OR I G INA L ART I C L E

Three-Dimensional Representations of Objects inDorsalCortex are Dissociable from Those in Ventral CortexErez Freud1,2, Tzvi Ganel2, Ilan Shelef3, Maxim D. Hammer4, Galia Avidan2

and Marlene Behrmann1

1Department of Psychology and Center for the Neural Basis of Cognition, Carnegie Mellon University, Pittsburgh,PA, USA, 2Department of Psychology and Zlotowski Center for Neuroscience, Ben-Gurion University of the Negev,Beer-Sheva, Israel, 3Radiology Unit, Soroka University Medical Center, Beer-Sheva, Israel, and 4Department ofNeurology, University of Pittsburgh Medical Center, Pittsburgh, PA, USA

Address correspondence to E. Freud, Department of Psychology, CarnegieMellon University, Baker Hall 331H. 5000 Forbes Avenue, Pittsburgh, PA 15213, USA.Email:erezfreud@gmail.com

AbstractAnestablishedconceptualizationof visual cortical function is one inwhichventral regionsmediate object perceptionwhile dorsalregions support spatial information processing and visually guided action. This division has been contested by evidence showingthat dorsal regions are also engaged in the representation of object shape, even when actions are not required. The criticalquestion is whether these dorsal, object-based representations are dissociable from ventral representations, and whether theyplay a functional role in object recognition.We examined the neural and behavioral profile of patients with impairments in objectrecognition following ventral cortex damage. In a functional magnetic resonanace imaging experiment, the blood oxygen level-dependent response in the ventral, but not dorsal, cortex of the patients evinced less sensitivity to object 3D structure comparedwith that of healthy controls. Consistently, in psychophysics experiments, the patients exhibited significant impairments inobject perception, but still revealed residual sensitivity to object-based structural information. Together, these findings suggestthat, although in the intact system there is considerable crosstalk between dorsal and ventral cortices, object representations indorsal cortex can be computed independently from those in ventral cortex. While dorsal representations alone are unable tosupport normal object perception, they can, nevertheless, support a coarse description of object structural information.

Key words: 3D perception, impossible objects, object agnosia, object recognition, two visual systems

IntroductionStandard conceptualizations of the organization of the visualcortical system espouse a segregation between ventral cortex,whichmediates object recognition, and dorsal cortex, which sup-ports spatial representations (Ungerleider andMishkin 1982) andgoal-directed actions (Goodale and Milner 1992). Consistent withthis distinction, many studies have confirmed that ventral cor-tical regions are highly sensitive to object shape and are invariantto changes in viewpoint or retinal size (Gross et al. 1972;Desimone et al. 1984; Malach et al. 1995; Booth and Rolls 1998;Grill-Spector et al. 1999; Kourtzi and Kanwisher 2000). More

surprising, then, is the evidence that object-selective responses

are also observed in dorsal cortical regions, even for tasks that

do not involve visually guided actions. These responses were

initially attributed to attentional or intentional processes,

(Grill-Spector et al. 1999; Kourtzi and Kanwisher 2000). However,

recent studies have shown that these responses are largely inde-

pendent of image transformation (Konen and Kastner 2008) and

support object-related spatial processing (Xu 2009; Bettencourt

and Xu 2013; Zachariou et al. 2014; Freud, Rosenthal, et al. 2015).The involvement of the dorsal cortex in object perception is

also in-line with behavioral and imaging studies that show that

© The Author 2015. Published by Oxford University Press. All rights reserved. For Permissions, please e-mail: journals.permissions@oup.com

Cerebral Cortex, 2015, 1–13

doi: 10.1093/cercor/bhv229Original Article

1

Cerebral Cortex Advance Access published October 18, 2015 at A

cquisitions DeptH

unt Library on O

ctober 22, 2015http://cercor.oxfordjournals.org/

Dow

nloaded from

action observation can facilitate object perception (Helbig et al.2006; and also see Creem and Proffitt 2001), and that this processis supported by parietal regions in the dorsal cortex that modu-late the activation profile in portions of the ventral cortex (Kieferet al. 2011; Sim et al. 2015). The existence of such crosstalk wasalso established by anatomical investigations that showed thatthe vertical occipital fasciculus travels between the ventral anddorsal cortices and provides the structural infrastructure for thecrosstalk between the two systems (Yeatman et al. 2014).

Despite the considerable evidence that suggests that thedorsal cortex is involved in object perception, the nature ofsuch representations and their behavioral significance are notyet clear. One possibility is that the object representations in dor-sal regions are merely the result of feedforward signals from theobject representations in ventral cortex. An alternative possibil-ity is that the dorsal representations are dissociable from ventralcortex and can be computed independently of the ventral repre-sentations, that is, they can be generated in the absence of theinput from the ventral cortex.

To adjudicate between these alternatives, we employed neuroi-maging and behavioral measures to examine whether individuals,with a lesion to ventral cortex (Fig. 1) and impaired object recogni-tion, still evince sensitivity to object structural information. Thesepatients all suffer from visual object agnosia (see Table 1 for casedescriptions), a deficit in object recognition which is not attributedto a general loss of knowledge about objects or to impaired intelli-gence (Farah 1994). This deficit is usually observed after ventrolat-eral or ventromedial occipitotemporal cortex lesions (e.g., Goodale

et al. 1991; Karnath et al. 2009; Konen et al. 2011). Importantly,dorsal cortex is intact in these individuals, thereby providing theopportunity toexplore object representations indorsal cortex large-ly without the contribution of ventral cortex. If dorsal representa-tions are sufficiently sensitive to object structural information,then the dorsal cortex of the agnosia patients should evince differ-ential sensitivity to visual displays that either obey or violate thelegality of objects’ spatial structure. This sensitivity should con-tinue to be apparent in behavioral tasks even if the patients’ overallperceptual performance is impaired relative to controls.

To assay the sensitivity of dorsal object representations, weemployed pictures of possible and impossible objects (Penroseand Penrose 1958; Fig. 1). Contrasting these 2 classes of objectsis informative: in contrastwith possible objects, in impossible ob-jects, there are inconsistencies between global and local informa-tion, that is, while the local cues are valid, the resulting 3D globalstructure is incoherent (shown from an accidental viewpoint)and violates the legality of the structure of real-world objects.Thus, although global depth cues may be extracted rapidly andeffortlessly (Di Luca et al. 2004), the contradicting sources of glo-bal and local information cannot be resolved and observers arehighly sensitive to these violations (Shuwairi et al. 2007; Regolinet al. 2011; Freud, Avidan, et al. 2015). Moreover, as we haveshown in a recent functional magnetic resonanace imaging(fMRI) study, there is greater activation for impossible than pos-sible objects in both ventral and dorsal cortices (Freud, Rosenthal,et al. 2015) and thus, this contrast can serve as a probe of thesensitivity of dorsal cortex to object structure.

Figure 1. Representative axial slice from the MRI or computed tomography (CT) scan of each of the 4 patients with object agnosia, all showing right (RN, SM, and EC) or

bilateral (CR) occipitotemporal lobe involvement (top row). Representative slices from a CT scan of a visual-form agnosia patient (JW) (bottom row, left). Representative

slice from the MRI scan of each of the brain-damaged controls showing damage to the left occipitotemporal lobe (GB) or right parietal lobe (KL) (bottom row, right). Slices

are presented in radiological convention. Details regarding etiology of lesion and time since onset may be found in Materials and Methods and in the Supplementary

Material.

2 | Cerebral Cortex

at Acquisitions D

eptHunt L

ibrary on October 22, 2015

http://cercor.oxfordjournals.org/D

ownloaded from

Materials and MethodsParticipants

All participants gave written informed consent to participate,and the protocol was approved by the Institutional Review Board,Carnegie Mellon University (CMU) and by the ethics committee ofBen-Gurion University of the Negev. The patients and a subset ofcontrol participants were tested in Pittsburgh and the remainingcontrol participants were tested in Israel. Patients were tested ei-ther at the university or in their home.

Patients with Visual Agnosia

Five premorbidly normal right-handed individuals participatedin these experiments. After a lesion sustained in adulthood(except for CR who was aged 16 years at the lesion onset), all in-dividuals are impaired at visual perception. Table 1 briefly sum-marizes the biographical and neurological history for eachpatient; additional information about the etiology of the lesionsand about the structural scan is available in the SupplementaryMaterial and in previous publications (Mapelli and Behrmann1997; Vecera and Behrmann 1997; Gauthier et al. 1999; Marottaet al. 2001; Marotta et al. 2002; Behrmann and Kimchi 2003;Rosenthal and Behrmann 2006; Behrmann and Williams 2007;Nishimura et al. 2010; Konen et al. 2011; Behrmann and Plaut2014; Gilaie-Dotan et al. 2013, 2015).

