Recognizing an Object from the Sum of Its Parts: An Intracranial Study on Alpha Rhythms Josie-Anne Bertrand1, Julie Tremblay2, Maryse Lassonde1,2, Phetsamone Vannasing2, Dang Khoa Nguyen3, Manon Robert1, Alain Bouthillier3, and Franco Lepore1,2

Abstract ■ Little is known about the relation of alpha rhythms and

object recognition. Alpha has been generally proposed to be associated with attention and memory and to be particularly important for the mediation of long-distance communication between neuronal populations. However, how these apply to object recognition is still unclear. This study aimed at describing the spatiotemporal dynamics of alpha rhythms while recognizing fragmented images of objects presented for the first time and presented again 24 hr later. Intracranial electroencephalography was performed in six epileptic patients undergoing presurgical evaluation. Time–frequency analysis revealed a strong alpha activity, mainly of the evoked type, propagating

INTRODUCTION In the past decades, several studies have tried to understand the cerebral electrophysiological mechanisms responsible for object recognition. The focus has been mainly on the induced gamma-band response (iGBR) recorded from scalp EEG. iGBR has been related to coherent perception of visual stimuli and is thought to represent the synchronization of neuronal populations involved in feature binding and comparisons with memory (Bertrand et al., 2013; Hassler, Barreto, & Gruber, 2011; Grützner et al., 2010; Gruber, Müller, & Keil, 2002; Tallon-Baudry & Bertrand, 1999). However, recent evidence attributes a significant association of alpha rhythm to visual perception ( Voytek et al., 2010; Freunberger, Klimesch, Griesmayr, Sauseng, & Gruber, 2008). Its specific relation is still unclear, and a number of interpretations have been proposed. First, many studies measuring alpha amplitude related a reduced alpha intensity, or event-related desynchronization (ERD), to brain activation because of the fact that eye opening, visual stimuli, or increased attention evoked a decrease in alpha amplitude (Klimesch, Fellinger, &

1

Université de Montréal, 2Centre de Recherche de lʼHôpital Ste-Justine, Montréal, 3Centre Hospitalier de lʼUniversité de Montréal, Hôpital Notre-Dame © 2014 Massachusetts Institute of Technology

from posterior cerebral areas to anterior regions, which was similar whether the objects were recognized or not. Phase coherence analysis, however, showed clear phase synchronization specific for the moment of recognition. Twenty-four hr later, frontal regions displayed stronger alpha activity and more distributed phase synchronization than when images were presented for the first time. In conclusion, alpha amplitude seems to be related to nonspecific mechanism. Phase coherence analysis suggests a communicational role of alpha activity in object recognition, which may be important for the comparison between bottom–up representations and memory templates. ■

Freunberger, 2011; Pfurscheller, 2001). Alternately, an increased alpha event-related synchronization (ERS), observed during retention intervals of working memory and STM tasks, may reflect cortical inhibition for irrelevant stimuli (Sauseng et al., 2005; Busch & Herrmann, 2003; Jensen, Gelfand, Kounios, & Lisman, 2002). At the end of the retention period, an ERD was triggered by the participantʼs response. On the basis of this result, Klimesch et al. (2011) suggested that ERD might be related to the access and retrieval from memory. Similarly, in object recognition tasks, ERD in posterior brain regions was strongest for recognized objects than for unrecognized stimuli, supporting the link between ERD and retrieval from semantic memory (Freunberger et al., 2008; Vanni, Revonsuo, & Hari, 1997). However, others argue against this ERD/ERS activation/ inhibition hypothesis, by linking an alpha power increase to attentional modulation. Namely, Palva and Palva (2007) questioned whether the ERD recorded after a working memory taskʼs retention interval may simply reflect the end of attentional processes rather than memory activation. Moreover, evoked alpha power has been shown to increase according to working memory load, and at a somatosensory detection task, a parietal ERS has been linked to better and faster performances (LinkenhaerHansen, Nikulin, Palva, Ilmoniemi, & Palva, 2004; Jensen et al., 2002). Other evidences of the association between Journal of Cognitive Neuroscience 26:8, pp. 1797–1805 doi:10.1162/jocn_a_00582

