Perceptual learning modulates electrophysiological and psychophysical response to visual texture segmentation in humans

Perceptual learning modulates electrophysiological and psychophysical response to visual texture segmentation in humans
of 6
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
  Neuroscience Letters 371 (2004) 18–23 Perceptual learning modulates electrophysiological and psychophysicalresponse to visual texture segmentation in humans Clara Casco a , ∗ , Gianluca Campana a , Alba Grieco a , Giorgio Fuggetta b a  Dipartimento di Psicologia Generale, Universit`a di Padova, Via Venezia, 8, 35131 Padova, Italy b  Dipartimento di Scienze Neurologiche e della Visione, Universit`a di Verona, Italy Received 14 April 2004; received in revised form 2 July 2004; accepted 2 August 2004 Abstract Weinvestigatedthemechanismsthatallow,viaperceptuallearning,selectivemodulationofavisualline-texturefiguresaliencyinaccordancewithtaskrelevance.Learning-dependentsaliencyincreasewasinferredbyincreasedaccuracyinorientationdiscriminationwithtaskrepetition.As a result of learning, accuracy increase was more pronounced when local and global orientation of the texture figure conflicted, and reachedceiling in both conflict and conflict-free conditions. This psychophysical effect was associated with a decrease in amplitude of negative VEPcomponents in the configurations where global and local orientation conflicted, and to a weak increase of VEP’s earliest negative componentin the conflict-free condition. The VEP result is a direct demonstration that learning, in addition to increasing response of relevant channels,also reduces the weight of channels whose receptive field size and orientation tuning conflict with the task.© 2004 Elsevier Ireland Ltd. All rights reserved. Keywords:  Perceptual learning; Textures; VEPs; Psychophysics Little is known about the mechanisms that allow the visualsystem to increase the salience of behaviourally importantstimuli and to decrease salience of irrelevant stimuli. Somephysiological evidence in both audition [13] and vision [6] indicates that receptive field tuning can be modulated via aHebbian learning mechanism. In line with these studies, weexplored the mechanisms that allow implicit processing of visual information in humans to be modulated by perceptuallearning according to task relevance. The processes inves-tigated are those involved in figure-ground segmentation of line-texture stimuli, on the basis of local orientation contrastor global orientation. This is a well-suited stimulus for as-sessing how saliency is modulated by learning, because ithas been shown that saliency of local and global orientationdepends on orientation contrast [18] and is increased by per-ceptual learning [12]. Here we asked how global and localprocesses are modulated by learning depending on task rele-vance.InaccordancewithHebb’srule[10],learningproduces ∗ Corresponding author. Tel.: +39 0498276611; fax: +39 0498276600.  E-mail address: (C. Casco). long-lasting regulation of visual responses, either excitatoryor inhibitory, and furthermore, it has been proposed [1] thatlearning can have the effect of increasing or decreasing thestrengthofinhibitoryconnectionsamongstchannels.DosherandLu[5]suggestthatperceptuallearningmightreflectplas-ticityintherelativeactivityofdifferentvisualchannelstunedto orientation and spatial frequency. Both facilitatory [5,8]and inhibitory [16] effects of learning are well-documentedpsychophysically,butnodirectelectrophysiologicalevidenceof inhibitory effect of learning has yet been provided.Here we asked not only whether saliency is increased bylearning from trial repetition, but also whether an increase insaliency is associated with an increase in amplitude of cor-tical response (VEPs) to relevant texture features and with adecrease for irrelevant ones. The effect of saliency increaseduetotrialrepetitionwaspsychophysicallyassessedbymea-suring accuracy in an orientation discrimination task, eitherglobal or local, as a function of block number. Indeed, thesaliency of a single or group of elements emerging froman array of texture elements can be psychophysically as-sessed either directly from subjective reports, or indirectly 0304-3940/$ – see front matter © 2004 Elsevier Ireland Ltd. All rights reserved.doi:10.1016/j.neulet.2004.08.005  C. Casco et al. / Neuroscience Letters 371 (2004) 18–23  