Brain-Damaged Controls

Two additional patients who sustained brain damagewere tested,onewith damage to the left ventral cortex (GB), and the other withdamage to the right parietal lobe (KL) (Fig. 1, bottompanel). GBhasan upper right visual field quadrantanopia and pure alexia. KLdoes not reveal any perceptual difficulties. Neither of these pa-tients is impaired in object recognition. More detailed informationis available in the Supplementary Material and in previous publi-cations (Mapelli and Behrmann 1997; Vecera and Behrmann 1997;Gauthier et al. 1999; Marotta et al. 2001; Marotta et al. 2002;Behrmann and Kimchi 2003; Rosenthal and Behrmann 2006;Behrmann and Williams 2007; Nishimura et al. 2010; Konen et al.2011; Behrmann and Plaut 2014; Gilaie-Dotan et al. 2013, 2015).

Healthy Controls

A group of 10 healthy participants participated in the behavioralstudy, with 2 controls matched to each visual agnosia patient onage and handedness. Three of the matched control participantsfor CR and SM also participated in the fMRI experiment. Two add-itionalmatched controlswere recruited for the fMRI studyandalsocompleted the behavioral session. These latter participants werenot included in the behavioral experiments analysis to maintainan equal number (2) of matched controls for each patient.

Stimuli

Stimuli were 71 pairs of grayscale line-drawings of possible andimpossible objects that had been used in previous studies(Freud, Avidan, et al. 2013; Freud, Ganel, et al. 2013, 2015; Freud,Avidan, et al. 2015; Freud, Rosenthal, et al. 2015). For each possibleobject, an impossible object was derived by altering one or a fewfeatures, resulting in a modification of the object’s global struc-ture from possible to impossible (Fig. 2). Low-level features wereequivalent between object categories as shown by statistical Tab

le1Su

mmaryofca

sehistories

Objec

tag

nosia

Visual-form

agnosia

Brain-d

amag

edco

ntrols

Patien

tSM

ECRN

CR

JWGB

KL

Age

(gen

der)

40(M

)51

(F)

58(M

)36

(M)

58(M

)74

(F)

50(M

)Controls’ag

e42

±5.6

51±0

62.5

±0.7

36±0

60±4.2

––

Tim

epost

lesiononse

t22

years

11ye

ars

16ye

ars

20ye

ars

23ye

ars

7ye

ars

1ye

arLe

sionsite

Rightve

ntral

occ

ipitotem

poralc

ortex

Bilateral

ventral

occ

ipitotem

poralco

rtex

Bilateral

ventral

portionof

occ

ipital

lobe

Left

ventral

occ

ipitotem

poralco

rtex

Rightparietalco

rtex

Neu

ropsy

chologica

ldiagn

osis

Objec

tag

nosia,

proso

pag

nosia

Objec

tag

nosia,

proso

pag

nosia

Visual-form

agnosia

Pure

alex

iaNovisu

alim

pairm

ent

Visual

fieldstatus

Fullvisu

alfield

Fullvisu

alfield

Fullvisu

alfield

Fullvisu

alfield

Inco

mplete

upper

left

scotom

aUpper

righ

tvisu

alfield

quad

ranta-n

opia

Fullvisu

alfield

Objec

tperce

ption

Agn

osic(3

SDs)

Objec

treco

gnition

difficu

lties(scree

ning)

Agn

osic(3

SDs)

Agn

osic(3

SDs)

Agn

osic(3

SDs)

Noobv

iousim

pairm

enton

screen

ing

Noobv

iousim

pairm

ent

onsc

reen

ing

Face

perce

ption

Proso

-pag

nosic(3

SDs)

Face

reco

gnition

difficu

lties(scree

ning)

Proso

-pag

nosic(3

SDs)

Proso

pag

nosic(3

SDs)

Proso

pag

nosic(3

SDs)

Unkn

own

Noobv

iousim

pairm

ent

onsc

reen

ing

Word

perce

ption

Mildim

pairm

ent

(1–2SD

s)Unkn

own

Mildim

pairm

ent

(1–2SD

s)Mildim

pairm

ent(1–2SD

s)Alexic(3SD

s)Pu

realex

ic(3

SDs)

Noobv

iousim

pairm

ent

onsc

reen

ing

Experim

ents

fMRI+

behav

ioral

experim

ents

Beh

aviorale

xperim

ents

Beh

avioral

experim

ents

fMRI+

behav

ioral

experim

ents

Beh

avioralex

perim

ents

Beh

aviorale

xperim

ents

Beh

aviorale

xperim

ents

Dissociable Dorsal Object Representations Freud et al. | 3

at Acquisitions D

eptHunt L

ibrary on October 22, 2015

http://cercor.oxfordjournals.org/D

ownloaded from

comparisons showing equivalence in the overall number ofpixels and the number of pixels that defined edges [t’s < 1].

Procedure

fMRI Experiment.General procedure. Three patients (CR, SM, andRN) and 5 matched controls completed a MRI session that wasconducted before the behavioral session described below. Thissession included a 3D anatomical scan, 2 localizer runs, and 4experimental runs. The data from RN were not analyzed due toexcessive headmovements. Additionally, to obtain a broader dis-tribution of data from control participants, we included the datafrom an identical experiment that was performed in a differentscanner and with young controls (age = 26 ± 1.5) (Experiment 1a;Freud, Rosenthal, et al. 2015). A preliminary analysis ensuredthat no differences were observed between the two controlgroups in terms of sensitivity to object possibility (at CMU versusFreud, Rosenthal, et al. 2015 scanned in Soroka Medical Center;Fs < 1) and, thus, we merged the control data.

MRI setup. Participants were scanned in a 3T head scanner (Verio,Siemens) at Carnegie Mellon University. Details of the scanning

parameters and of visual stimulation setup are described in theSupplementary Material.

Localizer scan. Participants were presented with a standardblocked-design localizer experiment used for delineating objectselective regions. Stimuli were presented in 10-sec blocks, eachcomprised of 20 images. The stimuli in a given block were froma specific visual category (faces, houses, musical instruments,novel objects, or scrambled objects). Note that we used this largevarietyof object categories to remain consistentwith other studiesthat perform localization of object–selectivity. To maintain theirattention, participants performed a one-back task (respondingby button press to a consecutive repeat of the same stimulus)and there was one image repetition within each block.

Possible/impossible scans. Participants completed 4 runs in whichthey viewed 30 blocks of possible and 30 blocks of impossible ob-jects. Each 20 s blocks was comprised of 8 stimuli, each presentedfor 2000ms followed by 500 ms fixation. Participants performed aone-back task and there was one image repetition within eachblock.

Data Analysis

fMRI data were processed using BrainVoyager 2.6 QX software(BrainInnovations, Maastricht, the Netherlands. RRID: nif-0000-00274) and complementary in-house software written in Matlab(TheMathWorks, Inc., Natick, MA, USA. RRID: nlx_153890). Prepro-cessing included 3D-motion correction and filtering of lowtemporal frequencies (slow drift) and was followed by concaten-ation of the 4 experimental runs for each participant. No spatialsmoothing was applied.

Selection of Regions of Interest

For each subject (i.e., patients and controls), regions of interest(ROIs) were defined individually based on the localizer scans at asignificance level of at least q(FDR) < 0.05. As in our previousstudy (Freud, Rosenthal, et al. 2015), object-selective regions weredefined by the contrast of (novel objects +musical instruments) >scrambled objects. The regions that were selected as ROIs wereshown to be selective to objects and to 3D structure of objects inprevious studies. The lateral occipital complex (LOC) (Malachet al. 1995) is the best known and well-studied object selective re-gion. Importantly, this region was also found to be involved in 3Dperception (e.g., Moore and Engle 2001). Another ventral region,the ITS, was also identified by previous studies, as an object select-ive region (e.g.,Hassonet al. 2003) andwas therefore included in thecurrent study. Additionally, 2 dorsal regions were defined as ROIs,the transverse occipital sulcus (TOS), which is usually known as ascene selective region (Hasson et al. 2003) but is also responsiveto spatial properties of objects (Troiani et al. 2014) and the pos-terior parietal cortex (PPC), which is involved in the perceptionof objects’ spatial features (Faillenot et al. 1999) and supportsobject individuation (Bettencourt and Xu 2013). SupplementaryTable 1 includes the mean Talairach coordinates of the selectedROIs and the mean cluster size for the controls and separatelyfor CR and SM.

Univariate Analysis

The 4 runs of each participant were concatenated and analyzedusing a general linear model (GLM) and possible and impossibleobjects were contrasted to identify whether a region is sensitive

Figure 2. Examples of possible (left panel), and their matched counterpart

impossible objects (right panel). Note that there are minimal physical differences

between object types, while the perceptual experience in viewing these 2 object

sets is substantially different.

4 | Cerebral Cortex

at Acquisitions D

eptHunt L

ibrary on October 22, 2015

http://cercor.oxfordjournals.org/D

ownloaded from

to object possibility. Next, for each participant and each ROI, weused the mean β-weights for impossible and possible objects tocalculate a ratio measuring the selectivity to possible/impossibleobjects:

Possible/impossible selectivity ratio

¼ ð Beta weightimpossible objects � Beta weightpossible objectsÞðBeta weightimpossible objectsþBeta weightpossible objectsÞ

:

Positive values indicate that impossible objects elicit greateractivation compared with possible objects while negative valuesindicate the opposite. Zero indicates equivalent activation for the2 object categories.

To compare between the patients and controls, we conducteda bootstrapping analysis which consisted of 10 000 iterations inwhich the data from 12 healthy controls were randomly sampledwith replacement, and their object possibility sensitivity indiceswere averaged. This analysis yielded a distribution of the meanselectivity index for each ROI and the mean selectivity index ofthe patients was compared with this distribution. To establishthe statistical significance of the difference between controlsand patients, we calculated the 99% confidence interval (CI) ofthe obtained bootstrap distribution of the mean.

Behavioral Experiments

General ProcedureThe behavioral session included 4 experiments. In all experi-ments, practice was first given and feedback provided. Duringthe experiment, trials were self-initiated and no feedback wasprovided.

Experiment 1—same/different classification with matched objects. Par-ticipants made speeded same/different judgments on 80 trials inwhich pairs of objects were presented. Half of the pairs were ofthe same object (possible/impossible) and half were of differentobjects. The “different” trialswere of possible andmatched impos-sible objects that differed fromeachother by few features. Thefirstobject in half of these trials was of the possible version, while inthe remaining trials the first object was of the impossible version.

On each trial, the first object was presented for 3000 ms, fol-lowed by amask screen (500 ms) that was composed of scrambled(400 fragments) objects. The second object was then presenteduntil the participant provided a response (Fig. 4A).

Experiment 2—depth comparison of spatial information. One greenand one red dot were superimposed on the stimuli and partici-pants were required to judge which of the 2 markers was locatedcloser in depth to the participant. The dots were located at thesame position for the matched possible and impossible object(Fig. 4B). The assignment of the dots was counterbalanced acrossstimuli such that for half of the stimuli, the closer dot was lowerin the vertical plane and for the other half the closer dot washigher in the vertical plane. Subject agreement about the close/far spatial location of the dots was validated in a prior pilotstudy. The experiment included 142 stimuli (half possible andhalf impossible). Each stimulus was presented until participantsprovided a response and was then followed by a masking screenthat was presented for 500 ms.

Experiment 3—possible/impossible classifications. Participants wereinstructed to classify object possibility (Schacter et al. 1990). Pos-sible objects were defined as objects that could legitimately exist

in 3D space, while impossible objects were defined as objects thatcould not exist in 3D space. The experiments included the stimuliused in Experiment 2. Stimuli were presented until participantresponse and then followed by amask shown for 500 ms (Fig. 4C).

Experiment 4—same/different classification with nonmatched objects.This last experiment serves as a control experiment to offersome specificity on the mechanisms that mediate the advantagefor possible over impossible objects. The procedurewas similar toExperiment 1, however the “different” trials were of two objectsthat remarkable differed from each other. Half of these trialswere of a possible object, while the remaining trials were consti-tuted from impossible object. The logic here is that because thetwo objects in a pair, when they differ, are so easily discriminableas being different, there is no need to compute a complex 3Dshape (discrimination can be based on simple features, com-puted in a 2D description). The prediction is that even if the con-trol participants perform better than the patients, there shouldbe no interaction with object type.

ResultsFunctional MRI

fMRI ExperimentTo examine the sensitivity of the dorsal and ventral cortices toobject structural information in individuals with a lesion to theoccipitotemporal cortex, we analyzed the fMRI response in theventral and the dorsal cortices of 2 patients with visual agnosiaand 12 healthy controls.

Behavioral Results During fMRI ScanningAccuracy was high in both patients and controls, with better per-formance for the controls regardless of object category single-case analyses (Crawford and Garthwaite 2002): t’s > 2.1, P < 0.06.No differences were found between possible and impossible ob-jects [F1,9 = 2.5, P > 0.1]. Unsurprisingly, reaction times (RT) wereslower for the patients than the controls [single-case analyses:t’s > 2.05, P < 0.07] but were similar across object categories [F < 1,ns] (Table 2). Thus, any differences in the fMRI signal betweenpossible and impossible objects could not obviously be attributedto the fact that signal was summated over a longer activation per-iod in one condition (or participant group) over the other.

Univariate AnalysisObject-selective ROIs in the occipitotemporal (LOC, ITS) and occi-pitoparietal (TOS, PPC) cortices were defined in each hemisphere(Fig. 3A). A GLM analysis was conducted using the data from thecontrols, separately for each object-selective region. The weightsfor possible and impossible objects were calculated and sub-jected to repeated-measures analysis of variance (ANOVA) withROI, hemisphere and object possibility as within-subject vari-ables. This analysis revealed robust effects of object possibility,with greater activity for impossible compared with possible ob-jects [F1,11 = 52.35, ηp2 ¼ 0:82, P < 0.001], and planned comparisonsconfirmed that all object selective regions exhibited a similarpattern [Fs1,11 > 19.6, P < 0.01]. Additionally, an interaction wasobserved between hemisphere and possibility [F1,11 = 9.26,ηp

2 ¼ 0:45, P < 0.05], with stronger sensitivity to object impossibil-ity in the right than left hemisphere, a finding that is consistentwith the lateralization of object-based spatial information pro-cessing to the right hemisphere (e.g., Ditunno and Mann 1990;Farah 1994; Harris and Miniussi 2003; Zacks et al. 2003).

Dissociable Dorsal Object Representations Freud et al. | 5

at Acquisitions D

eptHunt L

ibrary on October 22, 2015

http://cercor.oxfordjournals.org/D

ownloaded from

We analyzed the data from the patients in the same way but,first, we ensured that a reliable signal could be derived from theROIs. For SM, because his lesion overlapped the right ITS, no ac-tivationwas observed in this region in the localizer scans. For CR,although the right ITSwas defined in the localizer, no reliable sig-nal was obtained from this ROI in the experimental runs, andtherefore this ROI was excluded from further analysis. All otherregions exhibited normal event-related activation profiles andwere used in subsequent analyses (see Supplementary Fig. 1).

To compare the BOLD profile of the two patients to the controlgroup, we conducted a bootstrapping analysis of the distributionof normal sensitivity indices to object possibility in each ROI forthe controls: sensitivity to object impossibility (greater responseto impossible objects) is defined as a value greater than zero (seeMaterials and Methods for details). This analysis revealed that,in all dorsal ROIs, patients exhibited sensitivity to object possibil-ity and itwas equivalent to that of the controls (see Fig. 3B). In con-trast, in the ventral regions, for the patients, 2 of the 3 sampledROIs did not show sensitivity to object structure: this was true in

both the left LOC [meanpatients =−0.01, meanControls = 0.05, P < 0.01,99% CI (0.02, 0.09)] and left ITS [meanpatients = 0.001, meanControls =0.05, P < 0.01, 99% CI (0.02, 0.08)]. The mean sensitivity in the rightLOC was in the normal range [meanpatients = 0.04, meanControls =0.05, 99% CI (0.02, 0.09)] as a result of SM’s sensitivity to object pos-sibility in this ROI. This LOC region is posterior to SM’s lesion,which might account for his preserved sensitivity. A similarbootstrapping analysis that equalized the number of controls topatients by sampling just 2 controls in each iteration yielded asimilar pattern of results.