alpha rhythm and attentional top–down modulation come from other techniques, such as alpha phase coherence analyses (Palva & Palva, 2007). For instance, conscious awareness of visual stimuli was found to be dependent on the alpha phase angle (Dugué, Marque, & VanRullen, 2011; Busch, Dubois, & VanRullen, 2009; Matthewson, Gratton, Fabiani, Beck, & Ro, 2009). Moreover, decreased performances at a visual working memory task were also linked to a reduced prestimulus alpha coherence between frontal and posterior brain regions provoked by TMS over pFC (Zanto, Rubens, Thangavel, & Gazzaley, 2011). This supports the idea that the alpha rhythm mediates attentional top–down modulation of prefrontal cortices over occipital regions involved in visual processing. It has also been suggested that alpha rhythm may act as a means of long-distance communication between brain regions, namely, the frontal and posterior regions, whereas iGBR may represent local communication between neuronal assemblies (von Stein & Sarnthein, 2000). Alpha activity may modulate iGBR in precise regions such that parallel visual processing can efficiently occur (Voytek et al., 2010; Osipova, Hermes, & Jensen, 2008). In relation to this, Bar et al. (2006) proposed a model of object recognition in which visual information is rapidly sent through the magnocellular pathway to prefrontal cortices for rapid integration. From this blurred information processing, the pFC would be able to generate possible candidates of the objectʼs identity by accessing semantic memory and to provide them to the bottom–up processes. This top–down modulation would facilitate recognition by decreasing the number of comparisons between the objectʼs representation and memory templates. Using magnetoencephalography and fMRI, an early alpha phase

coherence between OFC and occipital regions and a later phase coherence between OFC and temporal regions during object recognition suggested that the OFC may subserve the function of generating possible candidates to bottom–up processes and may do so by using the alpha rhythm as a way to communicate (Bar et al., 2006). Alpha rhythmʼs relation to object recognition is thus difficult to understand as different data analyses were performed with different cognitive tasks in-between studies. Thus, this study generally aimed at describing alpha measures inside a single object recognition task. We used the intracranial EEG (iEEG) technique, which consists in the implantation of subdural grid and/or depth electrodes on the brain of epileptic patients undergoing presurgical evaluation for epileptic foci localization (Figure 1A). This technique has one of the best spatial and temporal resolutions and represents an outstanding opportunity to investigate the spatiotemporal dynamics of alpha rhythms while recognizing objects. Moreover, we decided to use an original task, designed by Doniger et al. (2000), during which fragmented images of objects are presented in an incremental manner such that the object became more and more recognizable (Figure 1B). This task is very unique in the sense that it allows comparison of alpha rhythm at recognition (threshold of recognition, T) to moments before recognition (T − 2 and T − 1) and after recognition (T + 1). We think this task will provide a precise description of alpha in relation to recognition of objects presented for the first time. Moreover, it has been shown that, when fragmented images were presented a second time, recognition happened at prior levels as a consequence of top–down facilitation (Viggiano & Kutas, 2000; Snodgrass & Corwin, 1988). Precise mechanisms about how this happens have

Figure 1. (A) Postimplantation MRIs of a patient implanted with a grid of electrodes. From left to right, sagittal, coronal, and horizontal sections representing an electrode implanted over the occipital cortex. (B) Example of a set of fragmented images at Session 1. The eight images composing each set were presented from the most fragmented image to the complete image. When patients recognized the object, this level was marked as the threshold of recognition (T). Prior levels were marked as T − 1 and T − 2 accordingly, and the next level was marked as T + 1. Images were presented again 24 hr later (Session 2). When images are presented a second time, recognition happens at prior levels. SAG = sagittal; COR = coronal; AXI = axial.