19Fig. 1. Stimuli. Texture consisted of white vertical line elements 19 ′ longarranged on a diamond raster, with raster step of 30.5 ′ and jittered aroundtheir raster centre by 0–2.7 ′ . The segmented texture bar comprised 6  ×  24line elements tilted 45 ◦ clockwise or counterclockwise at random. In theparallel condition (a) the bar had the same orientation as its line elements,while in the orthogonal configuration (b) it had opposite orientation. by measuring speed and accuracy of detection or discrimina-tion [18].We based the assessment of neural correlates of the ef-fect of perceptual learning on the evidence that variationsin D-VEP amplitudes (obtained as the difference betweenuniform-andsegmentation-textureVEPs)arespecificallyas-sociated with texture segmentation [12,14].Observers viewed line-texture bars, oriented in either thesame (parallel, Fig. 1a) or opposite (orthogonal, Fig. 1b) di- rection to their elements segregating out from a uniform tex-ture background. Separate groups performed a 2AFC task in which they had to discriminate either the global (bar) orlocal (elements) orientation. When parallel, both global andlocal orientations are relevant to task, whereas when orthog-onal either local or global conflict, depending on the task.Perceptual learning was inferred from block-dependent vari-ation of both D-VEP early components and percentage cor-rect responses for the various blocks of trials in independentgroups. To analyse the effect of perceptual learning, we firstconsidered the possibility that observers were not blind toorientation conflicting with that explicitly judged (either lo-cal or global). This was indeed the case, since accuracy waslowerwhentextureelementswereorthogonal(conflicting)toglobalbar,aneffectthatdisappearedwithstimulusrepetition.We then asked whether the improvement in accuracy withstimulus repetition was associated with a modulation of therelated D-VEP amplitude and whether this modulation wasin the same or in a different direction in the two conditions.Our results showed that increased accuracy with perceptuallearning was associated with opposite electrophysiologicaleffectsdependingontheconfiguration:anamplitudeincreasein conflict-free configurations, suggesting facilitation, and adecrease in the conflict configuration, suggesting inhibitionof mechanisms with receptive field size and orientation tun-ing conflicting with the task.Fourgroupsof10observers,aged20–30years,selectedonthe basis of absence of astigmatism, participated voluntarilyintheexperimentsafterapprovalbythehumansubjectreviewboard of the University of Padova and in accordance with theHelsinki Declaration.Different groups of observers had to be used in differenttasks of both psychophysical and VEP experiments to re-duce possible artefacts due to generalisation of learning [2].Two-way ANOVA, with block and stimulus configurationsas main factors were used to analyse the psychophysical andelectrophysiological data.StimuliweregeneratedbyaPC,displayedona15 ′′ colourmonitor (70Hz vertical refresh) viewed through a 16 ◦ diam-eter circular aperture from a distance of 57cm in a dark-ened room. The resolution of the monitor was 640  ×  480with square pixel 2.7  ×  2.7arc min. Textures were com-posedofwhitelineelements,presentedonadarkbackground(0.6cd/m 2 ). The look-up table was set in such a way that thespace average luminance of the texture was matched for bothvertical (11.45cd/m 2 ), and 45 ◦ and 135 ◦ (11.51cd/m 2 ) ori-entation of the texture line elements.In each block of trials, four stimulus conditions were ran-domly interleaved: bar-oriented right (parallel or orthogonalto its elements) or left (parallel or orthogonal).The psychophysical experiments consisted of five blocksof 64 trials each, performed consecutively during the samesession.Eachtrialconsistedofathree-framesequence,whichwasperceivedasatexturebarappearingordisappearingfromthe uniform texture background. Frame duration was 26ms(with 2977ms interval).Observersperformeda2AFCtaskeverytrialbyusingtwokeys. An acoustic feedback was given for errors and omis-sions. Performance was defined as percent correct responses.The VEP experiment consisted of two blocks of 200 tri-als. The two blocks were performed consecutively duringthe same session. Onset–offset stimulation consisted of thecyclical alternation of segmented and uniform texture, eachpresentedfor840mswithnointerval,whichwasperceivedasa