In addition to the group-level analysis, we performed single-case analyses using the revised standardized difference test(RSDT) (Crawford and Garthwaite 2005) which tests whether thedifference between an individual’s standardized score on 2 con-ditions (possible versus impossible) is significantly differentfrom the difference observed in the control sample. The RSDTconfirmed that SM was significantly impaired in his sensitivityto object possibility in the left ITS [t(11) = 2.11, P = 0.05], while CRexhibited reduced sensitivity to object possibility in the left LOC

Table 2 The behavioral performance during the fMRI scan

Accuracy RTs

Possible Impossible Avg. Possible Impossible Avg.

Controls 98.8% (±0.009%) 100% (±0%) 99.4% (±0.009%) 609 (±113 ms) 623 (±125 ms) 616 (±117 ms)SM 92% 90% 91%** 1104 ms 996 ms 1035 ms**CR 94% 100% 97%** 860 ms 880 ms 870 ms*

Note: The performance of the patients was compared with the controls using a single-case analysis (see text for details). *indicates a marginal effect (P < 0.1).

**indicates a significant effect (P < 0.05).

Figure 3. fMRI experiment. (A) ROI definition—ROIs were defined based on an independent localizer. The averaged ROIs are presented on inflated hemispheres of

representative subject. (B) For each ROI, a distribution of the mean differences between the beta weights of impossible and possible objects was created (see text for

details). The red vertical lines represent the 99% CI of the mean. The red dot in each graph represents the mean sensitivity index of the patients. For controls, all

objects’ elective regions exhibited stronger activation for impossible compared with possible objects as evident from the CI that did not include zero. In contrast, for

CR and SM dorsal ROIs were still sensitive to object possibility while most of the ventral ROIs were insensitive to structural information.

6 | Cerebral Cortex

at Acquisitions D

eptHunt L

ibrary on October 22, 2015

http://cercor.oxfordjournals.org/D

ownloaded from

[t(11) = 2.3, P < 0.05] and a marginal effect in the right LOC [t(11) =2.03, P = 0.06]. Critically, the patients showed normal sensitivityto object possibility in dorsal ROIs [all Ps > 0.15] (see Supplemen-tary Fig. 2).

The results of the fMRI experiment confirm that, as expected,visual agnosia is accompanied by reduced neural selectivity to 3Dstructure in the ventral cortex. More importantly, the resultsshow, for the first time, that the fMRI signal in dorsal cortex re-flects sensitivity to 3D object structural information equivalentto that of the controls. This finding provides strong support forthe claim that object representations in dorsal cortex are inde-pendent of those in ventral cortex.

Before reaching this conclusion definitively, however, weexplore 2 alternative interpretations. One possibility is that thesensitivity to 3D object structural information is a product ofthe left hemisphere ventral regions. This can be ruled out as CRhas bilateral lesions and SM, who has a unilateral right lesion,also exhibited impaired sensitivity in his intact left ventral cortex(see Konen et al. 2011 as well).

A second possibility is that the sensitivity noted in the dorsalcortex in the patientswasmediated by the observed residual ven-tral sensitivity and perhaps, then, is not entirely independent. Toevaluate this possibility, we applied a Bayesian method that en-ables us to determine whether sensitivity to object impossibilityin dorsal ROIswas still preserved in the patients even after the re-sidual ventral sensitivity was accounted for as a covariate (Craw-ford et al. 2011). The results confirmed that, in all dorsal ROIs,there was no difference in sensitivity to structural informationbetween the patients and the controls (Ps > 0.3). The onlyexception was the right PPC of SM that exhibited marginal over-sensitivity compared with the controls (Z-ccc = 4.6, P = 0.07).

Behavioral Experiments

The results of the fMRI experiment suggest that representations inthe dorsal ROIs are sensitive to object structure in patients with alesion to ventral cortex. These findings provide novel evidence forthe functional dissociations of object representations in dorsalcortex. If these dorsal representations play a functional role in per-ception, then we might predict a behavioral difference to possibleversus impossible objects in these patients even though their ob-ject perception accuracy may be lower than that of controls.

We evaluated the patients’ sensitivity to the coherence ofstructural information in a series of psychophysical tasks that var-ied in their perceptual requirements. Two additional patientswithvisual agnosia participated in these experiments (n = 4). Compari-sons were made between the patients versus two groups of con-trols, individuals with no neurological damage and two patientswho are not agnostic and have brain damage to regions outsideof right ventral cortex.

Experiment 1: Same–Different Classifications (matched objects)Experiment 1 utilized a task that could be completed solely on thebasis of 2D information but, at the same time, demanded fine-grained processing. Participants performed same–different clas-sifications onpairs of possible and impossible objects (Fig. 4A, leftpanel). To enforce detailed processing of the stimuli, the differ-ences between stimuli were minor even in the “different” trials(seeMaterials andMethods for details). In this task, a representa-tion is generated from the first object, which is later comparedwith the second, on-screen object. Thus, to the extent that struc-tural information is processed in this task, better performance ispredicted for trials in which a possible object was presented asthe first object compared with trials in which the first object

was impossible and a coherent representation could not be gen-erated (Schacter et al. 1990; Soldan et al. 2009; Freud, Hadad, et al.2015; Freud et al. 2015). Additionally, it was predicted thatpatients with visual agnosia would be impaired on this taskcompared with healthy controls, but, would nevertheless showpreserved sensitivity to object possibility which may rely on theintact dorsal object representations.

In this experiment, object possibility was orthogonal to thetask (same/different judgement) at hand, and therefore sensitiv-ity could be calculated independently for possible and impossibleobjects (note that analysis of the raw data, which included bothobject type (possible–impossible) and trial type (same–different)yielded similar results, with better accuracy and faster RTs forpossible objects, regardless of experimental group. Yet, patientsshowed better performance in the “same trials” mirroring a per-ceptual bias. Importantly, the d′ analysis, is not affected by thisperceptual bias, and reflects the perceptual sensitivity to objectpossibility.). Repeated-measures ANOVA revealed a robust differ-ence between the groups, with better performance for thematched controls than the patients [F1,12 = 44.97, ηp2 ¼ 0:78, P <0.001]. However, despite a quantitative trend in this direction,no advantage was observed for trials primed by possible com-paredwith impossible objects [F1,12 = 2.23, P = 0.16], with no inter-action with group [F1,12 < 1] (Fig. 4A). Single-case analyses(Crawford and Garthwaite 2002; Crawford et al. 2010) confirmedthat all patients were impaired in their overall performance [t’s >2.67, P < 0.05], while the brain-damaged controls performedequivalently to healthy controls [t’s < 1]. Additionally, the infer-ential methods for comparing 2 single cases (Crawford et al.2010), showed that brain-damaged controls performed betterfrom the agnosic patient [t’s > 2.26, P < 0.05], excluding EC[t’s < 1.68, P > 0.1].

Despite the absence of an effect for object type in terms of d′scores, RTanalysis revealedamaineffect of object possibility [F1,12= 8.28, ηp2 ¼ 0:4 P < 0.001] with faster RTs for trials primed bypossible objects. Importantly, this effect was evident to a similarextent in the patients and the control groups as shown by thelack of an interaction between group and object possibility[F1,12 < 1]. Specifically, all patients exhibited faster RTs for possiblecompared with impossible objects, excluding RN who performedslightly faster (5 ms) for impossible objects, butmight have tradedspeed and accuracy as he had a considerably better d′ score forpossible than impossible objects. To assess the similarity of eachpatient’s RT to the controls with respect to sensitivity to objectpossibility, we applied the RSTD analysis and confirmed that allpatients performed within the normal range [t’s < 1.3, P > 0.2].

Despite the lack of interaction between the groups and thedescriptive statics that showed the advantage for possible objectsin all patients, we wanted to ensure that the patient group wasstatistically significantly better for possible compared with im-possible objects. However, the utilization of simple comparisonsfor 4 subjects is highly unreliable. Therefore, we applied a ran-domization analysis, which consisted of 10 000 iterations inwhich the data from each patient was sampled with replace-ment, and was randomly assigned to the different conditions(i.e., possible or impossible). This method yielded 10 000 randomsamples, which were based on the patients’ data. Next, we sub-tracted the “impossible” condition from the “possible” conditionto create the random distribution of the possibility advantageand calculated the 95% CI of this distribution. Finally, we exam-ined whether the results obtained in the experiment exceededthe CI and therefore deviated from the random distribution.

The randomization analysis confirmed that the patients hadbetter sensitivity for possible objects compared with impossible

Dissociable Dorsal Object Representations Freud et al. | 7

at Acquisitions D

eptHunt L

ibrary on October 22, 2015

http://cercor.oxfordjournals.org/D

ownloaded from

Figure 4. Experiments 1–4 design (left panel) and results (middle and right panels). The data of each agnosic patient (RN, SM, CR, and EC) and brain-damaged controls (KL

and GB) is plotted separately in addition to the average results for the patients andmatched controls. Agnosic patients,matched controls and brain-damaged controls are,

respectively, designated by squares, circles, and triangles. Error bars across all figures represent CIs for the main effect of object possibility as calculated for repeated-

measures ANOVAs (Jarmasz and Hollands 2009). (A) Experiment 1: Participants performed same/different classifications on pairs of objects which were physically

highly similar.). Object possibility was orthogonal to the task, and sensitivity was calculated independently for possible (gray) and impossible (black) objects. The

overall performance of the matched controls was better (middle panel), but sensitivity to object possibility was equivalent between the two groups (right panel). (B)

Experiment 2: participants performed depth classifications—which dot is closer in depth—the red (black in figure) dot or the green dot (white in figure). Overall,

matched controls performed better than patients (middle panel). Both groups exhibited superior performance for possible compared with impossible objects (right

panel). (C) Experiment 3: participants performed overt possible/impossible classifications. Patients had only minimal awareness to object 3D structure possibility

while matched control exhibited high sensitivity to this information. (D) Experiment 4: Participants performed same/different classifications on pairs of objects which

were physically obviously different. Overall, matched controls performed better than patients (middle panel). Both groups exhibited similar performance for possible

and impossible objects (right panel).

8 | Cerebral Cortex

at Acquisitions D

eptHunt L

ibrary on October 22, 2015

http://cercor.oxfordjournals.org/D

ownloaded from

objects [mean d′ advantage = 0.24, mean random advantage =0.001, P < 0.05, 95% CI (−0.18, 0.18)] and this was also true interms of RTs [mean RT difference = 120 ms, mean random RTdifference = 0.48 ms, P < 0.05, 95% CI (−120, 119 ms)] (note thatanalysis of the raw data, which included both object type (pos-sible-impossible) and trial type (same-different) yielded similarresults, with better accuracy and faster RTs for possible objects,regardless of experimental group. Yet, patients showed betterperformance in the “same trials”mirroring a perceptual bias. Im-portantly, the d′ analysis, is not affected by this perceptual bias,and reflects the perceptual sensitivity to object possibility.).

The results of Experiment 1 suggest that, notwithstanding thelesion to the ventral cortex, patients with visual agnosia still pro-cess structural information and retain sensitivity to the validityof object structure. Although patients exhibited highly impairedperformance in the perceptual task, they were as sensitive toobject possibility as the controls.

Experiment 2: Depth Comparison of Spatial InformationExperiment 1 revealed that, despite their reduced accuracy in ob-ject perception, the patients were still sensitive to the legality ofobject structure. Because, arguably, this task does not necessarilydemand the processing of the object as a whole, the observedsensitivity to object possibility might reflect the patients’ intactlocal processing of the structural violations that differentiate be-tween object categories. To explore whether the observed sensi-tivity is a product of structural representation of objects per se, inExperiment 2, two dots were superimposed on each object andparticipants decided which dot is closer to the observer indepth (Fig. 3B, left panel). The dots were always located in differ-ent locations of the object to enforce the processing of the objectas a whole.

Consistent with the results of Experiment 1, with d′ as the de-pendentmeasure, patients were less accurate than controls [F1,12= 13.99, ηp2 ¼ 0:53, P < 0.05] and this was true for all patients, asreflected in the single-case statistics [t’s > 2.63; P < 0.05], withthe exception of CR [t < 1]. The two brain-damaged controlsperformed as well as the healthy control group [t’s < 1] (Fig. 3B,middle panel), and significantly better than the agnosic patients[t’s > 2.097, P < 0.06] excluding CR [t’s < 1].

The ANOVA also revealed an additionalmain effect with betterperformance for possible compared with impossible objects [F1,12= 5.56, ηp2 ¼ 0:31, P < 0.05]. Importantly, this superiority was ob-served to a similar extent for the patients and controls, as evidentfrom the lack of interaction between group and object possibility[F1,12 < 1; Fig. 3B, right panel]. Specifically, all patients exhibitedgreater sensitivity to possible compared with impossible objects,excluding SM who showed similar sensitivity levels to the twoobject categories in termsofd′, but exhibited slowerRTs for impos-sible objects compared with possible objects (note that SM exhib-ited a similar pattern in Experiment 1). The RSDTanalysis showedthat none of the patients was different from the controls in termsof sensitivity to object possibility in this task [t’s < 1].

RT analysis revealed a similar trend with faster responses forpossible compared with impossible objects [F1,12 = 3.13, ηp2 ¼ 0:2,P = 0.1], and with no interaction between object possibility andgroup [F1,12 = 1.59, P > 0.2]. Brain-damaged controls also showedthe same patternwith faster RTs for possible (1954 ms) comparedwith impossible objects (2060 ms) (see Supplementary Fig. 3), andthis did not differ from the other groups [t’s < 1].

To validate the advantage for possible over impossible objects,we applied the randomization analysis similarly to Experiment1. This analysis revealed significantly better sensitivity for pos-sible over impossible objects for the patients [mean d′ advantage

= 0.32, mean random advantage = 0.001, P < 0.05, 95% CI (−0.27,0.27)], while a marginal effect was also observed in terms of RTs[mean RT difference = 43 ms, mean random RT difference = 0.14ms, P = 0.06, 95% CI (−44, 44ms)].

These findings replicate and extend the results of Experiment1, albeit in a task that required computation of structural object-based information: patients with visual agnosia still evinced sen-sitivity to object structural information despite their overall im-pairment in object perception. This preserved sensitivity mightbe mediated by the intact dorsal object representations uncov-ered in the fMRI experiment. This finding suggests that, althoughthese dorsal representations are insufficient to support normalobject perception, they may suffice to support sensitivity tostructural, likely spatial, object information.

Experiment 3: Possible/Impossible ClassificationsThe ability to classify an object as possible or impossible requiresa fine-grained representation of object 3D structure. Specifically,object impossibility is a product of the inconsistency betweenlocal and global cues; only when the inconsistency of these con-tradicting pieces of information becomes apparent can partici-pants label the image as impossible. Based on previous studies(e.g., Delvenne et al. 2004; Turnbull et al. 2004), we predictedthat, despite the observed sensitivity to object possibility in Ex-periments 1 and 2, patients with visual agnosia will be impairedin the possibility classification task.

Participants were shown a single image and were asked toindicate explicitly whether it was possible or not. The agnosicpatients exhibited reduced sensitivity to the possible/impossiblestatus of the objects, compared with matched controls [F1,12 =24.4, ηp2 ¼ 0:67, P < 0.001] and this held true for every patient[t’s > 2.32; P < 0.05]. The sensitivity of the two brain-damaged con-trols did not differ from that of the matched controls [t’s < 1.03,P > 0.3] (Fig. 4C), however, their scores were found not be statis-tically different from the patients [t’s < 1.42, P > 0.15], excludingEC [t’s > 2.94, P < 0.01]. There were no differences in RT betweenthe agnosic patients and healthy controls [F1,13 < 1] nor betweenthe agnosic patients and the brain-damaged controls [t’s < 1](see Supplementary Fig. 3).

The results of Experiment 3 showed that, despite the observedsensitivity to structural representation in Experiment 1 and 2, ag-nosic patients were unable to explicitly determine whether anobject is possible or not. The apparent failure to integrate the3D information for this purpose is likely an outcome ofmore ven-tral computations and sets a boundary on the nature of the dorsalobject-based representations. These results indicate that the dor-sal object representations may be necessary for, but are clearlyinsufficient for, the normal perception of object structure.

Experiment 4: Same–Different Classifications (Nonmatched Objects)The behavioral results presented so far have shown that sensitiv-ity to object 3D structure is observed even when ventral cortex isimpaired. Based on this finding, we suggest that the dorsal cortexcan mediate 3D object representations and these are potentiallydissociable from ventral object representations. We have arguedthat the dorsal representations contribute to 3D processing and itis this sensitivity to spatial geometric structure that gives rise tothe advantage for possible over impossible objects. It is the case,however, that this observed advantage might arise for other rea-sons such as object complexity (i.e., number of junctions, linesetc.). Despite the fact that our stimulus set is highly controlledfor such factors (see Materials and Methods), we wanted to en-sure that the observed advantage for possible objects, truly re-flects 3D processing mechanisms. To this end, participants

Dissociable Dorsal Object Representations Freud et al. | 9

at Acquisitions D

eptHunt L

ibrary on October 22, 2015

http://cercor.oxfordjournals.org/D

ownloaded from

performed a same-different task, but unlike Experiment 1, stim-uli in the “different” trials were obviously different (Fig. 4D, leftpanel). Hence, this task could be performed based on coarse dif-ferences between objects (i.e., 2D properties) and did not requirefine-grained processing of the 3D properties of the stimuli.

