extended zones of activation not expressible in terms of isolated peaks. The peak distribution was then searched for significant signals using change-distribution analysis (Fox et al., 1988; Fox, 1991) and z-score thresholding. Peaks with significance levels of P < 0.05 are reported.
PET STUDY OF FACE AND OBJECT PROCESSING
21
The grating condition served as control for the experimental conditions, and the experimental conditions on face processing were compared with one another. The significant foci of activation are presented in Table 1, expressed in terms of Talairach and Tournoux's (1988) stereotactic coordinates, of their location in Brodmann's classification and of the common name of the activated area. The most significant foci of activation are shown in Fig. 1, superimposed on a MR image of the brain. Cerebral TABLE I. LOCALIZATION OF FOCI OF ACTIVATION WITH THE SUBTRACTION METHOD. EXPRESSED IN TERMS OF STEREOTACTIC COORDINATES. BRODMANN'S CLASSIFICATION AND USUAL NAME OF CORTICAL AREA
1
Cortical area
Level ofj . significance
Striate cortex Striate cortex Striate cortex Striate cortex Lingual gyrus Lingual gyrus Inferior occipital gyrus Parieto-occipito sulcus Parieto-occipito sulcus Primary sensory area Secondary sensory area Motor area Premotor area
*•
Cuneus Inferior occipital gyrus Inferior occipital-temporal gyrus Lateral occipital gyrus Middle occipital gyrus Lingual gyrus
*•
Medial anterior temporal gyrus Medial anterior temporal gyrus Temporal pole Temporal pole Parahippocampal gyrus Gyrus rectus Fusiform gyrus Fusiform gyrus Middle temporal gyrus Lingual gyrus Inferior temporal gyrus Fusiform gyrus Middle temporal gyrus Middle occipital gyrus Superior parietaJ lobe SupramarginaJ gyrus Supramarginal gyrus Gyrus rectus
***P < 0.001; **P < 0.01; *P < 0.05.
••• »** *•* *•* • «» * • •
*•• ** • •• **
*
• ** ** **
•*» ** *
• • ** • »
* *
*
Downloaded from by guest on April 24, 2015
Coordinates X Y Brodmann s area Z Gratings -Fixation 3 -82 3 Right 17 8 -87 9 Right 17 -12 -77 8 Left 17 -7 -85 Left 17 6 4 -81 Right 18 2 12 -79 2 Right 18 -27 -98 -6 Left 18 15 -82 30 Right 19 -19 -81 32 Left 19 -41 -28 59 Left 3 -44 -35 60 Left 1 -41 -18 62 Left 4 2 -33 60 Left 6 Gender discrimination—gratings Right 19 19 -92 18 40 -77 -2 Right 18 43 -73 -2 Right 19 29 -81 -8 Right 19 -32 -73 -3 Left 19 20 -81 -6 Right 18 Face identity—gender discrimination -23 -3 -32 Left 36 25 -5 -32 Right 36 37 20 -32 Right 38 -36 9 -27 Left 38 24 -15 -20 Right 36 -3 25 -17 Centre 11 -37 -60 -12 Left 37 37 -55 -11 Right 37 -52 -9 -9 Left 21 21 -60 2 Right 18 Object—gratings -55 Left -39 -17 20 -37 -58 -14 Left 37 -53 -9 -11 Left 21 -40 -76 -6 Left 19 -32 -60 60 Left 7 52 -23 32 Right 40 -53 -18 18 Left 40 -2 21 -19 Centre II
22
J. SERGENT AND OTHERS
Downloaded from by guest on April 24, 2015
localization is described along the three Euclidean axes: X corresponds to the lateral axis, 0 being the midline between two hemispheres and negative values indicating the left hemisphere; Y corresponds to the antero-posterior axis, 0 being at the level of the anterior commissure and negative values indicating the posterior regions; Z corresponds to the dorsoventral axis, 0 being at the level of the AC-PC line and negative values indicating areas below this line. Gratings minus fixation. The perception of the gratings, and their categorization as a function of their orientation signalled by a right-hand response, resulted in intense activation of the primary visual cortex and the left pre- and post-rolandic areas corresponding to the cortical representation of the right hand (Fig. 1). The main foci of activation were located in the striate cortex of both hemispheres, in the posterior part of the right lingual gyrus (area 18) and in the inferior lateral occipital area of the left hemisphere (area 18). All these cortical regions were equally activated in the three experimental conditions, and they reflect elementary perceptual operations, irrespective of the complexity and signification of the visual information being processed. The only cortical area activated by the presentation of the gratings and not by other visual stimuli was located in the occipito-parietal sulcus of both hemispheres (area 19, see Table 1), and this activation can therefore be attributed to the spatial discrimination inherent in the categorization of the gratings as a function of their orientation. A second focus of activation was specific to the right manual response and is presented in the central left part of Fig. 1. It is located in the pre- and post-central superior region of the left hemisphere encompassing the primary and secondary sensorimotor areas, and it was also present in the three experimental conditions that too required a right manual response. Gender categorization minus gratings. Gender categorization was tested with faces unfamiliar to the subjects, and this task does not require the identification of the faces nor does it call for the processing of faces as unique entities. This condition resulted in further activation of extrastriate visual areas 18 and 19, and the involved cerebral