Attentional blindness and perceptual blindness are two distinct
phenomena acting on different aspects of recurrent processing
Assignment: Bachelor thesis
Version: first version/final version /resit
Name student : Mika Mautner-Rohde Student number : 11679093
Institute (SILS, AMC, etc.) : Psychology Research Institute, University of Amsterdam
Department : Brain and Cognition Supervisor : Simon van Gaal
Mentor : Samuel Noorman
Date: 22.01.2021 Wordcount: 5061
Attentional blindness and perceptual blindness are two distinct
phenomena acting on different aspects of recurrent processing
Abstract
Many popular theories of consciousness agree on the fact that recurrent
processing is necessary for achieving conscious perception. Recurrent processing can occur either locally or globally, depending on how widespread the information is distributed in the brain. Attentional blindness and perceptual blindness are thought to be two distinct phenomena that might differ in the way in which they disturb local or global recurrent processes. When comparing these two phenomena it is apparent that performance difference is a common confounding factor in many previous studies. In this experiment we aim to explore the different impact of attentional and perceptual blindness on recurrent processing. This is done by inducing attentional blindness with an attentional blink paradigm, and perceptual blindness with backwards masking, while measuring electroencephalogram (EEG) data. Attentional blink and masking trials are performance matched to tackle the performance difference confounding factor. We show that, behaviorally, both manipulations impair subjects accuracy on discriminating Kanizsa illusion figures from control figures. Furthermore, we were able to decode the absence or presence of said stimuli from the EEG data. Here, a difference between attentional and
perceptual blindness trials could be detected. It appeared, that attentional blindness interferes strongly with global recurrent processing, while perceptual blindness interferes with both, local and global recurrent processes alike.
Keywords: consciousness; recurrent processing; attentional blindness; perceptual blindness; masking; attentional blink; Kanizsa illusion
Introduction
Many prominent theories of consciousness postulate that recurrent processing is needed to integrate information and create a conscious experience (Del Cul, Baillet, & Dehaene, 2007; Fahrenfort, Scholte, & Lamme, 2007; Lamme, 2006). Recurrent processing can be seen as a second phase of information processing in the brain. In a preceding phase, a stimulus is processed in hierarchically lower stages and
subsequently send to hierarchically higher stages. This process is often called the ‘feedforward sweep’(Lamme & Roelfsema, 2000). The second step is recurrent
processing (via recurrent connections) from higher cortical regions back to regions of the same or lower level (Lamme, 2003). Recurrent processing is found to occur
locally, within sensory modules and modules of close anatomical proximity, or globally, when information exchange is widespread throughout the brain and reaches frontoparietal networks (Lamme, 2003). Opinions on how those recurrent processes contribute to a conscious experience and how to label the emerging stages
of consciousness, diverge at this point (see Pitts, Lutsyshyna, & Hillyard, 2018 for an in depth analysis of the discussion). Most of the researchers however, agree on a broader four-stage model of consciousness (Dehaene, Changeux, Naccache, Sackur, & Sergent, 2006; Lamme, 2010). In this model consciousness depends on two factors: 1) bottom-up stimulus strength, which determines the presence or absence of
recurrent processing, and 2) top-down attention, which determines the depth of processing. With sufficient attention, given that also enough bottom-up stimulus strength is available, the processing of a stimulus can cross a threshold to be globally ignited. That way information becomes accessible for several cognitive processes such as working memory and verbal report (Dehaene et al., 2006; Mashour, Roelfsema, Changeux, & Dehaene, 2020).
Obstructing bottom-up stimulus strength or top-down attention interferes with the conscious perception of a stimulus and leads to distinct phenomena: perceptual blindness and attentional blindness (Kanai, Walsh, & Tseng, 2010). Perceptual blindness on the one hand occurs, when enough attention is directed towards the stimulus, but the bottom-up stimulus strength is too weak. On the other hand, if bottom-up stimulus strength is sufficient, but not enough top-down
attention is directed towards the stimulus, attentional blindness occurs. Dehaene and colleagues propose in their paper from 2006 that in perceptual blindness the stimulus elicits a feedforward sweep, but recurrent processing cannot occur. This is supported by earlier findings (Dehaene et al., 2001; Fahrenfort et al., 2007; Lamme, Zipser, & Spekreijse, 2002; Moutoussis & Zeki, 2002). Furthermore, they hypothesize that in attentional blindness the stimulus should elicit recurrent processing, but because there is not enough top-down attention towards the stimulus, the recurrent processing would stay local and would not reach frontoparietal networks. This is partly based on earlier findings, where a feedforward sweep could be detected for attentional blindness, but the later components, that we would classify as global recurrent processing, were missing (Sergent, Baillet, & Dehaene, 2005). Based on the theory of Dehaene and colleagues (2006) perceptual and attentional blindness seem to be fundamentally different. But can we actually conclude this by the literature? When diving deeper into the literature on these phenomena it is striking that attentional blindness and perceptual blindness are difficult to compare with each other. While often not being compared with each other in the same experiment, the difficulty to compare the two is mostly due to the fact that many studies report very different performance scores for perceptual blindness compared with attentional blindness. These performance scores were obtained by asking participants to report whether or not they perceived a target stimulus. Performance scores in experiments that induce perceptual blindness (e.g. with masking) are often at chance level (in forced choice paradigms) or have a very low detection rate (Dehaene et al., 2001; Moutoussis & Zeki, 2002), whereas scores using attentional blindness (e.g. AB) paradigms are often considerably higher (Gross et al., 2004; Kranczioch, Debener, Schwarzbach, Goebel, & Engel, 2005; Marois, Yi, & Chun, 2004; Sergent et al., 2005).