As in previous experiments, overall, patients were impairedcompared with controls [F1,12 = 4.5, ηp2 ¼ 0:27, P = 0.05] reflectingtheir perceptual, ventral-related, difficulties. Single-case analysisshowed that SM and CR were within the normal range while theother patientswere impaired even on this simple task [t’s >−2.01,P < 0.07]. More importantly for the purpose of the current experi-ment, similar performance was found for the 2 object categorieswith no interaction with group [Fs < 1] (Fig. 4D). Repeated-mea-sures ANOVA on the RT data did not revealed any significant ef-fects [Fs < 1] (see Supplementary Fig. 3).

The results of Experiment 4 indicates that the advantagefound in Experiments 1 and 2 for possible objects does not relyon 2D shape properties and, rather, is more specifically relatedto the processing of 3D information.

Experiment 5—The Representation of Object Structural Informationin Visual Form AgnosiaAlthough, in the imaging analysis, we attempted to rule out theidea that the sensitivity to object 3D structure might have beenmediated by residual functionality of the ventral cortex byusing ventral stream activity as a covariate, here we provide add-itional evidence to rule out the claim that the dorsal representa-tions are a result of ventral signal propagation. The approach is toreport data from a final patient, JW, a patient who has especiallymarked visual form agnosia following widespread anoxic dam-age affecting the occipitotemporal lobes bilaterally (Mapelli andBehrmann 1997; Vecera and Behrmann 1997; Rosenthal andBehrmann 2006). JW is profoundly impaired in his perceptualabilities even relative to the patients described above. Given thesubstantial ventral lesion, if JW reveals any sensitivity to objectstructural information, it is unlikely to be the product of residualfunctionality of the ventral cortex, licensing us to conclude thatthe object representations in dorsal cortex are independent andfunctionally relevant.

We were unable to obtain imaging data from JW, as we couldnot confirm definitively that his replacement mitral valve wasnot ferromagnetic. Nevertheless a lesion of a similar magnitudein another patient, dramatically reduces object-selective activa-tion in the ventral cortex (James et al. 2003).

Experiment 5a: Same–Different Classifications (Matched Objects)JW had significantly lower d′ in this task, relative to the healthycontrols [t = 5.2, P < 0.001]. The inferential methods (Crawfordet al. 2010) showed that his performance did not differ from theother visual agnosia patients [t’s < 1.9, P > 0.08], whichmay be ac-counted for by afloor effect. Remarkably, however, JW still exhib-ited greater sensitivity to possible than impossible objects, andthis was within the normal range in terms of sensitivity to objectpossibility [RSDT: t = 1.2, P = 0.25] (Fig. 5A).

The RT analysis showed that JW was significantly slowerthan both the healthy controls [t = 7.28, P < 0.0001] and all theagnosia patients [t’s > 4.7, P < 0.01], confirming his perceptualdifficulties. Notably his RT, too, was faster for possible thanimpossible objects, and his sensitivity to object possibility fellin the normal range [RSDT: t = 1.05, P = 0.31] (see SupplementaryFig. 4). As the number of correct trials was small, we re-analyzedthe RT data using all trials and faster RTs were still evident forpossible over impossible objects (possible: 3657 ms, impossible:4202 ms).

Experiment 5b: Depth Comparison of Spatial InformationBased on d′, JW’s performancewas significantly impaired relativeto controls [t = 10.9, P < 0.0001] and also relative to each patient[t’s > 5.35, P < 0.001] (Fig. 5B). Despite his profoundly impaired per-formance, JWretained sensitivity to object possibility similarly tothe healthy controls [t < 1].

This reduced d′ but preserved sensitivity to object structurewas also mirrored in the RT data. JW was significantly slowerthan the healthy controls [t = 9.4, P < 0.001] and also slower thaneach patient [t’s > 6.09, P < 0.001]. Nevertheless, he was evenmore sensitive than the healthy controls to object possibilitywith faster RT for possible objects [RSDT: t = 2.915, P < 0.05] (seeSupplementary Fig. 4) and this was true even when all trialswere included (possible objects 8757 ms; impossible objects10 165 ms).

Experiment 5c: Possible/Impossible ClassificationsAs expected, JW was unable to classify objects as possible or im-possible (Fig. 5C); his d′ was significantly lower than the controls[t = 5, P < 0.001] and marginally lower than the other patients [t’s< 1.82, P < 0.1], excluding EC [t < 1]. RT analysis showed no differ-ences between JWand the controls or the other patients [t < 1.26,P > 0.2] (see Supplementary Fig. 4).

Figure 5. Experiments 1–3 with respect to JW performance. (A,B) JW, a patient with visual form agnosia, was profoundly impaired relative to the controls and relative to

visual agnosia patients in Experiments 1 and 2, but evinced intact sensitivity to object possibility in these tasks with better performance for possible objects (gray)

compared with impossible objects (black). (C) JW was unable to classify objects as possible or impossible.

10 | Cerebral Cortex

at Acquisitions D

eptHunt L

ibrary on October 22, 2015

http://cercor.oxfordjournals.org/D

ownloaded from

DiscussionThe goal of the present study was to determine the nature of ob-ject representations in the dorsal regions of the visual cortex.While previous studies have demonstrated that these regionsare sensitive to object shape (Konen and Kastner 2008; Xu 2009;Bettencourt and Xu 2013; Freud, Rosenthal, et al. 2015) and inter-act closely with ventral cortex to aid recognition (Kiefer et al.2011; Sim et al. 2015), it is unclear whether these representationsresult from a cascade of signals from ventral cortex, or whetherthey are dissociable from ventral representations and independ-ently computed.

To address this, we examined the neural and behavioral re-sponses of 5 patients with visual agnosia, following a lesion tothe ventral cortex, on a range of tasks that required the represen-tation of object structural information. In the imaging results,despite a reduction of sensitivity to object structure in ventral re-gions, regions along the dorsal cortex evinced greater activationfor impossible than for possible objects. Moreover, notwithstand-ing the profound perceptual impairments, the patients revealedbehavioral sensitivity to the integrity of object structural infor-mation, evincing an advantage for geometrically coherent, pos-sible objects over impossible objects. These findings lead us toconclude that dorsal representations are computed independ-ently of ventral object representations. Moreover, the dorsal acti-vation patterns capture representations of object structure andits legality.

However, a possible alternative explanation is one inwhich thedorsal activation is simply the cascade of whatever residualventral function is available. Two lines of evidencemake this alter-native unlikely. The first is that the sensitivity to structural infor-mation in the dorsal cortex of the patients was found to be in thenormal range even when the residual ventral cortex sensitivityserved as a covariate. The second is that in a patientwithprofoundvisual-form agnosia following extensive bilateral occipitotempor-al lesion,withminimal, if any, ventral cortex object functions, per-formance was still better for possible compared with impossibleobjects in 2 different tasks (i.e., Experiment 5a and 5b) to anequal extent as in the control individuals.

Together, these findings provide evidence that dorsal cortex,while disconnected or dissociated from ventral cortex, has suffi-cient object-based information. The question is whether the dor-sal representations contribute to visual perception in normalparticipants. In the normal functioning brain, there are functional(e.g., Kiefer et al. 2011; Sim et al. 2015; Freud, Rosenthal, et al. 2015)and anatomical connections (Yeatman et al. 2014) between dorsaland ventral cortices and, thus, dorsal object representations arelikely modulated and shaped by such connections (and the samemay be true of ventral cortex). Nevertheless, our finding that 3Dobject representations can be generated in dorsal cortex evenwhen ventral inputs are highly reduced and impaired suggeststhat such dorsal representations can be dissociated from and arenot entirely reliant on ventral cortex.