texture bar cyclically appearing and disappearing from theuniform texture background. Note that these long exposures  20  C. Casco et al. / Neuroscience Letters 371 (2004) 18–23 Fig.2. Trialevents.Eventsinasingletrialconsistedinthealternationofthesegmented(parallelororthogonalbar,tiltedclockwiseorcounterclockwise)and uniform texture. During the blank interval between trials, presented ei-ther every trial (in the psychophysical experiment) or every three trials onaverage (in the VEP experiments), the subject was asked to discriminate ei-therthelocalorientationgradientortheglobalbarwhilemaintainingfixationon a central dot. allowed to obtain VEPs with sufficiently large amplitudes,so that all components (included those related to texture seg-mentation) could be easily seen on VEP traces, if present [3].However, such long exposure times were inadequate for psy-chophysical testing since they produced ceiling performancein both configurations and tasks.Individual D-VEPs were determined by algebraic sub-traction of the background-stimulus VEP from eachsegmentation-stimulus VEP; the four D-VEP peaks (N1, P1,N2, N3) were identified as the largest peak within the ap-propriate temporal window (60–90, 91–120, 121–160, and161–240ms) (Fig. 2). After presentation of three texture bar-uniformtexturepairsonaverage,theonset–offsetstimulationwas suspended and the monitor remained dark until responsewasgiven.Theobserver’sresponsere-startedtheonset–offsetstimulation,after2000msofauniformtexturedisplaytopre-pare fixation.The electroencephalogram (EEG) was recorded fromAg/AgCl-coated cup electrodes placed at Oz and left (ref-erence) and right (ground) earlobes, in accordance with theinternational 10/20 system. Electrode impedance was heldbelow 5k   . The EEG was amplified (BM 623) and digitallyconverted (CED 1401) under control of a second PC. Stimu-lation and recording onset were synchronised using the ver-tical retrace signal of the monitor that displaced the stimulus.The EEG was amplified 50,000 times, bandpass filtered at1–50Hz, sampled at 1kHz with a resolution of 12 bits, andstoredonharddisk.Artefactrejectionwascarriedoutoff-linewhen the signal amplitude exceeded 100  V. Fig. 3. Orientation discrimination accuracy as a function of block in globalpsychophysical task, separately for parallel and orthogonal configurations. The VEPs were obtained by averaging the signal sepa-rately for the three stimuli: background, bar parallel, and barorthogonal to its elements, and were then vertically alignedby taking their mean amplitude in the 0–50 range after stim-ulus onset as baseline.We recorded VEPs while eye movements were registeredthrough an independent channel, by placing electrodes onthe temporal side of both eyes, at 1cm from eye edge. Acalibration procedure was set so that, on average, any eyemovement along the horizontal axis equal to 1.4 ◦ or to 2.8 ◦ was discarded 50% or 100% of the trials, respectively. Thepercentage of trials with eye movements was less than 4% ineach subject.Fig.3showsresults,plottingpercentcorrectinglobalpsy-chophysicaltaskasafunctionofblockinparallelandorthog-onal configurations. Performance is seen to be significantlybetter for parallel than for orthogonal bar ( P  < 0.001). Thisindicates that irrelevant local information was processed andinterfered with relevant global orientation explicitly judged.Regression lines fitted to individual accuracy data had slopeslargerthanzero,indicatinganimprovementofaccuracywithperceptuallearninginbothconfigurations.Thissuggeststhatsaliency generally increases as a result of learning. The find-ing that configural effect was significant in the first four of five blocks but not in the last, as well as the finding that indi-vidual regression line slopes (indicating learning-dependentimprovement)weresignificantlylarger( P <0.01)fororthog-onal (10.5) than for parallel (3.96) configuration, indicatesthat learning increased performance more for orthogonal barso that accuracy, which reflected saliency, equated in the lastblock.IndividualD-VEPswereaveragedacross-subjecttoobtainmean D-VEPs. Mean D-VEPs are shown separately for thefirstandsecondblockforbothparallel(Fig.4a)andorthogo-nal (Fig. 4b) configurations. Across-subject averaging of in-dividual D-VEPs was done after shifting them by an amountequal to (latency(N3) i  – latency(N3) mean ) so that the latencyof the N3 peak coincided with average latency. Shifting wasallowed given a non-significant differences in latency in ei-ther the main variables or the interactions and allowed meanD-VEPs graphically to show (Fig. 4) N3 amplitude corre-spondingtothatresultingfromdescriptivestatistics(Table1).  