Research project 2
Student: Doris E. Dijksterhuis Student number: 10354409 Credits for project: 36 EC
Period of research project: 08/01/2018 – 29/06/2018 Final date: 29/06/2018
Supervisor: Nadine Dijkstra Examiner 1: Marcel van Gerven Examiner 2: Filip van Opstal
Research institute: Donders Institute for Brain, Cognition and Behavior - Radboud University Nijmegen
Education: Research Master Brain and Cognitive Sciences Track: Cognitive Science
2
Vividness of Visual Imagery and its Relationship with
Alpha Oscillations
Doris E. Dijksterhuis
Radboud University, Donders Institute for Brain, Cognition and Behavior, Nijmegen, the Netherlands
This study aims at getting a better understanding of the underlying mechanisms of visual imagery. Using MEG, we investigated the relationship between the vividness of visual imagery and alpha oscillations. Human participants performed a perception and imagery task and the development of alpha power over time was explored. We proposed two possible outcomes for the direction of this relationship. If visual imagery functions similar to visual perception, we expect that a more vivid visual mental experience results in stronger activation of the visual cortex. Because previous literature suggests that alpha oscillations in the brain have an inhibiting function, this leads to the hypothesis that there is a negative correlation between alpha power and the vividness of imagery, which we henceforth will call the activation hypothesis. The other possible outcome suggests that the visual cortex should be less active during imagery, to inhibit incoming bottom-up information to only focus on the mental experience. This leads to the hypothesis that there is a positive correlation between alpha power and the vividness of imagery which we refer to as the inhibition hypothesis. We found no correlation between the vividness of visual imagery and alpha power which means we cannot say which proposed hypothesis is likely to be true. However, we took the first step in investigating this relationship and we propose directions for future studies to find a lead towards one of the mentioned hypotheses. This lead will contribute to the understanding of the functioning of visual imagery.
Visual imagery gives rise to visual experiences about objects that are absent in the visual field. It allows us to create a mental representation of these objects. Mental representations are key for many human cognitive processes such as learning, memory and planning (Kosslyn et al., 2001). The vividness of visual imagery seems to be important as it positively correlates with performance during tasks that require imagery (Albers et al., 2013). Many neuroscientific studies aimed at unraveling visual imagery in humans, using a variety of methods and tasks.
3 This has led to different concepts of how imagery might work, two of which we will discuss here (Lee et al., 2013; Keogh et al., 2016).
The first concept that will be discussed highlights the overlap between the working of visual imagery and visual perception. Among others, Lee et al. (2013) suggest that the visual cortex plays a critical role in the imagery of certain visual stimuli. Furthermore, their findings suggest differential processing during imagery and perception, but within the same neural substrates. This is in line with studies that show a resemblance in the mechanisms on which perception and imagery depend. Albers et al. (2013) for example propose that imagery and perception both rely upon the visual cortex, as it is used for both bottom-up processing and top-down generation of mental representations. Based on this, we could hypothesize that imagery, like perception, depends on visual cortex activity, among others (Dijkstra et al., 2017). Further, Cui et al. (2007) showed that higher visual cortex activity correlates with more vivid imagery.
On the other hand, another concept suggests that bottom-up information should be inhibited during imagery to only focus on the mental experience. This would mean that the primary visual areas should be less excitable during imagery. The results of Keogh et al. (2016) showed, by relating brain imaging data to the imagery strength of their participants, that lower excitability levels in the visual cortex predict stronger sensory imagery, meaning more vivid images. This agrees with the findings of Daselaar et al. (2010). Their study used functional Magnetic Resonance Imaging to show imagery-related suppression in the primary sensory cortices for both visual and auditory imagery. Daselaar et al. (2010) proposed that this suppression might occur to shield relevant areas from input from primary regions. They also mention inconsistent findings in previous studies regarding the involvement of primary regions in imagery, which highlights the importance of our study.
The inhibition and disinhibition of the visual cortex is partly driven by alpha oscillations (Clayton et al., 2017). Alpha oscillations are frequency bands between the 7.5 and 12.5 Hz and are the most dominant frequency in the human adult scalp electroencephalography (Klimesch, 1999). They are thought to relate with cortical excitability because of their inhibiting function (Wang et al., 2016) and it has been shown that alpha oscillations are involved in a range of cognitive functions (Jokisch & Jensen, 2007; Klimesch, 2012).