1798

Journal of Cognitive Neuroscience

Volume 26, Number 8

not been identified yet. By presenting the same paradigm 24 hr later (Session 2), we aimed at describing alpha rhythmʼs relation to this top–down facilitation. Analyses were performed using separate as well as total (induced + evoked) time–frequency analyses to assess which response provides the most accurate information about the amplitude of alpha activity triggered by visual stimuli. Evoked analyses provide information about the activity that is phase-locked to stimuli. This type of analysis is known to mask information jittering temporally, commonly termed induced responses, which may be crucial for the understanding of underlying cerebral mechanisms (TallonBaudry & Bertrand, 1999). Evoked analyses being more commonly performed to study alpha rhythm, both types of analyses were compared in this study. Moreover, phase coherence analyses were performed to investigate the communicational nature of alpha rhythms. We hypothesize the following: 1. If alpha rhythm is more related to access in memory as proposed by Klimesch et al. (2011), total alpha power should be at its lowest level (strongest ERD/lowest ERS) at the moment of recognition (T) compared with other recognition levels because, at T, the observed object has been successfully retrieved from semantic memory. Because of the additional involvement of episodic memory as a top–down facilitation mechanism at Session 2, alpha at T should be more powerful at Session 2 compared with Session 1. However, if alpha rhythm is more related to attentional modulation, it appears logical that its total power would be equivalent across T − 2, T − 1, and T (because participants are equally working hard to make sense out of stimuli between conditions) or may slightly increase at T (because recognition may be more arousing). This pattern would be equivalent at both Sessions 1 and 2. 2. Alpha may mediate communication between top– down areas and regions involved in object processing. According to the model proposed by Bar et al. (2006), the OFC may act as a top–down modulator by generating potential candidates about the object identity, which would be transferred to occipito-temporal areas by means of alpha phase synchronization. From this, it is possible to believe that, as the object gets more and more defined, less potential candidates would have to be transmitted. That may result in a gradual decrease in alpha coherence from T − 2 to T + 1, which would be equivalent at both sessions. However, at Session 2, a higher number of phase coherence between cortical regions, namely with those involved in retrieval from episodic memory, might also be involved. Conversely, one may hypothesize that a match between a memory template and the visual cortical representation of the object at recognition would result in a stronger alpha coherence between regions compared with previous levels. A higher number of phase coherence between other cerebral regions at Session 2 may also apply in this case.

METHODS Participants Electrodes were implanted over the brain of six patients (four men, mean age = 27.5 years, SD = 14.2 years) undergoing presurgical evaluation for localization of epileptic foci. All patients were right-handed and were left-hemisphere dominant for language. They all had normal or corrected-to-normal vision. This study was approved by the Centre Hospitalier de lʼUniversité de Montréal Ethics Committee. All patients signed an informed consent form before participating in the study. Stimuli and Task Stimuli, taken from a bank of images designed by Snodgrass and Corwin (1988), were line drawings of common objects. Each image of an object was fragmented over eight levels, with level 1 representing the most fragmented version of the object and level 8 representing the image in its complete form (Figure 1B). Each image was presented in such a way that the objects gradually became recognizable. Stimuli appeared on a 17-in. monitor (1280 × 1024 pixels) using E-Prime software (Psychology Software Tools, Inc., Pittsburgh, PA). Images were displayed for 1 sec after which participants were asked if they recognized the object. Interstimuli intervals were variable and long enough to minimize the effect of the verbal response on recordings. Patients were asked at all time to fixate the center of the screen. Between 90 and 150 sets of images were presented depending on the patientʼs collaboration and level of fatigue. All eight images composing a set were presented even if patients recognized the objects earlier. Images were divided in blocks composed of 30 sets of images. Blocks were administered on different days, depending on the patientʼs tolerance to the task. Blocks were presented again about 24 hr later, without advising participants that the same fragmented images were going to be presented (Session 2). The order of stimulus presentation was randomized. On the first day of experiment, only new images were presented in blocks. On subsequent days, images seen the day before and new images were presented, such that both Session 1 (new) and Session 2 (old) stimuli could be administered on the same day. This was to control for time and learning effects that could otherwise bias the data. Localization of Electrodes Subdural grid and/or strip electrodes were implanted over the brain of patients with up to 124 contact points, strategically localized according to the presurgical evaluation. To avoid contamination of the data by epileptic activity, all selected electrodes in this study were situated away from epileptic zones identified by an expert epileptologist. Localization was performed by first coregistering postimplantation MRIs of all patients onto the Montreal Neurological Institute template and then converting the electrodeʼs Bertrand et al.