C. Casco et al. / Neuroscience Letters 371 (2004) 18–23  21Fig. 4. Mean D-VEPs in blocks 1 and 2 of parallel (a) and orthogonal (b)conditions in the global orientation task. Meanamplitudesofeachcomponentforeachblockandcon-figuration are shown in Table 1.Our results indicate that perceptual learning affected theamplitudeofearlyD-VEPcomponents(N1andP1)whenthebar was parallel: N1 becomes more negative ( P  < 0.05) andP1 less positive ( P  = 0.02). The effect of P1 is likely to de-pend on response preparation whereas the N1 effect reflectselectrophysiological correlates of (global) texture segmen-tation [3,14]. The important D-VEP result is that, throughlearning (from block 1 to 2), N2 and N3 amplitude presenteda reduction for orthogonal bar only ( P  < 0.005), so that theparallel–orthogonaldifferenceinN3componentbecamesig-nificant in block 2. Since psychophysical and D-VEP effectsover time are assessed with a similar number of trials, we Table 1Mean (S.E.) D-VEP amplitudes in N1, P1, N2 and N3, for parallel andorthogonal bars for blocks 1 and 2 in the global task, are shownGlobal task 1st block 2nd block Parallel Orthogonal Parallel OrthogonalMean S.E. Mean S.E. Mean S.E. Mean S.E.N1  − 0 . 4 0.4  − 1 0.6  − 1 . 2 0.3  − 0 . 6 0.3P1 1 . 3 0.5 0 . 81 0.6 0 . 34 0.5 10 . 6 0.7N2  − 4 0.9  − 3 . 9 1.1  − 5 . 1 0.9  − 3 . 8 0.8N3  − 6 . 5 1.0  − 6 . 2 0.98  − 5 . 8 1.0  − 3 . 8 0.7Fig. 5. Orientation discrimination. Accuracy in local task is shown as afunction of block number is shown separately for parallel and orthogonalconfigurations. can conclude that learning-dependent improvement in psy-chophysical response to texture bar is associated with an in-crease of D-VEP amplitude for parallel bar, but with a de-crease for orthogonal bar.Results referring to local psychophysical task are shownin Fig. 5, as percent correct as a function of block in paral-lel and orthogonal configurations. These results indicate thatperformance was significantly better for parallel bar than fororthogonal bar ( P  < 0.02), suggesting that global orienta-tion was also processed, and interfered with local orientationdiscrimination. Regression lines fitted to individual accuracydata had slopes different from zero, indicating improvementofaccuracyinbothconfigurationsasaresultoflearning.Theparallel–orthogonal difference in block 1 only ( P  < 0.0001),as well as regression line slopes larger ( P  < 0.005) for or-thogonal (5.5) than for parallel configuration (2.7), indicatethat improvement through learning was larger for orthogonalbar than for parallel bar so that accuracy equated from thesecond block.Mean D-VEPs, separately for the first and the secondblock, are shown for both parallel (Fig. 6a) and orthogonal(Fig.6b)configurations.Meanamplitudeforeachcomponentand each of the two blocks and configurations are shown inTable2.AsFig.6shows,perceptuallearningdecreasedearly D-VEP amplitude, but only when the texture bar was orthog-onal to its elements. Indeed, in block 2 but not block 1, N1,N2 and N3 amplitudes were significantly smaller ( P  < 0.05)in the orthogonal condition. These results indicate that, as in Table 2Mean (S.E.) D-VEP amplitudes for N1, P1, N2 and N3, for parallel andorthogonal bars and for blocks 1 and 2 in the local task, are shownLocal task 1st block 2nd block Parallel Orthogonal Parallel OrthogonalMean S.E. Mean S.E. Mean S.E. Mean S.E.N1  − 1 . 9 0.6  − 2 . 7 0.7  − 2 . 2 0.9  − 1 . 1 0.8P1 2 . 6 1.2 2 . 5 1.3 2 . 1 1.2 2 . 7 1.2N2  − 6 . 6 1.3  − 6 . 3 1.3  − 6 . 6 1.2  − 5 . 3 1.0N3  − 7 . 7 1.4  − 8 . 4 1.3  − 8 . 3 1.1  − 7 . 