Thus, it is difficult to conclude with certainty if the differences found are really due to distinctions in recurrent processing, or rather due to performance differences.
One study that found a very neat way to compare attentional and perceptual blindness in the same experiment is the study of Fahrenfort from 2017 ( Fahrenfort, Van Leeuwen, Olivers, & Hogendoorn, 2017). They used the well-studied Kanizsa illusion to explore their impact on recurrent processing. Kanizsa stimuli create visual illusions where illusory surface regions of shapes can be perceived, even though they are physically not present (see figure 1). Those illusions can only be
perceptually achieved by recurrent feedback from higher order areas to areas lower in the visual hierarchy (Kok, Bains, Van Mourik, Norris, & De Lange, 2016; Pak, Ryu, Li, & Chubykin, 2020). Kanizsa control images are very similar to Kanizsa images but lack the illusory surface area (see figure 1). When the brain responses to Kanizsa control images, which have the same feedforward processes as the Kanizsa images but lack the feedback to create the illusory surface area, are subtracted from the brain responses to Kanizsa images, recurrent activity will be isolated (Fahrenfort et al., 2017). This is why those stimuli are very useful to examine recurrent processing and the reason that Fahrenfort’s study design introduced a promising measurement of recurrent processing. Fahrenfort and colleagues manipulated the conscious perception of the Kanizsa illusions with backward masking and attentional blink (AB) trials in a rapid serial visual presentation task (RSVP). An AB was induced by presenting two Target stimuli with a short lag (300ms) between them to let the processing of the first target interfere with processing of the second, making the second invisible to the subject. They used a classifier with multivariate pattern
analysis (MVPA) on electroencephalogram (EEG) data to discriminate between trials with Kanizsa illusion or control, while comparing trials with perceptual blindness and attentional blindness. The classifier had three peaks, where classification was
Figure 1: Examples of stimuli containing a Kanizsa illusion (left) and control stimuli (right). The top row shows figures where the top half of the figure has more contrast than on the bottom half (hence called high-low contrast), while the figures in the bottom row have higher contrast on the bottom half of the figure (hence called low-high contrast).
significantly above chance. One significant peak appeared at ~264ms and is
described by the researchers to represent local recurrent processing, because it was found to be impaired only by masking, and not by the AB. A later peak had its maximum amplitude at ~406ms. That peak was reduced in both, the masking and the AB condition, leading to the assumption that this peak might represent global recurrent processing. This hypothesis is supported by the finding that, interestingly, this peak occurred in the time frame of the P300, an event related potential (ERP). The P300 is often associated with complex cognitive processes which require global recurrent processes (Gebuis & Reynvoet, 2013; Sergent et al., 2005) and has also been described as a neural correlate of consciousness (NCC) (Dehaene & Changeux, 2011). The fact that this peak was disrupted by the absence of top-down attention, while leaving the local recurrent processing peak intact implies that attentional blindness does indeed disrupt global recurrent processing more than local recurrent
processing. Furthermore, this is supported by the finding that also behavioral accuracy for detecting the illusory triangle was reduced. Fahrenfort and colleagues thus conclude that local recurrent processing is enough for conscious perception of a stimulus.
Since masking disrupted both of the classification peaks mentioned earlier, the researchers investigated whether masking disrupts all processing of stimuli. They decoded high- vs. low- contrast stimuli and found an early peak around 80-90ms, which was present in all conditions. If masking would have wiped out all processing of stimuli, the classifier would not have been able to discriminate between the high- and low- contrast stimuli. So, this peak was interpreted as contrast detection and thus the feedforward sweep, which was not affected by masking.