structures were more anterior and more ventral than those activated during the categorization of the gratings. However, the gender discrimination task also engaged the right cuneus as well as the lateral occipital area of the left hemisphere (area 19, see Table 1). Face identity minus gender categorization. All the cortical areas specifically activated during the face-identity task were located more anteriorly than those engaged in the face-gender categorization task, as indicated by the location of these areas on the anteroposterior Y axis (see Table 1). In addition, both hemispheres participated in the operations underlying face identification, but asymmetrically so. In the occipital and posterior temporal cortices, therightlingual gyrus (area 18), the fusiform gyrus of both hemispheres (areas 19 and 37), and the middle gyrus (area 21) of the left temporal cortex were significantly activated during the face-identity task. The highest level of activation was found in the parahippocampal gyrus (area 36) of the right hemisphere, and no activation of the homotopic area of the left hemisphere was observed (see Fig. 1, at level —20 on the Z axis). Also activated during the face-identification task were the most anterior regions of the temporal cortex, including the temporal pole of both hemispheres and the adjacent medial areas, as illustrated in Fig. 1. All these foci of activation are localized in the ventral parts of the cortex, as indicated by their values on the dorsoventral plane
FIG 1. Foci of activation superimposed on a magnetic resonance image of the brain (the numbers on the bottom left of Ihe images correspond to the slice level on the Z axis). Gratings: gratings minus fixation (a) activation at the level of the occipital cortex; (b) activation at the level of the left sensorimotor cortex. Face gender: face gender minus fixation; slices are 3 mm apart, starting at Z = - 8 . Face identity: (a) face identity minus face gender; (b) face identity minus fixation; slices are 3 mm apart, starting at Z = —23. Object: object-recognition minus gratings (a) absence of activation of the anterior temporal cortex, (b) activation of the left occipito-temporal cortex and the left temporal area 21.
Face Oender
Face Identity
Downloaded from by guest on April 24, 2015
z o
GO
oo
n m
O
•v
-i
m n
a
a o
2
m
O
•n
o
•
FUNCTIONAL NEUROANATOMY OF FACE AND OBJECT PROCESSING A POSITRON EMISSION TOMOGRAPHY STUDY by JUSTINE SERGENT, SHINSUKE OHTA and BRENNAN MACDONALD (From the Montreal Neurological Institute, McGill University, Montreal, Canada)
SUMMARY
INTRODUCTION
The human face holds a special place among visual objects. Any social animal must possess the capacity to differentiate and recognize members of its group and, in humans, the face is the most distinctive attribute for indexing identity reliably. For the face to assume such an indexing role, the processing organism must be endowed with structures and mechanisms adapted to performing the operations required by this function, and this implies a cognitive and structural architecture different from that required in the processing of other objects that are not called to play such a role. From a cognitive standpoint, there are several ways in which the processing of faces differs from that of other objects. For instance, most objects are processed at the basic category level (Rosch et al., 1976) regardless of the particularities of each object, whereas each instance of faces is treated as different within the category of faces and the differentiation of each face allows access to related stored biographical information. Correspondence to: Justine Sergent, Cognitive Neuroscience Laboratory, Montreal Neurological Institute, 3801 University Street, Montreal (Quebec) H3A 2B4, Canada. © Oxford University Press 1992
Downloaded from by guest on April 24, 2015
Studies of brain-damaged patients have revealed the existence of a selective impairment of face processing, prosopagnosia, resulting from lesions at different loci in the occipital and temporal lobes. The results of such studies have led to the identification of several cortical areas underlying the processing of faces, but it remains unclear what functional aspects of face processing are served by these areas and whether they are uniquely devoted to the processing of faces. The present study addresses these questions in a positron emission tomography (PET) study of regional cerebral blood flow in normal adults, using the "oxygen water bolus technique. The subjects participated in six tasks (with gratings, faces and objects), and the resulting level of cerebral activation was mapped on images of the subjects' cerebral structures obtained through magnetic resonance and was compared between tasks using the subtraction method. Compared with a fixation condition, regional cerebral blood flow (rCBF) changes were found in the striate and extrastriate cortex when subjects had to decide on the orientation of sine-wave gratings. A face-gender categorization resulted in activation changes in the right extrastriate cortex, and a face-identity condition produced additional activation of the fusiform gyrus and anterior temporal cortex of both hemispheres, and of the right parahippocampal gyrus and adjacent areas. Cerebral activation during an object-recognition task occurred essentially in the left occipito-temporal cortex and did not involve the right hemisphere regions specifically activated during the face-identity task. The results provide the first empirical evidence from normal subjects regarding the crucial role of the ventro-medial region of the right hemisphere in face recognition, and they offer new information about the dissociation between face and object processing.