1799

Stellate Gridview coordinates to Talairach coordinates. Electrode locations were determined according to the Talairach Daemon atlas (Lancaster et al., 2000). In Figure 3, all electrodes were represented over the right hemisphere for the lateral view of the brain and on the left hemisphere for the medial portion of the brain to facilitate comprehension. It was decided to do so because the activity in both hemispheres was similar. In total, 525 electrodes were analyzed for all patients combined: 80 in the occipital lobe (from two patients), 102 in the parietal lobe (from three patients), 134 in the temporal lobe (from all patients), and 209 in the frontal lobe (from five patients). OFC was defined as in Kringlebach (2005). Twenty-eight electrodes were localized over the OFC.

frequency f resulted from this analysis. Time–frequency analysis was also performed over the mean ERP and was subtracted from the total response to result in a purely induced response. Mean total power values of five time widows were calculated using MATLAB (version 7.9.1.705, The MathWorks, Inc., Natick, MA). The windows comprised all wavelet points from 8.65 to 13.86 Hz (consisting mainly of upper alpha frequency) in the time intervals 0–100, 100–200, 200–400, 400–600, and 600–800 msec (Figure 2B). Mean total power values of each electrode at each window were then plotted over an MRI template according to Talairach coordinates (Figure 2C). Phase Coherence Analysis

Data Analysis Data acquisition was performed with a sampling rate ranging from 200 to 2000 Hz, depending on patients and other undergoing research protocols. Using Brain Vision Analyzer (version 1.0, Brain Products, Gilching, Germany), EEG segments at the moment of recognition were marked as the threshold of recognition (T). Previous levels were marked as T − 1 and T − 2; and the next level, as T + 1 (Figure 1B). All segments containing epileptic activity or a signal amplitude of ±350 μV were rejected because they constituted an obvious artifact. A mean of 92.67% (SD = 12.73%) of trials were artifact free and considered for analysis. Data were filtered from 0.05 to 100 Hz (24 dB/oct). An average reference was applied to the data. Each trial was segmented from −200 to 1000 msec. Time–Frequency Analysis Time–frequency analyses were performed over each trial using complex Gaussian Morletʼs wavelets, which generate a complex wavelet, w, for different frequencies (σf) and time domain (σt) around a central frequency f0 according to: wðt; f0 Þ ¼ A exp ð−t2 =2σ2t Þ exp ð2iπf0 tÞ

ð1Þ

 pffiffiffi−1=2 A ¼ σt π

ð2Þ

σf ¼ 1=2πσt

ð3Þ

with

and

Wavelets were calculated in the frequency range of 6–100 Hz in 2-Hz linear steps. The ratio f/σf was set to 7, with f corresponding to the central frequency of the wavelet and σf corresponding to its standard deviation. Data were baseline corrected by dividing epochs of −200 to −50 msec. A total power value for each trial at time t and 1800

Journal of Cognitive Neuroscience

Coherence analyses were performed according to the event-related cross-coherence procedures described in Delorme and Makeig (2004) adapted to MATLAB. It consists in the trial-by-trial measure of coupling between signals of electrode pairs that reveal the degree of synchronization between the two selected electrodes. The obtained value represents the magnitude of crosscoherence centered at the time and frequency of the wavelets and varies from 0 (absence of synchronization) to 1 (perfect synchronization). Electrode pairs compared in the phase coherence analysis were all electrodes implanted in different cerebral lobes of each patient. Statistical Analyses Statistical analyses were performed within patients only. A significant finding had to be replicated in more than one patient before being reported to ensure reliability of the results. SPSS (version 19.0.0, IBM Company, Armonk, NY) was used for statistical analyses. A paired t test was applied to compare Session 1 and Session 2 levels at which T happened in each participant. Significance was set to a p value of

Recognizing an object from the sum of its parts: an intracranial study on alpha rhythms.

Little is known about the relation of alpha rhythms and object recognition. Alpha has been generally proposed to be associated with attention and memo...
573KB Sizes 0 Downloads 0 Views