Though, also in this study we find the problem mentioned earlier: While performance on the masked trials is at chance, performance on the AB trials is above chance. So, it can still not be stated with certainty that the effects found really are attributable to the differences of perceptual and attentional blindness alone, or even at all. Those effects might rather be due to performance differences.
So, with this current study we want to settle the still remaining question on how local and global recurrent processing are affected by inducing attentional blindness and perceptual blindness. Based on previous research mentioned earlier, perceptual blindness is expected to interfere with all recurrent processing (local and global), while attentional blindness is only expected to reduce global recurrent processing and leave local recurrent processing intact.
We are going to test these hypotheses with a well-controlled experiment, inspired by the procedures and findings of Fahrenfort et al (2017). We will also use a RSVP task with Kanizsa figures as target figures to be able to measure recurrent processing. The conscious perception of those targets will be manipulated by backward masking on the one hand and an AB paradigm on the other hand. This way perceptual and attentional blindness will be induced. Moreover, we will use MVPA analyses to train a classifier to discriminate between perceptual blindness trials and trials with attentional blindness.
We will further develop the experimental design by using performance matching for the masking and AB conditions. This step is crucial to rule out any effects of performance accuracy and to tackle this prominent confounding factor, found so often in this field of research. Performance on the masking trials will be matched with performance on the AB trials. Finally, we will, similar to Fahrenfort and colleagues (2017), decode the contrast of the target images to control if the feedforward sweep is uninterrupted by our manipulations.
We expect findings similar as the ones found by Fahrenfort and colleagues (2017). Performance accuracy for both manipulations is expected to be lower than in unmanipulated trials. Also, since we will match the performances of masking and AB trials, those are expected not to differ significantly from each other. Furthermore, we expect to find similar classification peaks as found in Fahrenfort et al. (2017). When decoding contrast of target figures, an early peak (in a time window between 50 - 125ms) is expected to be found, representing the feedforward sweep. We will, unlike Fahrenfort and his team, not use different figures to achieve different contrast levels, but we will let the classifier discriminate between the target figure’s inherent high-low and low-high contrast ratios (see figure 1) and expect to see the same early classification peak. This peak is expected to neither be affected by masking nor the AB. A second peak (in a time window between 125 - 375ms) is expected to be identified for unmanipulated trials as well as for the AB condition. However, this peak is expected to be reduced in the AB condition and even further reduced in the masking condition. Additionally, a late peak (in a time window between 375 - 600ms) is expected to be found only for unmanipulated trials, and thus being reduced for both, the masking and the AB condition. Lastly, in trials with both manipulations, masking and AB, we expect to find a reduced second peak as well as no (or a radically reduced) third peak.
Materials and Method
Participants
Thirty-three subjects enrolled for the experiment. Two had to be excluded due to non-sufficient AB and one because of too low task performance, leaving us with thirty subjects that were included for further analyses. The subjects, 10 of which men, had a mean age of 21,7 years (± SD = 2,6) and most of the participants were right-handed (2 left-handed). All had normal or corrected-to-normal vision and provided written informed consent to take part in the study beforehand. The experiment has been approved by the local ethics commission of the University of Amsterdam. Subjects received monetary compensation or psychology research credits for their participation.
Design and Procedure
The experimental design of the current study is to large parts based on the
experimental design of Fahrenfort et al. (2017). Stimuli were presented on a white
background at the center of a computer screen with a refreshing rate of 60Hz, using
Presentation (Neurobehavioral Systems). Target stimuli were eight black figures, half of which containing a Kanizsa illusion, the other half being Kanizsa control images (see figure
1). The illusory effect of the Kanizsa image was created by the Pac-man like shapes. The other shapes in the Kanizsa figures did not have a purpose for this experiment. Furthermore, red distractor images were used (see figure 2A). Those were similar to the Kanizsa control images, but the six individual pieces of the image were 180° rotated and thus never contained a Kanizsa illusion. Lastly, black masking stimuli were used, which never contained a Kanizsa illusion and were designed to overlap with large parts of the target stimuli (see figure 2B).