16
J. SERGENT AND OTHERS
Downloaded from by guest on April 24, 2015
In addition, a large diversity of information about an individual can be derived on the sole basis of the physical attributes of a face, irrespective of its identity, and inferences about age, gender, race, emotion, the charm of the face, and so on, can be made that provide valuable information relevant to the bearer of the face. Another way faces differ from other objects, therefore, seems to be the variety of categories into which faces can be classified, with the same facial attributes being processed and combined differently depending on the information one wishes to access about an individual (Sergent, 1989). In contrast, for practical purposes, most objects belong to a single category and do not lend themselves to multiple categorizations. What makes face recognition a complex process is the large number of instances with which we are confronted, which requires the extraction, from a general configuration common to all faces, of the particularities that make each face unique. This implies refined perceptual mechanisms capable of detecting subtle differences among faces and of achieving a structural representation that uniquely defines each face (Sergent, 1989). Each representation must then be stored reliably, and this imposes added constraints on the processing organism given the many faces which we have to remember. Moreover, for a perceived face to be recognized as familiar, further operations must be implemented to make contact between perceived and stored facial information, and these operations are concerned with extracting the physiognomic invariants embedded in a given facial representation that may take diverse appearances resulting from the variety of possible formats, poses, angles of view, expressions, distance or illumination. Additional processes must then be performed for a face to be identified, and these involve the reactivation of episodic and semantic (i.e. biographic) information related to that face and without which it would remain that of a stranger. In spite of its complexity, face recognition is an automatic, effortless and quasi-infallible process, and this attests to the high level of efficiency of the cerebral structures sustaining this function. A detailed specification of the anatomical architecture of face recognition has been difficult to achieve, however, and the only source of information bearing on this issue has so far come from the study of prosopagnosic patients. The correlation between, on the one hand, the behavioural impairments related to the inability to process faces and to associated deficits and, on the other hand, the location of the lesions, has offered the opportunity to identify cortical areas the damage of which produced prosopagnosia (Meadows, 1974; Damasio etal., 1982, 1990a). Yet the reliance on data from neurological patients to infer the structural and functional organization of the cerebral structures normally underlying face recognition is confronted with a series of theoretical and methodological difficulties. First, enquiring about the location of the lesions responsible for prosopagnosia and asking which cortical areas underlie the normal processes inherent in face recognition are distinct questions that may not have the same answers. For instance, a focal damage may have detrimental effects on the functions of distant intact areas and, within a damaged brain, local structural integrity does not guarantee normal functioning (Monakow, von, 1910; Sergent and Signoret, 1992). Secondly, there is such a large variability among patients in the aetiology and location of the damage associated with the occurrence of prosopagnosia that it has been difficult to achieve a comprehensive view of the biological architecture of face recognition. Thus, even though the ventro-medial occipito-temporal junction of the right hemisphere has been considered a critical area in the processing
PET STUDY OF FACE AND OBJECT PROCESSING
17
Downloaded from by guest on April 24, 2015