Considering the first concept discussed here, imagery, like perception, results in stronger activation of the visual cortex (Cui et al., 2007; Albers et al., 2013; Lee et al., 2013). If
4 we accept the assumption that alpha oscillations have an inhibiting function (Wang et al., 2016), then less alpha power is expected during imagery to limit the inhibition and establish stronger activation of the visual cortex. A traditional view is that alpha decreases in response to a visual stimulus (Klimesch et al., 2011). Taken together, this concept leads to the hypothesis that there is a negative correlation between alpha power and the vividness of imagery. In the current paper this is referred to as the activation hypothesis. The second concept states that less activity in the visual cortex facilitates imagery (Daselaar et al., 2010; Keogh et al., 2016). An increase in alpha results in more inhibition which leads to a decrease in activity. This concept therefore expects an increase in alpha power during visual imagery. A corresponding hypothesis thus suggests a positive correlation between alpha power and the vividness of imagery. We refer to this as the inhibition hypothesis.
In this study, by investigating the relationship between alpha oscillations and the vividness of visual imagery, we aim at determining which of the two hypotheses is more plausible, the activation- or the inhibition hypothesis. We did this by studying brain oscillations during perceived and imagined images over time, derived with magnetoencephalography (MEG) measurements, and the subjective vividness ratings of the participants. After obtaining the output of a frequency analysis, we will compare the time-frequency representations between low and high vividness ratings. If there is a difference in alpha power, the direction of it will show which hypothesis to focus on in future research.
Methods
Participants. Thirty volunteers with normal or corrected-to-normal vision gave written informed consent and participated. Five participants were excluded from the analysis, due to scanner- or other technical problems or due to a performance with less than 50% correct on the catch trials (as described below). Twenty-five participants (mean age 28.6, SD = 7.62) were included in the final analysis. Before the experiment the participants filled in the Vividness of Visual Imagery Questionnaire (VVIQ) (Marks, 1973). A more detailed overview of the methods and procedure and an initial analysis of these data is already published in Dijkstra et al. (2018).
5 Figure 1. Experimental Design. Two images were shown after each other (a face and a house or vice versa), followed by a
retro-cue that indicated which of the two images had to be imagined by the participant. Subsequently, a frame was shown in which the participants had to imagine the cued face or house as vividly as possible. After this, a continuous scale was shown on which the participant had to indicate their experienced imagery vividness by moving a bar (from Dijkstra et al., 2018).
Experimental design. The experimental task is depict in Figure 1. Two images were shown after each other (a face and a house or vice versa), followed by a retro-cue that indicated which of the two images had to be imagined by the participant. Subsequently, a frame was shown in which the participants had to imagine the cued face or house as vividly as possible. After this, a continuous scale was shown on which the participant had to indicate their subjective imagery vividness by moving a bar. To ensure that the participants did their best to imagine the stimuli with as great visual detail as possible, both categories existed of eight exemplars. 7% Of the trials were catch trials, which meant that the participants had to indicate which of four exemplars they had imagined (see ‘Catch trial’ in Figure 1).
MEG preprocessing. Data were recorded using a 275-channel MEG system at 1200 Hz with axial gradiometers and analyzed with MATLAB version R2017a and Fieldtrip (Oostenveld et al., 2011). Data from the sensors MRF66, MLC11, MLC32, MLF62 and MLO33 were not recorded for technical reasons. Two conditions were defined per trial. The first one represented the perception condition and was defined as 200 ms prior to onset of the first image until 200 ms after the offset of the first image. The second one represented the imagery
6 condition and was defined as 200 ms prior to the onset of the retro-cue until 500 ms after the offset of the imagery frame.
The variance of the data of each trial was calculated to identify the artifacts and trials with a high variance were visually inspected. After visual inspection, the trials with excessive artifacts were removed. Eye movements and heart rate artifacts were removed by calculating independent components of the MEG data and correlating these with the EOG and ECG signals. If a component had a high correlation it was manually inspected before removal. As a baseline correction, the activity of the first 300 ms from the onset of the initial fixation of the corresponding trial was averaged per channel and subtracted from those signals.