Subjects participated in two sessions. The first session was a training session to familiarize participants with the task and the stimuli. Also, during this first session the size of the AB was calculated, and a staircase was used to determine the mask strength (contrast of the masking figures), that produced same behavioral accuracies as AB trials. To familiarize subjects with the target images and the
illusion, participants were instructed to discriminate between target stimuli with and without a Kanizsa illusion, presented between distractor stimuli, in an RSVP task. Participants reported whether or not they perceived a Kanizsa illusion by pressing “A” or “L” (for “yes” or “no”) on a keyboard. Button meanings were swapped between blocks to counterbalance for response related processes. In task one and two, five red distractor images were presented, followed by a black target image, which was then followed by six more red distractor images. The red distractor images were always presented for 17ms each. Black target images were presented for either 67ms (in task 1) or 17ms (in task 2). The stimulus onset asynchrony (SOA) was always held at 100ms. Participants could advance with the following task when they had an accuracy of 80% or higher. In the third task participants were shown five distractor images, followed by the first target image (T1), which was presented for 67ms, again followed by eight distractor images, then the second target (T2), presented for 17ms, and again followed by six red distractor images. This task was designed to acquaint the subjects to the long lag condition (no-AB), resulting in a long-lag of 900ms. Then, in task four, participants completed the same task with two-thirds of the trials containing a short lag between the two targets to induce an AB. In task five, two latencies were tested to check which one induced the larger AB: one red distractor between the two targets (resulting in a short lag of 200ms) or two
Figure 2: Examples of a red distractor stimulus (A), and a masking stimulus (B), both of which never contain a Kanizsa illusion.
red distractor stimuli (resulting in a short lag of 300ms). If subjects did not show a sufficient AB, they were excluded from the experiment. An AB was defined as sufficient when the performance accuracy on short-lag trials subtracted from the performance on long-lag trials was above 5%. After determining the short lag resulting in the greatest AB for each subject, masking was introduced in task six, leading to our 2 x 2 design with the factors lag length and masking. The four
different trial types were: Long-lag unmasked (L1M1), short-lag unmasked (L2M1), long-lag masked (L1M2) and short-lag masked (L2M2). In the following seventh task the first staircase was used to match the performance of the L1M2 trials to the
performance on the L2M1 trials. This crucial procedure was done by adjusting the mask strength until the one leading to a similar performance accuracy as the AB was found. The staircase used the weighted up-down method (Kaernbach, 1991) with the formula: Sup * p = Sdown * (1 – p). P is the percentage correctly identified T2, which the
staircase should strive for. Sup is the step size leading to a decrease in difficulty
(decrease of mask contrast) and Sdown is the step size leading to an increase in
difficulty (increase of mask contrast). The minimal step size was 1% of the maximum mask contrast, on a spectrum from white to black in 256 steps. A correctly detected Kanizsa figure resulted in an increased difficulty (Sdown), while an incorrect response
to the Kanizsa figure decreased the difficulty (Sup). After 25 reversals the staircase
Figure 3: Experimental design. Example of a short-lag, masked trial (L2M2). Presence of masking stimuli and number of distractor stimuli between the two targets determine the experimental condition of the trial. Stimulus onset asynchrony (SOA) was held constant at 100ms.
* one and two distractors leading to short-lag trials (200ms and 300ms respectively), eight distractors leading to a long-lag trial (900ms).
was stopped and the average mask strength of the last 20 reversals was calculated. The resulting masking strength was used as starting mask strength in session two.
During the second session participants were asked to complete short versions of tasks 1,2,3 and 6 from session one, while being connected to the EEG equipment. After the training, the experimental task started, and EEG was recorded. During this task a second staircase was used. This crucial staircase constantly checked how well the performances on L1M2 and L2M1 conditions were matched, while the
participant performed the task. Mask strength was adjusted by 1.5 levels of contrast for every percentage point that the trial types differed in performance. These
dynamic adjustments started after 32 trials were completed for both conditions. New mask strength adjustments could be made after completion of at least 16 L1M2 trials after the last adjustment. Furthermore, a third staircase controlled that the average overall performance on attentional blindness and perceptual blindness trials was also matched across all subjects. The adjustments coming from this across-subjects-staircase were dominant over the adjustments of the previously described across-subjects-staircases. Subjects completed 16 blocks with 64 trials of the experimental task. Trials were equally divided over the four conditions, leading to 256 trials per experimental condition.
Behavioral analysis
Responses were defined as hits (Kanizsa illusion correct), misses (Kanizsa illusion incorrect), correct rejections (control stimulus correct) and false alarms (control stimulus incorrect). The hit rate (HR) was calculated as the fraction of hits from all trials, while the correct rejection rate (CRR) was calculated by the fraction of correct rejections. Balanced accuracy for each experimental condition on T2 was computed for each participant. This was done by calculating the average of the HR and the CRR. The balanced accuracies were compared with a two-way repeated measures ANOVA to determine main and interaction effects for the factors lag length and mask strength. Then, post-hoc t-tests were used to test which
experimental conditions differed significantly from one another.