1 0.9  22  C. Casco et al. / Neuroscience Letters 371 (2004) 18–23 Fig. 6. Mean D-VEPs in blocks 1 and 2 of parallel (a) and orthogonal (b)conditions in the local orientation task. theglobaltask,theeffectofperceptuallearningonamplitudeofearlyD-VEPcomponentswasdependentonconfiguration.Acomparisonofresultsofexperiments1and2showsthatglobal orientation discrimination was initially, before learn-ing, less accurate than local ( P  < 0.01). Local orientationaccuracy is almost at ceiling in the first block. These resultsare in apparent contradiction with results showings prece-dence of global over local level of processing with complexfeatures [17,9]. However, it is well-known that basic features(orientedbars),andcomplexfiguresorcombinedfeaturesaresegmented on the basis of different rules [4].To explain the initial advantage of orientation discrimi-nation in the local task, one could call into cause a classicalmodel of texture segmentation [15], based and a two-stagefiltering interleaved by a rectification operation. Accordingto this model, different levels of absolute accuracy in theglobal and local tasks could indicate different levels of pro-cessing:lateprocessingatlow-spatialfrequencieslevelcouldbe required for the global task but not for the local. How-ever, since late low-spatial-frequency filtering would abolishthe parallel–orthogonal asymmetry, our results, showing aparallel–orthogonal asymmetry before learning, are at oddwith the two-stage filtering model.Our findings of a reduced accuracy in discriminating lo-cal and global orientation when configurations are orthogo-nal, despite orientation contrast being the same indicates thatorientation judgement relies on both global and local orien-tation information, regardless of their relevance for the task.This shows that irrelevant texture characteristics are implic-itlyprocessed,therebyaffectingongoingexplicitprocessing.Our primary result is that explicit and implicit process-ing both depend on learning, but in different ways. Stimulusrepetition improves performance more when orientations arepresent that conflict with that being judged (orthogonal), andequates discrimination accuracy in conflict and conflict-freeconfigurations. This indicates that the two configurations be-came equally salient as a result of learning. The increaseof accuracy in the two configurations is associated with anincrease(decrease)ofD-VEPamplitudeforparallel(orthog-onal) configurations. On the basis of the evidence that VEPamplitude is increased by texture segmentation [3,14], theeffect of an increased accuracy associated with an amplitudereduction suggests inhibition of the texture orientation con-flicting with the orientation being judged.Thesuggestionthatalearning-specificdecreaseinD-VEPamplitude reflects weighting of the input from conflictingorientation is reinforced by replication of this result in bothtasks:whenglobalorlocalorientationconflicts,theirrelevantorientationintheorthogonalconfigurationhastobeinhibitedto allow the opposite relevant orientation to be judged; this isassociatedwithadecreaseinD-VEPamplitude.Ontheotherhand, in the parallel configuration, the learning-dependentimproved accuracy has, as electrophysiological counterpart,an increase of negativity of earliest D-VEP component, aneffect suggesting facilitation of texture segmentation [14].To explain how learning produces either inhibition or fa-cilitation of cortical response, depending on the configura-tion,wehavetopinpointthemechanismaffectedbylearning.Psychophysical [7,18], neurophysiological [11] and electro- physiological data [3] show that besides orientation contrast,texture segmentation relies on grouping of texture elements.In our stimuli, differences in neither absolute orientationnor orientation contrast can explain the parallel–orthogonalasymmetry in accuracy and D-VEP amplitude. Indeed, ele-ment orientation and local orientation contrast are the samein the two configurations. The asymmetry must then relyon grouping factors. Using the same stimulus, Caputo andCasco [3] already showed an electrophysiological correlateof grouping, reflected in a parallel–orthogonal D-VEP asym-metry.Onthebasisofourresultswearguethat,intheglobaltask,a grouping process of similarly oriented elements is affectedby learning in both parallel and orthogonal configurations.Grouping is increased in the parallel condition but decreasedin the orthogonal, and the result is, in both configurations, anincrease in saliency.We first discuss the possibility that grouping mechanismmay be inhibited by learning. A learning-dependent inhibi-tion of grouping is suggested by accuracy data in the firstblockoforthogonalconfiguration(Fig.3)inglobaltask.Theyshow that accuracy is less than 50%, indicating that someobservers have a response bias for the opposite orientation.ComparisonbetweenFig.1bandcgivesaphenomenological
Similar documents
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks