1
What is the Nature of the Perceptual Deficit in Congenital Prosopagnosia? Irving Biederman 1,2 *, Eshed Margalit 2,4 , Rafael S. Maarek 3 , Emily X. Meschke 2 , Bryan E. Shilowich 1 , Jordan J. Juarez 2 , Catrina M. Hacker 2 , Taylor J. Seamans 2 , Sarah B. Herald 5 1 Department of Psychology 2 Neuroscience Program 3 Department of Biomedical Engineering, University of Southern California, 4 Neuroscience Program, Stanford University, 5 Department of Psychological and Brain Sciences, Dartmouth College Image Understanding Lab Is there a perceptual deficit in congenital prosopagnosia (CP)? If so, what is its nature? The standard tests for diagnosing prosopagnosia, e.g., CFMT, PI20, Celebrity, etc., document deficits in face recognition, but they leave open whether the deficit is perceptual as opposed to, for example, memory. The Benton, a match-to-sample test that has served as the standard diagnostic instrument for almost three decades, allows the subject to exploit small local features, e.g., the pattern of hairs in the eyebrow, thus enabling some prosopagnosics to score normally on the test (Duchaine & Nakayama, 2006). References Duchaine, B., & Nakayama, K. (2006). The Cambridge Face Memory Test ... . Neuropsychologia, 44, 576-585. Yue, X., Tjan, B.S., & Biederman, I. (2006). What makes faces special? Vision Research, 46, 3802-3811. Xu, X., Biederman, I., & Shah, M. S. (2014). A neurocomputational account of the face configural effect. Journal of Vision, 14, 1-9. Assessing the perceptual sensitivity of CPs and Controls to small metric differences in Faces and Blobs In these three trials, which of the bottom two images is an exact match to the top image? Similarity of distractors for Faces and Blobs were scaled to be equally dissimilar according to the Gabor Jet model. Sample trials with blob stimuli (left) and face stimuli (right). Subjects were instructed to indicate via button press which of the bottom two images was an exact match to the top image. Correct answer: Left Correct answer: Left Correct answer: Right CONCLUSIONS 1. CPs are deficient in their perceptual discrimination of small differences in complex, smoothly curved, metrically-varying stimuli such as Faces and Blobs. 2. This deficit does not extend to the discrimination of metric differences of the same magnitude in simple geometric volumes, suggesting a selective impairment in the discrimination of biologically plausible stimuli. SPECULATION The deficit in metric discrimination of complex, biologically appearing shapes by CPs may be a symptom rather than the underlying cause of face recognition deficits. Yue et al. (2006) presented evidence that FFA retained the spatial (Gabor-like) tuning characteristic of earlier visual stages. Xu et al. (2014) showed that face configural effects were produced by large r.f.s. which served to magnify the impact of small metric differences; a local difference would affect the activation of kernals with r.f.s. centered all over the face. Thus the core deficit in congenital prosopagnosia may be an inability to benefit from the configural effects—with their attendant magnification of small differences—produced by large receptive fields in face-selective areas. Results: Even though no information was conveyed by either the presence of a head or the ordering of the parts, enclosing the parts within a head and having the parts in a normal order resulted in a significant reduction in RTs. The presence of a head did not facilitate RTs when the face parts were inverted. Preliminary results from two prosopagnosics show no evidence of these configural effects— neither the benefit of a head nor an upright order of face parts. 1. With upright parts, embedding the parts in a head reduced RTs by 300 msec! 2. Inverting the order of the parts (but not the parts themselves), increased RTs by 210 msec. 3. There was no benefit of embedding the parts within a head when the parts were in inverted order. 4. Accuracy for the four conditions was high (between 91-93%) with slight advantages for the presence of a head (.5%) and an upright order (1.5%). Do CPs show face configural effects? We devised an odd-man-out task that reflects configural face effects but does not require identification or matching of a previously learned face (which puts CPs at a disadvantage in the standard condition as they have difficulty in storing faces). Subjects viewed a display with three “faces,” arranged diagonally, with each face composed of three face parts: eyes, nose, and mouth. In each trial, one of the faces differed subtly in the shape of one of its parts. The task was to detect which face was the “odd man out”: left, middle, or right. On half the trials, the three sets of face parts were each enclosed within a head (absent on the other half). Additionally, on half the trials, the parts were in the normal upright order (eyes on top) and, on the other half, in an inverted order (with the mouth on top). The individual parts themselves were never inverted. The configural effect is taken as an advantage of enclosing the parts by the head and the parts being in their normal ordering. Correct answer: Middle Correct answer: Right Correct answer: Left Correct answer: Right Do CPs show a deficit when discriminating metric differences between simple shapes (i.e., geons)? NO There was no difference between CPs and Controls in discriminating metric differences between geons of equal similarity as the faces and blobs. Controls (light blue lines) made fewer errors than DPs (dark blue lines) on the Face trials, p < .001, and on the Blob trials, p < 0.001. The higher error rates of CPs on both Faces and Blobs suggest that the deficit in DP may not be restricted to faces. Blobs Faces Controls Controls CPs CPs CPs are deficient in discriminating both Faces and Blobs Note: Error bars are sometimes encompassed within the data points. Ordering of Parts Does performance on the Match-to-Sample Face task correlate with standard measures of prosopagnosia? YES. Even though the M2S face discrimination task looks nothing like the standard tests for assessing prosopagnosia, the correlations with these measures are as high (and typically higher) than any other measure. Correlations of Error Rates on the M2S Face task with: PI20: -.61*** (df = 50) CFMT: .64*** (df = 52) Faceblind.org: .46*** (df = 53) Doppelgänger Discrimination: .65*** (df = 31) USC IUL Celebrity Faces: .51*** (df = 54) It is somewhat remarkable that a minimal Match-to-Sample task for discriminating a small, metric difference between two faces, with no requirements for memory or view invariance, accurately predicts responses to items on the PI20 such as ”Anxiety about face recognition has led me to avoid certain social or professional situations.” *** = p < .001 Correct Answer: Left Acknowledgments Supported by NSF BCS 0617699 and the Dornsife Research Fund. [email protected] http://geon.usc.edu/ C CPs Scan to download online copy of poster!

What is the Nature of the Perceptual Deficit in Congenital ...geon.usc.edu/~biederman/presentations/PerceptualDeficit_VSS_17.pdf · What is the Nature of the Perceptual Deficit in

Embed Size (px)

Citation preview

What is the Nature of the Perceptual Deficit in Congenital Prosopagnosia?IrvingBiederman1,2*,EshedMargalit2,4,RafaelS.Maarek3,EmilyX.Meschke2,BryanE.Shilowich1,

JordanJ.Juarez2,CatrinaM.Hacker2,TaylorJ.Seamans2,SarahB.Herald51DepartmentofPsychology2NeuroscienceProgram 3DepartmentofBiomedicalEngineering,UniversityofSouthernCalifornia,4NeuroscienceProgram,StanfordUniversity,5DepartmentofPsychologicalandBrainSciences,DartmouthCollege

Image Understanding Lab

Isthereaperceptualdeficitincongenitalprosopagnosia(CP)?Ifso,whatisitsnature?

Thestandardtestsfordiagnosingprosopagnosia,e.g.,CFMT,PI20,Celebrity,etc.,documentdeficitsinfacerecognition,buttheyleaveopenwhetherthedeficitisperceptualasopposedto,forexample,memory.TheBenton,amatch-to-sampletestthathasservedasthestandarddiagnosticinstrumentforalmostthreedecades,allowsthesubjecttoexploitsmalllocalfeatures,e.g.,thepatternofhairsintheeyebrow,thusenablingsomeprosopagnosicstoscorenormallyonthetest(Duchaine&Nakayama,2006).

ReferencesDuchaine,B.,&Nakayama,K.(2006).TheCambridgeFaceMemoryTest....Neuropsychologia,44,576-585.Yue,X.,Tjan,B.S.,&Biederman,I.(2006).Whatmakesfacesspecial?VisionResearch,46,3802-3811.Xu,X.,Biederman,I.,&Shah,M.S.(2014).Aneurocomputational accountofthefaceconfiguraleffect. JournalofVision,14,1-9.

AssessingtheperceptualsensitivityofCPsandControlstosmallmetricdifferencesinFacesandBlobs

Inthesethreetrials,whichofthebottomtwoimagesisanexactmatchtothetopimage?

SimilarityofdistractorsforFacesandBlobswerescaledtobeequallydissimilaraccordingtotheGaborJetmodel.

Sampletrialswithblobstimuli(left)andfacestimuli(right).Subjectswereinstructedtoindicateviabuttonpresswhichofthebottomtwoimageswasanexactmatchtothetopimage.Correctanswer:Left Correctanswer:Left Correctanswer:Right

CONCLUSIONS1. CPsaredeficientintheirperceptualdiscriminationofsmalldifferencesin

complex,smoothlycurved,metrically-varyingstimulisuchasFacesandBlobs.2. Thisdeficitdoesnotextendtothediscriminationofmetricdifferencesofthe

samemagnitudeinsimplegeometricvolumes,suggestingaselectiveimpairmentinthediscriminationofbiologicallyplausiblestimuli.