EEG data collection and preprocessing
EEG data was collected at 1024Hz using a 64-channel ActiveTwo system (BioSemi). The signal was re-referenced to the average signal of electrodes placed on the earlobes. The acquired data was analyzed using Matlab (Mathworks), the
EEGLAB toolbox (Delorme & Makeig, 2004), the fieldtrip toolbox (Oostenveld, Fries, Maris, & Schoffelen, 2011), and the ADAM toolbox (Fahrenfort, van Driel, van Gaal, & Olivers, 2018). First, a high-pass filter of 0.1Hz was applied, to circumvent
problems of baseline corrections on an AB task (Fahrenfort and colleagues describe this issue in detail in their paper from 2017). Second, data was epoched between -500 and 1000ms relative to stimulus onset. Then, T1 data was baseline corrected in a
period of -250 to 0ms before stimulus onset. Lastly, data was down sampled to 128Hz.
EEG MVPA
For each participant a backward decoding classification algorithm was used on the raw data of all scalp electrodes, to discriminate between T2 trials with a Kanizsa illusion and controls. This was done by using T1 data as training data and testing on the T2 data. Doing this for every time point in the EEG signal, this yielded us the classification accuracy over time for every experimental condition. Moments in time where the classification accuracy was greatest, appeared as peaks. To check whether the mean classification accuracy for all subjects was significantly above chance, a double-sided t-test with a multiple comparison correction was used. This was done with the use of a cluster-based permutation test with 1000 iterations. To detect the early classification peak (in the time window between 50 - 125ms), associated with the feedforward sweep, the classifier was trained to distinguish between contrasts (high-low/low-high) of the target figures (see figure 1). The assumption, that the four conditions do not differ significantly at this peak, was tested with a two-way repeated measures ANOVA with the variables lag length (short/long) and masking (present/absent).
Furthermore, to check for main and interaction effects, a 2 x 2 x 2 ANOVA with lag length (short/long), masking(absent/present) and peak number (2nd/3rd) was used. Post hoc t-tests then could determine which conditions differ significantly from each other. This way it could be tested, if there is a significant difference
between L1M2 and L2M1 at the classification in the time window between 125 - 375ms. Additionally, it tested if the difference between L1M1 and L2M1 is larger at the third classification peak (in a time window between 375 - 600ms) than at the second.
Results
Behavioral results
The balanced accuracy for T2 was calculated for each participant on each of the four trial types and plotted together with the average performance across subjects in Figure 4. As expected, the manipulations with masking and AB led to a lower balanced T2 accuracy. This finding was endorsed by the repeated measures ANOVA, which showed a strong main effect of both masking (F1,29 = 503.02, P < 10−8)
and lag length (F1,29 = 450.74, P < 10−8). Also, an interaction effect between masking
and lag length was found (F1,29 = 20.00, P = 1.10x10−4). Post-hoc t-tests revealed that
lag length significantly lowered accuracy on the unmasked trials (t(29) = 14.57, P =
6.99x10-15) as well as on the masked trials(t(29) = 9.39, P = 2.67x10-10). It was also shown
that masking had a significant adverse effect on performance accuracy in both the long lag trials (t(29) = -15.02, P = 3.22x10-15) and the short lag trials (t(29 )= -8.55, P =
2.03x10-9). Furthermore, it was apparent that the L1M2 and L2M1 trials did not differ
significantly from each other (t(29) = 0.144, P = 0.887) and thus, that performance
matching was successful.
EEG data
The classification accuracy was plotted as the area under the curve (AUC) over time for each of the four experimental conditions. When decoding figure contrast, we selected a time window between 50 - 125ms, where the first classification peak representing the feedforward sweep could be detected.
Classification accuracy peaked at ~89ms and was visible for all 4 conditions (Figure 5). The 2 x 2 ANOVA did neither show significant main effects, nor an interaction effect for the factors masking and lag length. This verified the hypothesis, that the
Figure 4: Behavioral accuracy of T2 detection for the factors mask (absent/present) and lag length (long/short). Data for each subject is plotted in light colors and the mean of all participants is plotted as a darker line. Error bars represent the mean ± SEM. ns, not significant * P < 0.05, *** P < 10-3, ***** P < 10-5, ********** P < 10-10. T2 behavior ** ** ** ** ** ** ** *
feedforward sweep is not interrupted by our manipulations, which is especially of interest for the masking condition.
Classification accuracy in the time window between 125 - 375ms yielded, as expected, the second classification peak. Here the AUC peaked around 214ms for L1M1, L1M2 and L2M1 trials, while peaking a little bit earlier for L2M2 trials, at 152ms (Figure 6). When selecting a time window between 375 and 600ms, the third classification peak was also detected. This third classification peak was highest at ~449ms (Figure 6). The 2 x 2 x 2 repeated measures ANOVA revealed main effects for all of the three factors: lag length (F1,29 = 39.23, P = 7.75x10-7), masking (F1,29 =
104.66, P = 3.93x10−11) and peak number (F1,29 = 10.47, P = 0.003). Our most important
goal was to reveal possible differences between the long-lag-mask and the short-lag unmasked condition during the second classification peak. So, a post-hoc t-test
Figure 5: Classification accuracies for contrast detection of T2 over time for all four experimental conditions. Long lag unmasked (L1M1) in red, long lag masked (L1M2) in green, short lag unmasked (L2M1) in blue and short lag masked (L2M2) in yellow. The lightly colored area around the lines represents ±SEM. Thick lines represent p < 0.05.
confirmed that these two conditions did indeed differ significantly in the second classification peak (t(29) = -3.84, P = 6.09x10-4). This result can also be seen in figure 7A.
Furthermore, it became clear that lag length had a greater impact on the classification accuracy during the third classification peak, than during the second peak. This result is supported by the significant three-way interaction effect found with the aforementioned 2 x 2 x 2 ANOVA (F1,29 = 4.72, P = 0.04). This conclusion can
be visualized when comparing the slope of the blue line in Figure 7A with its slope in Figure 7B, which is way steeper for peak three.
Figure 6: T2 classification accuracies over time for all four experimental conditions. Long lag unmasked (L1M1) in red, long lag masked (L1M2) in green, short lag unmasked (L2M1) in blue and short lag masked (L2M2) in yellow. The lightly colored area around the lines represents ±SEM. Thick lines represent p < 0.05. Straight lines below the plot represent the duration of p < 0.05.
max ~449ms max ~214ms
Lastly, it was found that the third classification peak is congruent with behavior. This is supported by the fact that the post-hoc t-test for the performance matched conditions did not reach significance at the third classification peak (t(29) =
-0.30, P = 0.77). Moreover, figure 7 illustrates that the third classification peak behaves qualitatively more like the behavioral data than the second peak.
Discussion
The aim of this study was to reproduce findings of Fahrenfort and colleagues (2007) and, by introducing performance matching to the experimental design, shed more light on the neural difference between attentional and perceptual blindness and their impact on local and global recurrent processing. We were able to find similar classification peaks as Fahrenfort et al. (2007) when decoding our EEG data. The current results support our hypothesis, that attentional blindness and perceptual blindness do indeed interfere differently with local and global recurrent processing. While attentional blindness seems to interfere especially with global recurrent processing, perceptual blindness seems to obstruct both forms of recurrent processing. We can conclude this from the fact that masked trials show reduced classification accuracy during the second and the third classification peak.
Furthermore, short lag trials only had a slightly negative impact on the second peak, while showing great negative impact on the third classification peak. Additionally, since the performance matched attentional blindness and perceptual blindness trials
Figure 7: EEG classification accuracy as AUC for T2 plotted at peak classification performance. (A) For the second classification peak (peaking at ~214ms) and (B) for the third classification peak (peaking at ~449ms). Blue lines represent unmasked conditions and red lines represent masked conditions. Classification accuracy at 449ms is qualitatively more analogous to the behavioral accuracy than at 214ms. Data for each subject is plotted in light colors and the mean of all participants is plotted as a darker line. Error bars represent the mean ± SEM. ns, not significant, * P < 0.05, *** P < 10-3, ***** P < 10-5, ********** P < 10-10. long short lag 0.4 0.5 0.6 0.7 0.8 0.9 1 A U C 0.6802 0.6312 0.5776 0.5453 absent present mask
A
long short lag 0.4 0.5 0.6 0.7 0.8 0.9 1 A U C 0.6407 0.5526 0.5487 0.5206 absent present maskB
T2 classification T2 classification at ~ 449ms at ~ 214msLag length Lag length
********** *****
*****
showed no significant difference during the third classification peak, these findings further strengthen the hypothesis that the third classification peak represents global recurrent processing, due to its similarity to behavior.
Based on the behavioral results we can conclude that performance matching of the attentional blindness and perceptual blindness trials was successful.