of faces, cases have been reported in whom the lesion did not invade this particular region (e.g. Damasio et al., 1982; Bruyer et al, 1983; Sergent and Poncet, 1990). Similarly, although it is widely held that bilateral damage underlies the occurrence of prosopagnosia (e.g. Meadows, 1974; Damasio, 1985), cases have been reported whose Magnetic Resonance Imaging (MRI) scans showed no sign of left-hemisphere involvement {see Michel et al., 1989, for a review). Thirdly, the problem is further complicated by the fact that prosopagnosia, albeit an essentially selective impairment, has so far never occurred in a pure form. Every case of prosopagnosia that has been thoroughly studied displayed associated deficits of a perceptual and/or mnesic nature {but see De Renzi, 1986). On the other hand, the occurrence of associated deficits could help refine the identification of the functional locus of prosopagnosia in each individual case (e.g. Sergent and Poncet, 1990), but two factors have so far prevented this possibility. It cannot as yet be established if the non-facial impairments accompanying prosopagnosia result from a common basis to the processing of faces and the other deficient functions or reflect adjacent cortical territories respectively dedicated to different functions but conjointly damaged. Moreover, even though the nature of the prosopagnosic and associated deficits appears to suggest perceptual impairments in some cases and memory impairments in others (c/He"caen, 1981; Sergent and Villemure, 1989), the informative value of such observations has not yet been fully exploited. It thus seems that the knowledge and understanding of the anatomical substrates of face recognition cannot be entirely achieved on the sole basis of data from prosopagnosic patients. The purpose of the present study is to contribute to a clarification of the anatomical architecture of face recognition and object recognition in the normal brain, using positron emission tomography (PET) measures of regional cerebral blood flow (rCBF) in normal adults performing cognitive tasks of face and object recognition. Positron emission tomography studies have opened the way to a better understanding of neurobiological substrates of psychological functions and, with short-lived isotopes such as I5 O, have led to a localization of cerebral areas activated during the performance of cognitive tasks (e.g. Posner et al., 1988). Although this approach is confronted with numerous technical, methodological and theoretical problems, these will be dealt with in the final discussion, assuming that the PET technique provides sufficiently significant findings not to engage in a critical evaluation of its merits at this point. The approach adopted in the present study is based on the technique developed by Raichle, Fox and Posner (e.g. Petersen et al., 1988; Posner et al., 1988), with "O as a radioactive tracer and successive and complementary tasks allowing, through the subtraction method, the comparison of the localization and level of cerebral activation as a function of the specific processing demands inherent in the experimental tasks. Some characteristics of the present approach differ from earlier studies. It consists in the first PET study on face identification, comparing the recognition of faces with that of common objects. In an attempt to identify the cerebral activation specific to face recognition, this particular condition was compared with a condition requiring the processing of another property of faces than identity, namely the gender of faces. Given that gender categorization is typically preserved in prosopagnosic patients (e.g. Tranel et al., 1988; Sergent and Villemure, 1989), a comparison of gender discrimination and face recognition will provide an 'isolation' of the structures specifically involved in the process of identifying faces, rather than performing a comparison of this condition
18
J. SERGENT AND OTHERS
with a rest condition or a passive-looking condition (e.g. Posner et ai, 1988) which would result in the activation of cerebral areas not uniquely engaged in the processes of face recognition. In addition, one of the difficulties inherent in the interpretation of PET results lies in the localization of the foci of activation in the cerebral structures. In the present study, functional data from PET scans are superimposed on structural data from MRI scans obtained from each subject. This procedure (Evans et al., 1988) offers the opportunity to identify the cerebral areas activated during the performance of a given task and directly to compare the localization of foci of activation in the normal brain with the site of lesions in the brains of prosopagnosic patients. This study is thus concerned with clarifying the anatomical substrates of face and object processing, and it addresses a series of specific questions. On what cortical structures does the processing of faces rely? Does the processing of different properties of faces (e.g. gender and identity) engage different cortical structures? Does the processing of faces and objects rely on common anatomical areas? Do findings from a PET study on normal subjects concur with the anatomical and radiological evidence obtained from prosopagnosic patients? What is the respective role of the cerebral hemispheres in the processing of faces and of objects? What is the relation between hemisphere processing contribution as inferred in divided visual-field studies, and the actual involvement of the cerebral hemispheres as indicated by PET measures of cerebral activation?