Data analysis. The planar gradients of the MEG field distribution were calculated for both conditions. After the time-frequency representations (TFRs) were calculated, the horizontal and vertical components of the estimated planar gradients were combined and are now close to the MEG signal as if measured with planar gradiometers (Jokisch & Jensen, 2007). This is helpful for interpreting spectral power at the sensor level and to make a better comparison between participants. To control the spectral smoothing around the frequency of interest we applied a Hanning taper method to an adaptive sliding time window (Jacobsen & Lyons, 2003). We used an adaptive time window of 3 cycles for each frequency ( time window in seconds = amount of cycles / frequency ) (Jokisch & Jensen, 2007). Prior to any statistical analyses, the power was averaged over the alpha frequency range (8-12 Hz).
To investigate the relationship between alpha oscillations and the vividness of imagery, we did a median split on the trial by trial vividness ratings within each participant (Barrowclif et al., 2004). We also performed a Pearson correlation test per subject to search for correlations between alpha power and the vividness ratings over trials.
Statistical analysis. To test for significant differences between low and high vivid trials and to compare the Pearson correlations with zero, nonparametric cluster based permutation tests were done. This test controls for false positives and solves the multiple comparison problem (Maris & Oostenveld, 2007; Maris, 2012). The test computes t-statistics for all channels and identifies clusters of neighboring channels having a threshold below p < 0.05. The cluster with the maximum sum of t-values of the channels defines the cluster-level test statistic (Jokisch & Jensen, 2007).
7
Results
Spectral changes in the alpha band over time
Due to a lack of a proper baseline condition, the perception and imagery conditions could not be normalized and therefore could not be compared with each other by means of statistical tests. By calculating the power for each sensor, we plotted the topographies for alpha power over time for each condition (Figure 2). We used time intervals of 0.4 seconds for the imagery condition and intervals of 0.2 seconds for the perception condition. The imagery condition has a longer time range than the perception condition and therefore has more topographies. Based on a visual inspection, we see a decrease in alpha power in the topographies of both conditions, right after the stimulus or retro-cue came on. We then see in the imagery condition that alpha power increases again and stays relatively stable over time.
Time = [-0.2 0] s Time = [0 0.2] s Time = [0.2 0.4] s
Time = [0.4 0.6] s Time = [0.6 0.8] s
Time = [-0.7 -0.3] s Time = [-0.3 0.1] s Time = [0.1 0.5] s Time = [0.5 0.9] s
Time = [0.9 1.3] s Time = [1.3 1.7] s Time = [1.7 2.1] s Time = [2.1 2.5] s
Time = [2.5 2.9] s Time = [2.9 3.3] s Time = [3.3 3.7] s
Figure 2. Topographies of alpha power over time during the imagery and perception condition. In both
conditions, the first 0.2 seconds were used as a baseline correction for this visualization. A. Topographies of alpha power over time are shown for the imagery condition with time intervals of 0.4 seconds. The cue onset is at time = -0.5 s and retro-cue offset is at time = 0 s. The imagery condition has a longer time range than the perception condition and therefore has more topographies. B. Topographies of alpha power over time are shown for the perception condition with time intervals of 0.2 seconds. The stimulus onset is at time = 0 s and stimulus offset is at time = 0.8 s. A B -1.5 x 10-27 0 1.5 x 10-27 -1.5 x 10-27 0 1.5 x 10-27
8 Relationship between vividness of visual imagery and alpha power
To investigate the relationship between vividness of imagery and alpha power, we divided all trials into ‘high vividness trials’ and ‘low vividness trials’ (Figure 3A), by combining the median splits of all participants. The difference between the powerspectrums of the low vividness trials and the high vividness trials was plotted into topographies of alpha power over time with time intervals of 0.4 seconds (Figure 3B). We found no significant difference between the high and low vividness trials. To further investigate the relationship between vividness and alpha power, we performed a Pearson correlation test per subject (Figure 4A), but again, there were no significant clusters. We plotted the correlations, averaged over subjects, into topographies of alpha power over time with time intervals with 0.4 seconds (Figure 4B). Based on visual inspection, there does seem to be a change in the direction of the correlation after the retro-cue dissappears and the participants start with imaging (Figure 4). The correlation is first negative, but turns positive over time.
Figure 3. Median Split on trial by trial vividness ratings within each
participant for the imagery
condition plotted over time and in topographies. The retro-cue onset is
at time = -0.5 s and retro-cue offset is at time = 0 s in both plots. A. We divided all trails by combining the median splits of all participants into ‘high vividness trials’ (blue line) and ‘low vividness trials’ (red line). The spectral power of alpha is plotted against time for both categories for the imagery condition for the occipital sensors. B. The first 0.2 seconds were used as a baseline correction for this visualization. The powerspectrum of the ‘low vividness trials’ was subtracted from the
powerspectrum of the ‘high
vividness trials’. The result is the difference between the high and low vividness trials and this is plotted into topographies of alpha power over time for the imagery condition with time intervals of 0.4 seconds.
4.3 4.2 4.1 4 3.9 3.8 3.7 3.6 3.5 3.4 3.3 -1 -0.5 0 0.5 1 1.5 2 2.5 3 3.5 4 Time (s) Spect ral po we r x 10-27 High vividness Low vividness A Time = [0.5 0.9]
Time = [-0.7 -0.3] s Time = [-0.3 0.1] s Time = [0.1 0.5] s
Time = [0.9 1.3] s Time = [1.3 1.7] s Time = [1.7 2.1] s
Time = [2.5 2.9] s Time = [2.9 3.3] s Time = [3.3 3.7] s
0 1 x 10-27 -1 x 10-27 Time = [0.5 0.9] s Time = [2.1 2.5] s B
9
Discussion
The aim of this study was to investigate the relationship between the vividness of visual imagery and alpha oscillations. Based on previous literature, we proposed two hypotheses about this relationship: the activation hypothesis predicted a negative correlation between alpha power and the vividness of imagery and the inhibition hypothesis predicted a positive correlation. By studying how alpha oscillations during imagery relate to the vividness ratings of the participants, we aimed at finding a lead towards one hypothesis. We did not observe a significant relation between alterations in alpha power and vividness over trials. This means that, based on these results, there is no lead towards the activation- or the inhibition
hypothesis.
This study is one of the first investigating the underlying mechanisms of visual imagery by looking at alpha oscillations. The current findings are in contrast with previous studies
Time = [-0.7 -0.3] s Time = [-0.3 0.1] s Time = [0.1 0.5] s Time = [0.5 0.9] s
Time = [0.9 1.3] s Time = [1.3 1.7] s Time = [1.7 2.1] s Time = [2.1 2.5] s
Time = [2.5 2.9] s Time = [2.9 3.3] s Time = [3.3 3.7] s -0.05
0 0.05 0.02 0.015 0.01 0.005 0 -0.005 -0.01 -0.015 -0.02 -0.025 -1 0 1 2 3 4 Time (s) Co rr elat io n c o ef fic ient s ( r)
Figure 4. Pearson Correlations between alpha power and trial by trial vividness ratings during the imagery condition plotted over time and in topographies. The Pearson
correlations are averaged over subjects. The retro-cue onset is at time = -0.5 s and retro-cue offset is at time = 0 s in both plots. A. A Pearson correlation test per subject was performed and the results were plotted against time. B. The first 0.2 seconds were used as a baseline correction for this visualization. Pearson correlations between alpha power and trial by trial vividness
ratings were plotted into
topographies with time intervals of 0.4 seconds.
A
10 which found that visual cortex activity does correlate negatively (Cui et al., 2007; Albers et al., 2013; Lee et al., 2013) or positively with the vividness of visual imagery (Daselaar et al., 2010; Keogh et al., 2016). By assuming that alpha oscillations have an inhibiting function (Wang et al., 2016), we suggested that alpha oscillations drive the activity of the visual cortex and therefore correlate with vividness as well. Even though previous literature suggests otherwise, our results propose that alpha oscillations are not the feature of activation that correlates with vividness. This is in line with the study of Clayton et al. (2017) that suggests that alpha oscillations only partly drive the inhibition and disinhibition of the visual cortex. Future studies could consider investigating the relation between other features of brain activation and vividness of visual imagery.
Another difference between the current study and previous studies which could explain the unpredicted results were the tasks and the content of the visual imagery. The task in this study involved imagining a face or a house. These two categories were chosen because they both elicit distinguishable neural responses throughout the visual hierarchy (Ishai et al., 2000). Using another content than a face or a house in combination with a different paradigm, most likely elicits a different neural response. In Daselaar et al. (2010) for example, the visual stimuli used in their experiment consisted of 456 Dutch nouns presented on a screen and presented as auditory stimuli. Their study found imagery-related suppression in the primary sensory cortices for both visual and auditory imagery. Another example is the study of Albers et al. (2013) where they used individual letters. This study found analogous representations during visual working memory and mental imagery in the early visual cortex. The study of Daselaar et al. (2010) and Albers et al. (2013) both use different visual stimuli and different paradigms which could explain why their outcomes are different from ours.
In the current study, the vividness of the visual imagery was rated by the participants by moving a bar to the left or to the right, indicating a low or high vivid experience on that trial. Further, before the participants started with the experiment they filled in the VVIQ, which is a measure of the imagery ability of people (Marks, 1973). Even though the results of the VVIQ of the participants correlated positively with their vividness ratings during the experiment (Dijkstra et al., 2018), the difference between participants is a problem when you want to look at the relationship between alpha power and vividness over participants. Some participants might be conservative and others might be liberal in their ratings, which makes the chance to find an effect on group level very small. This could explain why we did not find
11 a significant difference between the low vividness trials and the high vividness trials, because the median splits of all participants were combined. A future goal would be to find a way to relate self-reports of different subjects to each other.
Many MEG studies do source reconstruction to identify the brain areas that are responsible for the signals that are captured by the sensors (Oostenveld et al., 2011). This makes the interpretation of the location of the observed activity more reliable (Jokisch & Jensen, 2007). In the current study, no source reconstruction was done which means that there is no certainty about the location of the source of the measured alpha activity. Another explanation for our lack of significant results could be that the effects that the current study focused on were very specific to the visual brain area (Daselaar et al., 2010; Lee et al., 2013). Such effects are difficult to find if there is no clear localization of the alpha activity. Repeating this study and including a source reconstruction analysis plus a proper baseline condition could give interesting insights in the relation between the development of alpha power over time in the visual system and the vividness of imagery. Besides, Keogh et al. (2016) found that higher excitability levels in the prefrontal cortex predict more vivid imagery. Investigating the relationship between other brain areas and vividness should be considered in future research as well to get more insight in the underlying mechanisms of visual imagery.
Another solution for the source localization problem mentioned above, would be to use a different technique with a higher spatial resolution. This can be achieved by using human patients implanted with depth electrodes as part of their treatment for epilepsy. The locations of the electrodes are highly accurate and they are able to measure oscillations by recording local field potentials (Self et al., 2016). Using the same technology, Quiroga et al. (2008) found that certain cells in the human medial temporal lobe are tuned for high-level concepts such as a particular animal, person or place. A new study could use these high-level concepts in a similar paradigm as used in our study while performing intracranial human recordings. This creates the opportunity to measure alpha oscillations and relating these to the trial by trial vividness ratings of the corresponding participant while knowing exactly where the oscillations are coming from.
There were two conditions in the current study: a perception and an imagery condition. Due to a lack of a proper baseline condition, the two conditions could not be normalized and therefore no valid comparison could be made. Replicating this study and incorporating a baseline condition could provide interesting insights. The perception condition
12 should have the same duration as the imagery condition, in order to compare the development of alpha power after stimulus offset between both conditions. If the relation between alpha power and vividness during imagery could be compared with the relation between alpha power and vividness during perception, one might be able to find a lead towards the activation- or inhibition hypothesis. If this relation is similar in both conditions, then the activation hypothesis should be the focus of future studies. If the relation between alpha power and vividness is significantly different between both conditions, then the
inhibition hypothesis is the hypothesis to focus on.
Based on the results of this study, the question if the relationship between the vividness of visual imagery and alpha power leans towards the activation hypothesis or the
inhibition hypothesis remains. However, the directions that were proposed here for future
studies will help towards a better understanding of the underlying mechanisms of visual imagery.
References
Albers, A.M., Kok, P., Toni, I., Dijkerman, H.C. & de Lange, F.P. (2013). Shared
representations for working memory and mental imagery in early visual cortex.
Current Biology, 23(15), 1427-1431.
Barrowcliff, A.L., Gray, N.S., Freeman, T.C.A. & MacCulloch, M.J. (2004). Eye-movements reduce the vividness, emotional valence and electrodermal arousal associated with negative autobiographical memories. Journal of Forensic Psychiatry and Psychology,
15(2), 325-345.