Moreover, it can be concluded that the two manipulations were effective, since both, masking and short lag, lowered performance accuracy of the subjects and combining the two lead to lowest performance accuracy. Considering the EEG decoding results during the early classification peak, representing the feedforward sweep, the
conclusion can be drawn that the feedforward sweep is not touched by perceptual blindness. The early classification peak (at ~89ms) was found in all conditions. This is in line with findings of Fahrenfort and colleagues and also in line with our own hypothesis. The same strong conclusion for attentional blindness cannot be drawn on that matter. This is due to the fact that subjects were not told to attend to the different contrast levels of the targets, and thus attention was not actually
manipulated for this analysis. However, considering the points made on attentional blindness in the introduction, we would still not expect attentional blindness to interfere with the feedforward sweep. One noticeable detail might be that the
unmanipulated condition could be decoded significantly for a longer period of time than the other conditions. However, this might be due to the fact that this
classification accuracy, labeled as significant, is stretching into the time window of the second classification peak. Given that the second classification peak is stronger in the unmanipulated condition than in the other conditions, it is plausible that the later aspects of the first peak might be driven by this.
A possible limitation to this study might be the fact that in this study we only used on-diagonal decoding. We might therefore have missed interesting insights that an off-diagonal decoding could add to the study design. Future research could also incorporate off-diagonal decoding to create temporal generalization matrices and visualize activity patterns in the brain over time. Another interesting addition to this study would be to incorporate a measure of meta cognition to the design. Earlier studies did incorporate meta cognition as an alternative measure to the decoding of EEG signals when studying attentional and perceptual blindness and demonstrated its value (Kanai et al., 2010). That way, even stronger conclusions could be drawn concerning the impact of attentional and perceptual blindness on conscious
perception. A third suggestion for future research is to use a similar study design to the one used in this study while administrating a drug to half of the participants. This drug should preferably be a safe, well established pharmaceutical that has a well-studied effect on receptor types that are associated with feedforward and recurrent activity, such as AMPA and NMDA receptors respectively (Self, Kooijmans, Supèr, Lamme, & Roelfsema, 2012). That way more insights into the exact mechanisms underlying recurrent processing and in what way they relate to consciousness might be studied.
This study succeeded in its goal to bring more clarity to the still open questions regarding attentional and perceptual blindness. Our results support the ideas of Dehaene et al. (2006); attentional blindness mainly interferes with global recurrent processing, and perceptual blindness strongly diminishes both types of recurrent processing, local and global. We demonstrated that attentional and
perceptual blindness are indeed two distinct phenomena, acting on different aspects of information processing in the brain. This insight should be valuable for
consciousness researchers to make an informed decision on which paradigms to choose to study consciousness.
References
Dehaene, S., & Changeux, J. P. (2011). Experimental and Theoretical Approaches to Conscious Processing. Neuron, 70(2), 200–227.
https://doi.org/10.1016/j.neuron.2011.03.018
Dehaene, S., Changeux, J. P., Naccache, L., Sackur, J., & Sergent, C. (2006). Conscious, preconscious, and subliminal processing: a testable taxonomy.
Trends in Cognitive Sciences, 10(5), 204–211.
https://doi.org/10.1016/j.tics.2006.03.007
Dehaene, S., Naccache, L., Cohen, L., Bihan, D. Le, Mangin, J.-F., Poline, J.-B., & Rivière, D. (2001). Cerebral mechanisms of word masking and unconscious repetition priming. Nature Neuroscience, 4(7), 752–758.
https://doi.org/10.1038/89551
Del Cul, A., Baillet, S., & Dehaene, S. (2007). Brain dynamics underlying the
nonlinear threshold for access to consciousness. PLoS Biology, 5(10), 2408–2423. https://doi.org/10.1371/journal.pbio.0050260
Delorme, A., & Makeig, S. (2004). EEGLAB: an open source toolbox for analysis of single-trial EEG dynamics including independent component analysis. Journal of
Neuroscience Methods, 134(1), 9–21.
Fahrenfort, J. J., Scholte, H. S., & Lamme, V. A. F. (2007). Masking disrupts reentrant processing in human visual cortex. Journal of Cognitive Neuroscience, 19(9), 1488– 1497. https://doi.org/10.1162/jocn.2007.19.9.1488
Fahrenfort, Johannes J., van Driel, J., van Gaal, S., & Olivers, C. N. L. (2018). From ERPs to MVPA using the Amsterdam Decoding and Modeling toolbox (ADAM). Frontiers in Neuroscience, 12(JUL).
https://doi.org/10.3389/fnins.2018.00368
Fahrenfort, Johannes J., Van Leeuwen, J., Olivers, C. N. L., & Hogendoorn, H. (2017). Perceptual integration without conscious access. Proceedings of the National
Academy of Sciences of the United States of America, 114(14), 3744–3749.
https://doi.org/10.1073/pnas.1617268114
Gebuis, T., & Reynvoet, B. (2013). The neural mechanisms underlying passive and active processing of numerosity. NeuroImage, 70, 301–307.