Subjects Seven male adults, aged between 22 and 31 yrs, participated in the experiment. They were in good health, under no medication, and had no neurological or psychiatric disorders. They were right handed, as assessed with Bryden's (1982) questionnaire, with no left handers among their close relatives, and they had normal visual acuity and contrast sensitivity. They were fully informed of the risks associated with exposure to radioactive material, in accordance with the regulations of the Medical Research Council of Canada and the Control Board of Atomic Energy of Canada, and they read and signed consent forms for their participation. They were remunerated $100. Procedures and equipments The experiment comprised four phases, described in the order they took place. Preparatory phase. This phase consisted in introducing the subject to the experiment, explaining the risks, obtaining his formal consent, testing his handedness and vision, and preparing him to the experimental tasks in order to lessen potential anxiety and to reduce hesitations during the experiment proper. Each experimental task was explained and run, in conditions identical to those prevailing in the PET study, except that different stimuli were used. The subjects were also requested to fill in a questionnaire containing a list of 250 names of famous persons, and to indicate whether or not the name was familiar, and, if so, whether they could image a representation of the corresponding face and how clear this representation was. The faces used in one condition of the PET study were selected from this list for each subject, using those faces that subjects could represent clearly. PET experiment. The PET study took place the day following the preparatory phase and consisted in measuring rCBF in six different conditions run at 15 min intervals, with a different order of conditions for each subject. The stimuli were presented on a high-resolution Mitsubishi monitor, driven by a PS/2 IBM computer, in total darkness. The computer also controlled the duration of stimulus presentation (1 s), the interstimulus interval (3 s) and recorded the speed and accuracy of the subject's responses. The monitor was located above the subject's abdomen, at 85 cm from his eyes, and was oriented obliquely so that the screen was perpendicular to the subject's line of gaze. The stimuli appeared in the centre of the screen, and their size (in degrees of visual angle: 8° in height and 6° in width) and luminance were constant across conditions. In each condition, 40 stimuli were presented, except for a 'fixation' condition.
Downloaded from by guest on April 24, 2015
METHODS
PET STUDY OF FACE AND OBJECT PROCESSING
19
Downloaded from by guest on April 24, 2015
The subjects were in a supine position, the head firmly held in a customized frame containing rapidly hardening self-inflating foam-like material and located in the centre of the tomograph scanner. Ears were occluded with wax, and an intravenous catheter was placed into the left brachial vein for injection of H215O (40 mCi per injection) which served as CBF tracer. In each experimental condition, stimulus presentation started conjointly with the injection of the radioactive solution, and the recording of the emission of positrons began 15 s later and lasted for 60 s. Stimulus presentation went on for one more minute after the end of scanning in order to obtain additional data about the speed and accuracy of the subjects' responses. In four conditions, the subjects responded by pressing, with the index or middle finger of the right hand, one of two buttons of the computer 'mouse' which was placed on their abdomen. The study comprised three control conditions and three experimental conditions: (1) fixation: the subject looked at a fixation point located in the centre of the screen, and was required to concentrate on this point. The illuminated area of the screen was of the same size and average luminance as the faces and objects presented in the other conditions; (2) passive face-looking: the stimuli were black-and-white faces of individuals unfamiliar to the subjects, half of each gender, which the subject was requested to look at passively; (3) gratings: the stimuli were sine-wave gratings, with spatial frequency varying between 0.2 and 16 cycles per degree of visual angle, and contrast being held constant at 0.5. Half the gratings were oriented horizontally and the other half vertically, and the subject's task was to press one button for one orientation and the other button for the other orientation; (4) gender discrimination: the stimuli were blackand-white faces of individuals unfamiliar to the subjects, half of each gender, which the subject had to categorize as male or female by pressing one of the two buttons; (5) face identity: the stimuli were blackand-white faces of famous persons, drawn from the list presented to the subject the day before, and they were therefore all known by the subjects. Half the faces depicted actors and the other half depicted persons from other professional categories (politicians, sportsmen, newsmen, singers). The subject's task was to decide whether or not the face was that of an actor, and to press one of the two buttons to indicate the response; (6) object recognition: the stimuli were black-and-white photographs of common objects, half of which were 'living' or natural (cat, tree, fish) and the other half 'non-living' or man-made (car, fork, table). The subject had to press one button for one category and the other button for the other category. It must be noted that the face- and object-recognition tasks are, strictly speaking, categorization tasks, as the specific name or identity of the object or the face does not have to be accessed. However, such a categorization cannot be performed unless the stimulus is actually recognized (e.g. Sergent, 1985). In all conditions, and particularly in conditions 5 and 6, the subject was specifically instructed not to name the faces and objects, neither explicitly or implicitly, nor to attempt to retrieve their names. However, after the preparatory phase, the PET experiment, and the subsequent divided visual-field study, all the subjects commented that they had been unable to refrain from thinking of the names of the famous faces and of the objects. Divided visual-field study. This phase took place between 2 and 4 wks after the PET study. It consisted in a divided visual-field experiment using the same conditions and stimuli as in the PET study, without PET scanning and with two main differences. Only the conditions 3 to 6 were tested (i.e. gratings, gender discrimination, face identity, and object recognition). The stimuli were presented twice, once in each visual field at an eccentricity of 5° of visual angle, resulting in the presentation of 80 stimuli in each condition. To ensure as much similarity of stimulus presentation in this and the PET experiments, the same monitor, computer, stimulus duration and interstimulus interval were used in the two studies. Because of the lateral presentation, the size of the stimuli was reduced by half; however, to keep the actual visual angle of the stimuli the same in the two studies, the subjects viewed the stimuli from half the distance used in the PET experiment (43 cm). In addition, the stimuli were presented for 1 s as in the PET study. This made it necessary to control for eye fixation in the centre of the screen in order to ensure that the laterally presented stimuli would effectively be projected to the contralateral hemisphere. This was achieved on-line with a Gulf and Western (Model 1994B) Eye-View-Monitor system using comeal reflection to determine the locus of fixation. Before the experiment, central fixation was established for each subject, and the digital coordinates of central eye position were entered in the computer. During the experiment, eye position was sampled every 16.7 ms. Stimulus presentation required five consecutive samples of eye position to be within 0.5° of visual angle of the central fixation coordinates; any departure of more than 0.5° would delay stimulus presentation and, during the 1 s of exposure, would interrupt the presentation of the stimulus, in which case the trial was discarded. This procedure guaranteed that initial stimulation was projected to the contralateral hemisphere. Subjects were very accurate in maintaining central fixation, and less than 1 % of the trials had to be discarded because of eye deviation from the centre.
20
J. SERGENT AND OTHERS
MR1 study. This phase took place immediately after the lateraJity study. It consisted in a MRI scan of each subject's brain, with a Philips Gyroscan (1.5 Telsa), using the same head-holder and the same orientation of the head as for the PET scan.
RESULTS
Two main sets of data were collected in this investigation, one in the PET study, and the other in the divided visual-field study, and they will be discussed in turn. PET study The PET study comprised three control and three experimental conditions. However, the passive face-looking condition turned out to be inadequate as a control and, compared with the grating condition, resulted in no significant change in activation. This can be attributed to the variety of information conveyed by faces, which implies that faces lend themselves to a diversity of cognitive operations, and the subjects thus had the possibility to consider different properties of the faces during this condition. This suggests that a passive-looking condition, often used in PET studies (e.g. Posner et al., 1988), may not always serve as an adequate control given the impossibility to determine the cognitive operations actually performed by the subjects in the absence of specific instructions. The results of the other four conditions (gratings and the three experimental conditions) proved reliable as response accuracy was controlled and subjects did perform the requested tasks. In each condition, each subject's response accuracy was equal or superior to 95%, except in the face-identity condition in which accuracy dropped to 85% in two subjects.