Cui, X., Jeter, C.B., Yang, D., Montague, P.R. & Eagleman, D.M. (2007). Vividness of mental imagery: Individual variability can be measured objectively. Vision Research, 47(4), 474-478.
Daselaar, S.M., Porat, Y. Huijbers, W. & Pennartz, C.M.A. (2010). Modality-specific and modality-independent components of the human imagery system. NeuroImage,
13 Dijkstra, N., Bosch, S.E. & van Gerven, M.A.J. (2017). Vividness of Visual imagery depends on
the neural overlap with perception in visual areas. The Journal of Neuroscience, 37(5), 1367-1373.
Dijkstra, N., Mostert, P., de Lange, F.P., Bosch, S. & van Gerven, M.A.J. (2018). Differential temporal dynamics during visual imagery and perception. eLIFE, 1-16.
Dugué, L., Marque, P. & Van Rulen, R. (2011). The phase of ongoing oscillations mediates the causal relation between brain excitation and visual perception. Journal of
Neuroscience, 31(33), 11889-11893.
Feinberg, D.A & Yacoub, E. (2012). The rapid development of high speed, resolution and precision in fMRI. Neuroimage, 6(2), 247-253.
Ishai, A., Ungerleider, L. G. & Haxby, J. V. (2000). Distributed neural systems for the generation of visual images. Neuron, 28, 979–990.
Jokisch, D. & Jensen, O. (2007). Modulation of gamma and alpha activity during a working memory task engaging the dorsal or ventral stream. Journal of Neuroscience, 27(12), 3244-3251.
Keogh, R. & Pearson, J. (2014). The sensory strength of voluntary visual imagery predicts visual working memory capacity. Journal of Vision, 14(12), 1-13.
Keogh, R., Bergmann, J. & Pearson, J. (2016). Cortical excitability controls the strength of mental imagery. BioRxiv, 1-48.
Klimesch, W. (1999). EEG alpha and theta oscillations reflect cognitive and memory performance: a review and analysis. Brain Research Reviews, 29, 169-195. Klimesch, W. (2012). Alpha-band oscillations, attention, and controlled access to stored
information. Trends in Cognitive Sciences, 16(12), 606-617.
Klimesch, W., Doppelmayer, M., Russeger, H., Pachinger, T. & Schwaiger, J. (1998). Induced alpha band power changes in the human EEG and attention. Neuroscience Letters,
244(2), 73-76.
Klimesch, W., Fellinger, R. & Freinberger, R. (2011). Alpha oscillations and early stages of visual encoding. Frontiers in Psychology, 2, 1-11.
Kosslyn, S., Ganis, G. & Thompson, W. (2001). Neural foundations of imagery. Nature
Reviews Neuroscience, 2(9), 635-642.
Lee, S., Kravitz, D.J. & Baker, C.I. (2013). Disentangling visual imagery and perception of real-world objects. Neuroimage, 59(4), 4064-4073.
14 Maris, E. & Oostenveld, R. (2007). Nonparametric statistical testing of EEG- and MEG data.
Journal of Neuroscience Methods, 164(1), 177-190.
Maris, E. (2012). Statistical testing in electrophysiological studies. Psychophysiology, 49(4), 549-565.
Marks, D.F. (1973).Visual imagery differences in the recall of pictures. British Journal of
Psychology, 64(1), 17-24.
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, 1-9.
Self., M.W., Peters, J.C., Possel, J.K., Reithler, J., Goebel, R., Ris, P., Jeurissen, D., Reddy, L., Clause, S., Baayen, J.C. & Roelfsema, P.R. (2016). The effects of context and attention on spiking acitivty in human early visual cortex. PLoS Biology, 14(3), 1-26.
Wang, C., Rajagovindan, R., Han, S. & Ding, M. (2016). Topdown control of visual alpha oscillations: Sources of control signals and their mechanisms of action. Frontiers in
Human Neuroscience, 10(15), 1-14.
Jacobsen, E. & Lyons, R. (2003). The sliding DFT. IEEE Signal processing magazine, 20(2), 74-80.
Quiroga, R.Q., Kreiman, G., Koch, C. & Fried, I. (2008). Sparse but not ‘Grandmother-cell’ coding in the medial temporal lobe. Trends in Cognitive Sciences, 12(3), 87-91.