SPECULATIONThedeficitinmetricdiscriminationofcomplex,biologicallyappearingshapesbyCPsmaybeasymptom ratherthantheunderlyingcauseoffacerecognitiondeficits.Yueetal.(2006)presentedevidencethatFFAretainedthespatial(Gabor-like)tuningcharacteristicofearliervisualstages.Xuetal.(2014)showedthatfaceconfiguraleffectswereproducedbylarger.f.s.whichservedtomagnifytheimpactofsmallmetricdifferences;alocaldifferencewouldaffecttheactivationofkernalswithr.f.s.centeredallovertheface.Thusthecoredeficitincongenitalprosopagnosiamaybeaninabilitytobenefitfromtheconfiguraleffects—withtheirattendantmagnificationofsmalldifferences—producedbylargereceptivefieldsinface-selectiveareas.

Results:

Eventhoughnoinformationwasconveyedbyeitherthepresenceofaheadortheorderingoftheparts,enclosingthepartswithinaheadandhavingthepartsinanormalorderresultedinasignificantreductioninRTs.ThepresenceofaheaddidnotfacilitateRTswhenthefacepartswereinverted.Preliminaryresultsfromtwoprosopagnosicsshownoevidenceoftheseconfiguraleffects—neitherthebenefitofaheadnoranuprightorderoffaceparts.

1. Withuprightparts,embeddingthepartsinaheadreducedRTsby300msec!

2. Invertingtheorderoftheparts(butnotthepartsthemselves),increasedRTsby210msec.

3. Therewasnobenefitofembeddingthepartswithinaheadwhenthepartswereininvertedorder.

4. Accuracyforthefourconditionswashigh(between91-93%)withslightadvantagesforthepresenceofahead(.5%)andanuprightorder(1.5%).

DoCPsshowfaceconfiguraleffects?Wedevisedanodd-man-outtaskthatreflectsconfiguralfaceeffectsbutdoesnotrequireidentificationormatchingofapreviouslylearnedface(whichputsCPsatadisadvantageinthestandardconditionastheyhavedifficultyinstoringfaces).

Subjectsviewedadisplaywiththree“faces,”arrangeddiagonally,witheachfacecomposedofthreefaceparts:eyes,nose,andmouth.Ineachtrial,oneofthefacesdifferedsubtlyintheshapeofone ofitsparts.Thetaskwastodetectwhichfacewasthe“oddmanout”:left,middle,orright.Onhalfthetrials,thethreesetsoffacepartswereeachenclosedwithinahead(absentontheotherhalf).Additionally,onhalfthetrials,thepartswereinthenormaluprightorder(eyesontop)and,ontheotherhalf,inaninvertedorder(withthemouthontop).Theindividualpartsthemselveswereneverinverted.Theconfiguraleffectistakenasanadvantageofenclosingthepartsbytheheadandthepartsbeingintheirnormalordering.

Correctanswer:Middle Correctanswer:RightCorrectanswer:Left Correctanswer:Right

DoCPsshowadeficitwhendiscriminatingmetricdifferencesbetweensimpleshapes(i.e.,geons)?NO

TherewasnodifferencebetweenCPsandControlsindiscriminatingmetricdifferencesbetweengeonsofequalsimilarityasthefacesandblobs.

Controls(lightbluelines)madefewererrorsthanDPs(darkbluelines)ontheFacetrials,p<.001,andontheBlobtrials,p<0.001.

ThehighererrorratesofCPsonbothFacesandBlobssuggestthatthedeficitinDPmaynotbe

restrictedtofaces.

Blobs

Faces

Controls

Controls

CPs

CPs

CPsaredeficientindiscriminatingboth FacesandBlobsNote:Errorbarsaresometimesencompassedwithinthedatapoints.

OrderingofParts

DoesperformanceontheMatch-to-SampleFacetaskcorrelatewithstandardmeasuresofprosopagnosia?

YES.EventhoughtheM2Sfacediscriminationtasklooksnothinglikethestandardtestsforassessingprosopagnosia,thecorrelationswiththesemeasuresareashigh(andtypicallyhigher)thananyothermeasure.

CorrelationsofErrorRatesontheM2SFacetaskwith:PI20: -.61***(df=50)CFMT: .64***(df=52)Faceblind.org: .46***(df=53)DoppelgängerDiscrimination: .65***(df=31)USCIULCelebrityFaces: .51***(df =54)

ItissomewhatremarkablethataminimalMatch-to-Sampletaskfordiscriminatingasmall,metricdifferencebetweentwofaces,withnorequirementsformemoryorviewinvariance,accuratelypredictsresponsestoitemsonthePI20suchas”Anxietyaboutfacerecognitionhasledmetoavoidcertainsocialorprofessionalsituations.”

***=p<.001

CorrectAnswer:Left

AcknowledgmentsSupportedbyNSFBCS0617699andtheDornsifeResearchFund.

[email protected]://geon.usc.edu/

C

CPs

Scantodownload

onlinecopyofposter!