Downloaded from by guest on April 24, 2015
Scanning and analyses PET scanning was performed by measuring rCBF using the intravenous l5oxygen water bolus technique (Raichle et al., 1983) and a Scanditronix PC-2048 PET scanner with in-plane and spatial resolutions of 5 — 6 mm full width at half maximum. Fifteen slices with a centre-to-centre distance of 6.8 mm were simultaneously imaged, which could be further divided by bilinear interpolation to produce 80 transaxial slices. A rotating ^germanium source was used to correct the images for attenuation of the gamma rays in the skull and brain tissue. The image planes were chosen parallel to the subject's glabella-inion line which was aligned with the tomograph's laser reference line and transcribed onto the head-holder as a marker for the correlation of PET and MR imaging. The MR images were obtained at the same 15 planes as in the PET study. Slice registration was ensured using a thin CuSO4-filled (5 mm) catheter attached to the side of the customized head-holder, and visible in an MR image, coincident with the reference line identifying the PET scan reference plane (Evans et al., 1989). The PET activation functional images were mapped onto the MR structural images using a PIXAR threedimensional computer and landmark-matching software (Evans et al., 1989). Interactive three-dimensional image software was used to establish an orthogonal coordinate frame based on the anterior commissureposterior commissure (AC-PC) line as identified in the MR image volume. These coordinates were used to apply a linear resampling of each matched pair of MRI and PETdatasets into a standardized stereotactic coordinate system (TaJairach and Tournoux, 1988). PET images were then normalized for global CBF and the difference between control and experimental conditions determined for each subject. The mean state-dependent change image volume was obtained by averaging across subjects (Fox et al., 1985) and then converted to a r-statistic by dividing the mean state-dependent change by the mean standard deviation in normalized CBF for all intra-cerebral voxels and by multiplying this quotient by the square root of N (number of subjects). Anatomical and functional images were merged to allow direct localization on the MR images of /-statistic peaks identified by an automatic peak-detection algorithm (Mintun et al., 1989) and for the anatomical correlation of extended zones of activation not expressible in terms of isolated peaks. The peak distribution was then searched for significant signals using change-distribution analysis (Fox et al., 1988; Fox, 1991) and z-score thresholding. Peaks with significance levels of P < 0.05 are reported.
PET STUDY OF FACE AND OBJECT PROCESSING
21
The grating condition served as control for the experimental conditions, and the experimental conditions on face processing were compared with one another. The significant foci of activation are presented in Table 1, expressed in terms of Talairach and Tournoux's (1988) stereotactic coordinates, of their location in Brodmann's classification and of the common name of the activated area. The most significant foci of activation are shown in Fig. 1, superimposed on a MR image of the brain. Cerebral TABLE I. LOCALIZATION OF FOCI OF ACTIVATION WITH THE SUBTRACTION METHOD. EXPRESSED IN TERMS OF STEREOTACTIC COORDINATES. BRODMANN'S CLASSIFICATION AND USUAL NAME OF CORTICAL AREA
1
Cortical area
Level ofj . significance
Striate cortex Striate cortex Striate cortex Striate cortex Lingual gyrus Lingual gyrus Inferior occipital gyrus Parieto-occipito sulcus Parieto-occipito sulcus Primary sensory area Secondary sensory area Motor area Premotor area
*•
Cuneus Inferior occipital gyrus Inferior occipital-temporal gyrus Lateral occipital gyrus Middle occipital gyrus Lingual gyrus
*•
Medial anterior temporal gyrus Medial anterior temporal gyrus Temporal pole Temporal pole Parahippocampal gyrus Gyrus rectus Fusiform gyrus Fusiform gyrus Middle temporal gyrus Lingual gyrus Inferior temporal gyrus Fusiform gyrus Middle temporal gyrus Middle occipital gyrus Superior parietaJ lobe SupramarginaJ gyrus Supramarginal gyrus Gyrus rectus