Gross, J., Schmitz, F., Schnitzler, I., Kessler, K., Shapiro, K., Hommel, B., & Schnitzler, A. (2004). Modulation of long-range neural synchrony reflects temporal
limitations of visual attention in humans. Proceedings of the National Academy of
Sciences, 101(35), 13050–13055. https://doi.org/10.1073/pnas.0404944101
Kaernbach, C. (1991). Simple adaptive testing with the weighted up-down method.
Perception & Psychophysics, 49(3), 227–229.
Kanai, R., Walsh, V., & Tseng, C. H. (2010). Subjective discriminability of invisibility: A framework for distinguishing perceptual and attentional failures of
awareness. Consciousness and Cognition, 19(4), 1045–1057. https://doi.org/10.1016/j.concog.2010.06.003
Kok, P., Bains, L. J., Van Mourik, T., Norris, D. G., & De Lange, F. P. (2016). Selective activation of the deep layers of the human primary visual cortex by top-down feedback. Current Biology, 26(3), 371–376.
https://doi.org/10.1016/j.cub.2015.12.038
Kranczioch, C., Debener, S., Schwarzbach, J., Goebel, R., & Engel, A. K. (2005). Neural correlates of conscious perception in the attentional blink. NeuroImage,
24(3), 704–714. https://doi.org/https://doi.org/10.1016/j.neuroimage.2004.09.024
Lamme, V. A.F., Zipser, K., & Spekreijse, H. (2002). Masking interrupts figure-ground signals in V1. Journal of Vision, 1(3), 1044–1053.
https://doi.org/10.1167/1.3.32
Lamme, Victor A.F. (2003). Why visual attention and awareness are different. Trends
in Cognitive Sciences, 7(1), 12–18. https://doi.org/10.1016/S1364-6613(02)00013-X
Lamme, Victor A.F. (2006). Towards a true neural stance on consciousness. Trends in
Cognitive Sciences, 10(11), 494–501. https://doi.org/10.1016/j.tics.2006.09.001
Lamme, Victor A.F. (2010). How neuroscience will change our view on consciousness. Cognitive Neuroscience, 1(3), 204–220.
https://doi.org/10.1080/17588921003731586
Lamme, Victor A.F., & Roelfsema, P. R. (2000). The distinct modes of vision offered by feedforward and recurrent processing. Trends in Neuroscience, 23(11), 571– 579. https://doi.org/https://doi.org/10.1016/S0166-2236(00)01657-X
Marois, R., Yi, D.-J., & Chun, M. M. (2004). The Neural Fate of Consciously Perceived and Missed Events in the Attentional Blink. Neuron, 41(3), 465–472.
https://doi.org/https://doi.org/10.1016/S0896-6273(04)00012-1
Mashour, G. A., Roelfsema, P., Changeux, J. P., & Dehaene, S. (2020). Conscious Processing and the Global Neuronal Workspace Hypothesis. Neuron, 105(5), 776–798. https://doi.org/10.1016/j.neuron.2020.01.026
Moutoussis, K., & Zeki, S. (2002). The relationship between cortical activation and perception investigated with invisible stimuli. Proceedings of the National
Academy of Sciences, 99(14), 9527–9532. https://doi.org/10.1073/pnas.142305699
Oostenveld, R., Fries, P., Maris, E., & Schoffelen, J.-M. (2011). FieldTrip: open source software for advanced analysis of MEG, EEG, and invasive electrophysiological data. Computational Intelligence and Neuroscience, 2011.
Cortical Representation of Illusory Contours in Mouse Primary Visual Cortex.
The Journal of Neuroscience, 40(3), 648 LP – 660.
https://doi.org/10.1523/JNEUROSCI.1998-19.2019
Pitts, M. A., Lutsyshyna, L. A., & Hillyard, S. A. (2018). The relationship between attention and consciousness: An expanded taxonomy and implications for no-report paradigms. Philosophical Transactions of the Royal Society B: Biological
Sciences, 373(1755). https://doi.org/10.1098/rstb.2017.0348
Self, M. W., Kooijmans, R. N., Supèr, H., Lamme, V. A., & Roelfsema, P. R. (2012). Different glutamate receptors convey feedforward and recurrent processing in macaque V1. Proceedings of the National Academy of Sciences, 109(27), 11031 LP – 11036. https://doi.org/10.1073/pnas.1119527109
Sergent, C., Baillet, S., & Dehaene, S. (2005). Timing of the brain events underlying access to consciousness during the attentional blink. Nature Neuroscience, 8(10), 1391–1400. https://doi.org/10.1038/nn1549