The apparent dissociability of dorsal and ventral object repre-sentations raises a number of important questions for the inter-activity of the ventral–dorsal modulations and for the temporalcourse of the activations. One possible scenario is that the initialcomputation of object representations begins (separately) in thedorsal and ventral cortices and only then the bidirectional con-nections modulate these emerging or existing representations.There are also questions about the weighting of the bidirectionalconnections and potential asymmetries in the strength of theirmodulations. These questions go beyond the scope of the currentstudy and ought to be addressed in future investigations.

Finally, the apparent dissociability of dorsal object represen-tations is consistent with findings showing that the dorsal por-tion of the LOC in visual agnosia patient retained sensitivity toobjects presented tactilely but not visually (Allen andHumphreys2009). Here, we extend this finding and show that a (perhaps par-tial) visual representation of object structure is generated in thedorsal cortex although ventral vision is highly impaired. The re-sults we report here are also compatible with data from anotherimaging study, which includes patient SM (who is one of the par-ticipants in the present study) (Kassuba et al. in preparation). Inthat study, an fMRI adaptation paradigm revealed that SM’s vis-ual responses to both 2D-objects and line drawings were im-paired in all dorsal ROIs relative to healthy control subjects(and to a greater degree in the left than right hemisphere). How-ever, responses to 3D-objects were comparable to the controlgroup, suggesting that 3D structural information may be repre-sented in parietal cortices.

The Behavioral Signature of Object Representationsin the Dorsal Cortex

The imaging experiment demonstrated that the dorsal cortex ofthe visual agnosic patients was differentially activated by pos-sible and impossible objects. Despite this, the patients werestill profoundly impaired in various perceptual tasks, indicatingthat dorsal object representations are not sufficient for normalobject perception. What then is the functional contribution ofdorsal object representations?

The profound impairment in the object possibility classifica-tion task replicates previous studies with visual agnosia patients(Riddoch et al. 2003; Delvenne et al. 2004; Turnbull et al. 2004).However, because we used additional tasks in which the re-sponse was orthogonal to object possibility (e.g., red/greendepth decision), we were able to uncover the sensitivity to 3Dstructural object information (for a similar idea see Farah et al.1991) despite the reduced accuracy and increased RTs of the pa-tients. These results suggest that the dissociable dorsal object re-presentations might primarily be involved in deriving objectstructural information processing. This hypothesis is compatiblewith the importance of parietal regions for spatial and 3D infor-mation processing (e.g., Durand et al. 2009; Georgieva et al. 2009;Srivastava et al. 2009; Orban 2011; Verhoef et al. 2015). The keycontribution of dorsal object representations, then, is in thederivation of the global geometry of objects.

Hemispheric Specialization in the Processingof Structural Information

The present results also point to potential differences betweenthe 2 hemispheres in the processing of structural information.In particular, the fMRI experiment showed greater differential ac-tivation to impossible and possible objects in the right than lefthemisphere. Moreover, while all the visual agnosia patients suf-fered from a right ventral lesion (except CR), GB, who sufferedfrom a large unilateral lesion to her left ventral cortex showedneither a deficit in object perception nor any decrement in per-ceptual sensitivity to structural information.

This left–right hemispheric difference is compatible withprevious studies that have documented right hemisphere spe-cialization in tasks that require 3D rotations (Ditunno andMann 1990; Harris and Miniussi 2003). Furthermore, this spe-cialization, especially in parietal regions, held only when the ro-tationswere object-based, rather than viewer-based (Zacks et al.2003).

Dissociable Dorsal Object Representations Freud et al. | 11

at Acquisitions D

eptHunt L

ibrary on October 22, 2015

http://cercor.oxfordjournals.org/D

ownloaded from

Conclusions

The current study explored the nature and functional contribu-tion of object representations in the dorsal cortex. Together, theimaging and behavioral findings suggest, for the first time, thatthe dorsal cortex canmediate visual information to generate dis-sociable object representations from the ventral cortex. These re-presentations support the processing of object-related structuralinformation, but are not sufficient for normal object perception.

Supplementary MaterialSupplementary material can be found at: http://www.cercor.oxfordjournals.org/.

FundingThis researchwas supported by grants from the National ScienceFoundation (BCS0923763; #SMA-1041755 to the Temporal Dy-namics of Learning Center: PI: G. Cottrell) and by the UnitedStates-Israel Binational Science Foundation (T-2013219 to EF).

NotesConflict of Interest: None declared.

ReferencesAllen HA, Humphreys GW. 2009. Direct tactile stimulation of dor-

sal occipito-temporal cortex in a visual agnosic. Curr Biol.19:1044–1049.

Behrmann M, Kimchi R. 2003. What does visual agnosia tell usabout perceptual organization and its relationship to objectperception? J Exp Psychol Hum Percept Perform. 29:19–42.

Behrmann M, Plaut DC. 2014. Bilateral hemispheric processing ofwords and faces: evidence from word impairments in proso-pagnosia and face impairments in pure alexia. Cereb Cortex.24:1102–1118.

Behrmann M, Williams P. 2007. Impairments in part–wholerepresentations of objects in two cases of integrative visualagnosia. Cogn Neuropsychol. 24:701–730.

Bettencourt KC, Xu Y. 2013. The role of transverse occipital sulcusin scene perception and its relationship to object individu-ation in inferior intraparietal sulcus. J Cogn Neurosci.25:1711–1722.

Booth MC, Rolls ET. 1998. View-invariant representations of fa-miliar objects by neurons in the inferior temporal visual cor-tex. Cereb Cortex. 8:510–523.

Crawford JR, Garthwaite PH. 2002. Investigation of the single casein neuropsychology: confidence limits on the abnormality oftest scores and test score differences. Neuropsychologia.40:1196–1208.

Crawford JR, Garthwaite PH. 2005. Testing for suspected impair-ments and dissociations in single-case studies in neuropsych-ology: evaluation of alternatives using Monte Carlo simulationsand revised tests for dissociations. Neuropsychology. 19:318–331.

Crawford JR, Garthwaite PH, Porter S. 2010. Point and interval esti-mates of effect sizes for the case-controls design in neuro-psychology: rationale, methods, implementations, andproposed reporting standards. CognNeuropsychol. 27:245–260.

Crawford JR, Garthwaite PH, Ryan K. 2011. Comparing a singlecase to a control sample: testing for neuropsychological defi-cits and dissociations in the presence of covariates. Cortex.47:1166–1178.

Creem SH, Proffitt DR. 2001. Grasping objects by their handles: anecessary interaction between cognition and action. J ExpPsychol Hum Percept Perform. 27:218.

Delvenne J-F, Seron X, Coyette F, Rossion B. 2004. Evidence forperceptual deficits in associative visual (prosop) agnosia: Asingle-case study. Neuropsychologia. 42:597–612.

Desimone R, Albright TD, Gross CG, Bruce C. 1984. Stimulus-se-lective properties of inferior temporal neurons in the ma-caque. J Neurosci. 4:2051–2062.

Di Luca M, Domini F, Caudek C. 2004. Spatial integration in struc-ture from motion. Vision Res. 44:3001–3013.

Ditunno PL, Mann VA. 1990. Right hemisphere specialization formental rotation in normals and brain damaged subjects.Cortex J Devoted Study Nerv Syst Behav. 26:177–188.

Durand JB, Peeters R, Norman JF, Todd JT, Orban GA. 2009. Parietalregions processing visual 3D shape extracted from disparity.Neuroimage. 46:1114–1126.

Faillenot I, Decety J, Jeannerod M. 1999. Human brain activity re-lated to the perception of spatial features of objects.NeuroImage. 10:114–124.

Farah MJ. 1994. Visual agnosia. 2nd ed. Cambridge (MA): MITPress/Bradford Books.

Farah MJ, Monheit MA, Wallace MA. 1991. Unconscious percep-tion of “extinguished” visual stimuli: reassessing the evi-dence. Neuropsychologia. 29:949–958.

Freud E, AvidanG, Ganel T. 2013. Holistic processing of impossibleobjects: evidence from Garner’s speeded-classification task.Vision Res. 93:10–18.

Freud E, Avidan G, Ganel T. 2015. The highs and lows of object im-possibility: effects of spatial frequency on holistic processingof impossible objects. Psychon Bull Rev. 22:297–306.

Freud E, Ganel T, Avidan G. 2013. Representation of possible andimpossible objects in the human visual cortex: Evidence fromfMRI adaptation. NeuroImage. 64:685–692.

Freud E, Ganel T, Avidan G. 2015. Impossible expectations: fMRIadaptation in the lateral occipital complex (LOC) ismodulatedby the statistical regularities of 3D structural information.NeuroImage. 122:188–194.

Freud E, Hadad B, Avidan G, Ganel T. 2015. Evidence for similarearly but not late representation of possible and impossibleobjects. Name Front Psychol. 6:94.

Freud E, Rosenthal G, Ganel T, AvidanG. 2015. Sensitivity to objectimpossibility in the human visual cortex: evidence from func-tional connectivity. J Cogn Neurosci. 27:1029–1043.

Gauthier I, BehrmannM, Tarr MJ. 1999. Can face recognition real-ly be dissociated from object recognition? J Cogn Neurosci.11:349–370.

Georgieva S, Peeters R, Kolster H, Todd JT, Orban GA. 2009. Theprocessing of three-dimensional shape from disparity in thehuman brain. J Neurosci. 29:727–742.

Gilaie-Dotan S, Saygin AP, Lorenzi LJ, Egan R, Rees G,Behrmann M. 2013. The role of human ventral visual cortexin motion perception. Brain. 136:2784–2798.

Gilaie-Dotan S, Saygin AP, Lorenzi LJ, Rees G, Behrmann M. 2015.Ventral aspect of the visual formpathway is not critical for theperception of biological motion. Proc Natl Acad Sci. 112:E361–E370.

Goodale MA, Milner AD. 1992. Separate visual pathways for per-ception and action. Trends Neurosci. 15:20–25.

Goodale MA, Milner AD, Jakobson LS, Carey DP. 1991. A neuro-logical dissociation between perceiving objects and graspingthem. Nature. 349:154–156.

Grill-Spector K, Kushnir T, Edelman S, Avidan G, Itzchak Y,Malach R. 1999. Differential processing of objects under

12 | Cerebral Cortex

at Acquisitions D

eptHunt L

ibrary on October 22, 2015

http://cercor.oxfordjournals.org/D

ownloaded from

various viewing conditions in the human lateral occipitalcomplex. Neuron. 24:187–203.

Gross CG, Rocha-Miranda CE, Bender DB. 1972. Visual propertiesof neurons in inferotemporal cortex of the Macaque.J Neurophysiol. 35:96–111.

Harris IM,Miniussi C. 2003. Parietal lobe contribution tomental ro-tation demonstrated with rTMS. J Cogn Neurosci. 15:315–323.

Hasson U, Harel M, Levy I, Malach R. 2003. Large-scale mirror-symmetry organization of human occipito-temporal objectareas. Neuron. 37:1027–1041.

HelbigHB, GrafM, KieferM. 2006. The role of action representationsin visual object recognition. Exp Brain Res. 174:221–228.

James TW, Culham J, Humphrey GK, Milner AD, Goodale MA. 2003.Ventral occipital lesions impair object recognition but not ob-ject-directed grasping: an fMRI study. Brain. 126(11):2463–2475.

Jarmasz J, Hollands JG. 2009. Confidence intervals in repeated-measures designs: The number of observations principle.Can J Exp Psychol. 63:124–138.

Karnath H-O, Rüter J, Mandler A, Himmelbach M. 2009. The anat-omy of object recognition—visual form agnosia caused bymedial occipitotemporal stroke. J Neurosci. 29:5854–5862.

Kiefer M, Sim E-J, Helbig H, Graf M. 2011. Tracking the time course ofaction priming on object recognition: evidence for fast and slowinfluencesof actiononperception. JCognNeurosci. 23:1864–1874.

Konen CS, BehrmannM, NishimuraM, Kastner S. 2011. The func-tional neuroanatomy of object agnosia: a case study. Neuron.71:49–60.

Konen CS, Kastner S. 2008. Two hierarchically organized neuralsystems for object information in human visual cortex. NatNeurosci. 11:224–231.

Kourtzi Z, Kanwisher N. 2000. Cortical regions involved in per-ceiving object shape. J Neurosci. 20:3310–3318.

Malach R, Reppas JB, Benson RR, Kwong KK, Jiang H, KennedyWA,Ledden PJ, Brady TJ, Rosen BR, Tootell RB. 1995. Object-relatedactivity revealed by functional magnetic resonance imaging inhuman occipital cortex. Proc Natl Acad Sci USA. 92:8135–8139.

Mapelli D, Behrmann M. 1997. The role of color in object recogni-tion: Evidence from visual agnosia. Neurocase. 3:237–247.

Marotta JJ, Genovese CR, Behrmann M. 2001. A functional MRIstudy of face recognition in patients with prosopagnosia.Neuroreport. 12:1581–1587.

Marotta JJ, McKeeff TJ, Behrmann M. 2002. The effects of rotationand inversion on face processing in prosopagnosia. CognNeuropsychol. 19:31–47.

Moore C, Engel SA. 2001. Neural response to perception of volumein the lateral occipital complex. Neuron. 29:277–286.

Nishimura M, Doyle J, Humphreys K, Behrmann M. 2010.Probing the face-space of individuals with prosopagnosia.Neuropsychologia. 48:1828–1841.

Orban GA. 2011. The extraction of 3D shape in the visual systemof human and nonhuman primates. Annu Rev Neurosci.34:361–388.

Penrose LS, Penrose R. 1958. Impossible objects: a special type ofvisual illusion. Br J Psychol. 49:31–33.

Regolin L, Rugani R, Stancher G, Vallortigara G. 2011. Spontan-eous discrimination of possible and impossible objects bynewly hatched chicks. Biol Lett. 7:654–657.

Riddoch MJ, Humphreys GW, Blott W, Hardy E, Smith AD,Smith AD. 2003. Visual and spatial short-term memory in in-tegrative agnosia. Cogn Neuropsychol. 20:641–671.

Rosenthal O, Behrmann M. 2006. Acquiring long-term represen-tations of visual classes following extensive extrastriate dam-age. Neuropsychologia. 44:799–815.

Schacter DL, Cooper LA, Delaney SM. 1990. Implicit memory forunfamiliar objects depends on access to structural descrip-tions. J Exp Psychol Gen. 119:5–24.

Shuwairi SM, Albert MK, Johnson SP. 2007. Discrimination of pos-sible and impossible objects in infancy. Psychol Sci.18:303–307.

Sim E-J, Helbig HB, Graf M, Kiefer M. 2015. When action observa-tion facilitates visual perception: activation in visuo-motorareas contributes to object recognition. Cereb Cortex.25:2907–2918.

Soldan A, Hilton HJ, Stern Y. 2009. Bias effects in the possible/im-possible object decision test with matching objects. MemCogn. 37:235–247.

Srivastava S, Orban GA, DeMaziere PA, Janssen P. 2009. A distinctrepresentation of three-dimensional shape in macaque an-terior intraparietal area: fast, metric, and coarse. J Neurosci.29:10613–10626.

Troiani V, Stigliani A, Smith ME, Epstein RA. 2014. Multiple objectproperties drive scene-selective regions. Cereb Cortex.24:883–897.

Turnbull OH, Driver J, McCarthy RA. 2004. 2D but not 3D: pictor-ial-depth deficits in a case of visual agnosia. Cortex.40:723–738.

Ungerleider LG, Mishkin M. 1982. Two cortical visual sys-tems. In: Ingle DJ, Goodale MA, Mansfield RJW, editors.Analysis of visual behavior. Cambridge (MA): MIT Press.p. 549–586.

Vecera SP, Behrmann M. 1997. Spatial attention does not requirepreattentive grouping. Neuropsychology. 11:30–43.

Verhoef B-E, Michelet P, Vogels R, Janssen P. 2015. Choice-relatedactivity in the anterior intraparietal area during 3-D structurecategorization. J Cogn Neurosci. 27:1104–1115.

Xu Y. 2009. Distinctive neural mechanisms supporting visualobject individuation and identification. J Cogn Neurosci.21:511–518.

Yeatman JD,Weiner KS, Pestilli F, RokemA, Mezer A,Wandell BA.2014. The vertical occipital fasciculus: a century of contro-versy resolved by in vivo measurements. Proc Natl Acad Sci.111:E5214–E5223.

Zachariou V, Klatzky R, Behrmann M. 2014. Ventral and dorsalvisual stream contributions to the perception of objectshape and object location. J Cogn Neurosci. 26:189–209.

Zacks JM, Vettel JM, Michelon P. 2003. Imagined viewer andobject rotations dissociated with event-related fMRI. J CognNeurosci. 15:1002–1018.

Dissociable Dorsal Object Representations Freud et al. | 13

at Acquisitions D

eptHunt L

ibrary on October 22, 2015

http://cercor.oxfordjournals.org/D

ownloaded from