***P < 0.001; **P < 0.01; *P < 0.05.
••• »** *•* *•* • «» * • •
*•• ** • •• **
*
• ** ** **
•*» ** *
• • ** • »
* *
*
Downloaded from by guest on April 24, 2015
Coordinates X Y Brodmann s area Z Gratings -Fixation 3 -82 3 Right 17 8 -87 9 Right 17 -12 -77 8 Left 17 -7 -85 Left 17 6 4 -81 Right 18 2 12 -79 2 Right 18 -27 -98 -6 Left 18 15 -82 30 Right 19 -19 -81 32 Left 19 -41 -28 59 Left 3 -44 -35 60 Left 1 -41 -18 62 Left 4 2 -33 60 Left 6 Gender discrimination—gratings Right 19 19 -92 18 40 -77 -2 Right 18 43 -73 -2 Right 19 29 -81 -8 Right 19 -32 -73 -3 Left 19 20 -81 -6 Right 18 Face identity—gender discrimination -23 -3 -32 Left 36 25 -5 -32 Right 36 37 20 -32 Right 38 -36 9 -27 Left 38 24 -15 -20 Right 36 -3 25 -17 Centre 11 -37 -60 -12 Left 37 37 -55 -11 Right 37 -52 -9 -9 Left 21 21 -60 2 Right 18 Object—gratings -55 Left -39 -17 20 -37 -58 -14 Left 37 -53 -9 -11 Left 21 -40 -76 -6 Left 19 -32 -60 60 Left 7 52 -23 32 Right 40 -53 -18 18 Left 40 -2 21 -19 Centre II
22
J. SERGENT AND OTHERS
Downloaded from by guest on April 24, 2015
localization is described along the three Euclidean axes: X corresponds to the lateral axis, 0 being the midline between two hemispheres and negative values indicating the left hemisphere; Y corresponds to the antero-posterior axis, 0 being at the level of the anterior commissure and negative values indicating the posterior regions; Z corresponds to the dorsoventral axis, 0 being at the level of the AC-PC line and negative values indicating areas below this line. Gratings minus fixation. The perception of the gratings, and their categorization as a function of their orientation signalled by a right-hand response, resulted in intense activation of the primary visual cortex and the left pre- and post-rolandic areas corresponding to the cortical representation of the right hand (Fig. 1). The main foci of activation were located in the striate cortex of both hemispheres, in the posterior part of the right lingual gyrus (area 18) and in the inferior lateral occipital area of the left hemisphere (area 18). All these cortical regions were equally activated in the three experimental conditions, and they reflect elementary perceptual operations, irrespective of the complexity and signification of the visual information being processed. The only cortical area activated by the presentation of the gratings and not by other visual stimuli was located in the occipito-parietal sulcus of both hemispheres (area 19, see Table 1), and this activation can therefore be attributed to the spatial discrimination inherent in the categorization of the gratings as a function of their orientation. A second focus of activation was specific to the right manual response and is presented in the central left part of Fig. 1. It is located in the pre- and post-central superior region of the left hemisphere encompassing the primary and secondary sensorimotor areas, and it was also present in the three experimental conditions that too required a right manual response. Gender categorization minus gratings. Gender categorization was tested with faces unfamiliar to the subjects, and this task does not require the identification of the faces nor does it call for the processing of faces as unique entities. This condition resulted in further activation of extrastriate visual areas 18 and 19, and the involved cerebral structures were more anterior and more ventral than those activated during the categorization of the gratings. However, the gender discrimination task also engaged the right cuneus as well as the lateral occipital area of the left hemisphere (area 19, see Table 1). Face identity minus gender categorization. All the cortical areas specifically activated during the face-identity task were located more anteriorly than those engaged in the face-gender categorization task, as indicated by the location of these areas on the anteroposterior Y axis (see Table 1). In addition, both hemispheres participated in the operations underlying face identification, but asymmetrically so. In the occipital and posterior temporal cortices, therightlingual gyrus (area 18), the fusiform gyrus of both hemispheres (areas 19 and 37), and the middle gyrus (area 21) of the left temporal cortex were significantly activated during the face-identity task. The highest level of activation was found in the parahippocampal gyrus (area 36) of the right hemisphere, and no activation of the homotopic area of the left hemisphere was observed (see Fig. 1, at level —20 on the Z axis). Also activated during the face-identification task were the most anterior regions of the temporal cortex, including the temporal pole of both hemispheres and the adjacent medial areas, as illustrated in Fig. 1. All these foci of activation are localized in the ventral parts of the cortex, as indicated by their values on the dorsoventral plane
FIG 1. Foci of activation superimposed on a magnetic resonance image of the brain (the numbers on the bottom left of Ihe images correspond to the slice level on the Z axis). Gratings: gratings minus fixation (a) activation at the level of the occipital cortex; (b) activation at the level of the left sensorimotor cortex. Face gender: face gender minus fixation; slices are 3 mm apart, starting at Z = - 8 . Face identity: (a) face identity minus face gender; (b) face identity minus fixation; slices are 3 mm apart, starting at Z = —23. Object: object-recognition minus gratings (a) absence of activation of the anterior temporal cortex, (b) activation of the left occipito-temporal cortex and the left temporal area 21.
Face Oender
Face Identity
Downloaded from by guest on April 24, 2015
z o
GO
oo
n m
O
•v
-i
m n
a
a o
2
m
O
•n
o
•
