Wednesday 29 April 2015

Peering directly into the human brain

Wiki Commons
With the rise of non-invasive brain imaging such as functional magnetic resonance imaging (fMRI), researchers have been granted unprecedented access to the inner workings of the brain. It is now relatively straightforward to put your experimental subjects in an fMRI machine and measure activity 'blobs' in the brain. This approach has undoubtedly revolutionised cognitive neuroscience, and looms very large in people's idea of contemporary brain science. But fMRI has it's limitations. As every student in the business should know, fMRI has poor temporal resolution. fMRI is like a very long-exposure photograph: the activity snapshot actually reflects an average over many seconds. Yet the mind operates at the millisecond scale. This is obviously a problem. Neural dynamics are simply blurred with fMRI. However, probably more important is the theoretical limit.

Wiki in ECoG
Electricity is the language of the brain, but fMRI only measures changes in blood flow that are coupled to these electrical signals. This coupling is complex, therefore fMRI can only provide a relatively indirect measure of neural activity. Electroencephalography (EEG) is a classic method for measuring actual electrical activity. It has been around for more than 100 years, but again, as every student should know: EEG has poor spatial resolution. It is difficult to know exactly where the activity is coming from. Magnetoencephalography (MEG) is a close cousin of EEG. Developed more recently, MEG is better at localising the source of brain activity. But the fundamental laws of physics mean that any measure of electromagnetic activity from outside the head will always be spatially ambiguous (the inverse problem). The best solution is to record directly from the surface of the brain. Here we discuss the unique opportunities in that arise in the clinic to measure electrical activity directly from the human brain using electrocorticography (ECoG).

Epilepsy can be a seriously debilitating neurological condition. Although the symptoms can often be managed with medication, some patients continue to have major seizures despite a cocktail of anti-epileptic drugs. So-called intractable epilepsy affects every aspect of life, and can even be life-threatening. Sometimes the only option is neurosurgery: careful removal of the specific brain area responsible for seizures can dramatically improve quality of life.

Psychology students should be familiar with the case of Henry Molaison (aka HM). Probably the most famous neuropsychology patient in history, HM suffered intractable epilepsy until the neurosurgeon William Scoville removed two large areas of tissue in the medial temporal lobe, including left and right hippocampus. This pioneering surgery successfully treated his epilepsy, but this is not why the case became so famous in neuropsychology. Unfortunately, the treatment also left HM profoundly amnesic. It turns out that removing both sides of the medial temporal lobe effectively removes the brain circuitry for forming new memories. This lesson in functional neuroanatomy is what made the case of HM so important, but there was also a important lesson for neurosurgery – be careful which parts of the brain you remove!

The best way to plan a neurosurgical resection of epileptic tissue is to identify exactly where the seizure is comping from. The best way to map out the affected region is to record activity directly from the surface of the brain. This typically involves neurosurgical implantation of recording electrodes directly in the brain to be absolutely sure of the exact location of the seizure focus. Activity can then be monitored over a number of days, or even weeks, for seizure related abnormalities. This invasive procedure allows neurosurgeons to monitor activity in specific areas that could be the source of epileptic seizures, but also provides a unique opportunity for neuroscientific research.

From Pasley et al., 2012 PLoS Biol. Listen to audio here
During the clinical observation period, patients are typically stuck on the hospital ward with electrodes implanted in their brain literally waiting for a seizure to happen so that the epileptic brain activity can be ‘caught on camera’. This observation period provides a unique opportunity to also explore healthy brain function. If patients are interested, they can perform some simple experiments using computer based tasks to determine how different parts of the brain perform different functions. Previous studies from some of the great pioneers in neuroscience mapped out the motor cortex by stimulating different brain areas during neurosurgery. Current experiments are continuing in this tradition to explore less well charted brain areas involved in high-level thought. For example, in a recent study from Berkeley, researchers used novel brain decoding algorithms to convert brain activity associated with internal speech into actual words. This research helps us understand the fundamental neural code for the internal dialogue that underlies much of conscious thought, but could also help develop novel tools for providing communication to those otherwise unable to general natural speech.

From Dastjerdi et al 2013 Nature Communications (watch video below)

In Stanford, researchers were recently able to identify a brain area that codes for numbers and quantity estimation (read study here). Critically, they were even able to show that this area is involved in everyday use for numerical cognition, rather than just under their specific experimental conditions. See video below.
Wiki Commons

The great generosity of these patients vitally contributes to the broader understanding of brain function. They have dedicated their valuable time in otherwise adverse circumstances to help neuroscientists explore the very frontiers of the brain. These patients are true pioneers.

Key References

Dastjerdi, M., Ozker, M., Foster, B. L., Rangarajan, V., & Parvizi, J. (2013). Numerical processing in the human parietal cortex during experimental and natural conditions. Nat Commun, 4, 2528.

Pasley, B. N., David, S. V., Mesgarani, N., Flinker, A., Shamma, S. A., Crone, N. E., Knight, R. T., & Chang, E. F. (2012). Reconstructing speech from human auditory cortex. PLoS Biol, 10, e1001251.

Video showing the use of a number processing brain area in everyday use:

Friday 24 April 2015

Research Briefing: organising the contents of working memory

Figure 1. Nicholas Myers
Research Briefing, by Nicholas Myers

Everyone has been in this situation: you are stuck in an endless meeting, and a colleague drones on about a topic of marginal relevance. You begin to zone out and focus on the art hanging in your boss’s office, when suddenly you hear your name mentioned. On high alert, you suddenly shift back to the meeting and scramble to retrieve your colleague’s last sentences. Miraculously, you are able to retrieve a few key words – they must have entered your memory a moment ago, but would have been quickly forgotten if hearing your name had not cued them as potentially vital bits of information.

This phenomenon, while elusive in everyday situations, has been studied experimentally for a number of years now: cues indicating the relevance of a particular item in working memory have a striking benefit to our ability to recall it, even if the cue is presented after the item has already entered memory. See our previous Research Briefing on how retrospective cueing can restore information to the focus of attention in working memory.

In a new article, published in the Journal of Cognitive Neuroscience, we describe a recent experiment that set out to add to our expanding knowledge of how the brain orchestrates these retrospective shifts of attention. We were particularly interested in the potential role of neural synchronization of 10 Hz (or alpha-band) oscillations, because they are important in similar prospective shifts of attention.

Figure 2. Experimental Task Design. [from Myers et al, 2014]
We wanted to examine the similarity of alpha-band responses (and other neural signatures of the engagement of attention) both to retrospective and prospective attention shifts. We needed to come up with a new task that allowed for this comparison. On each trial in our task, experiment volunteers first memorized two visual stimuli. Two seconds later, a second set of two stimuli appeared, so that a total of four stimuli was kept in mind. After a further delay, participants recalled one of the four items.  

In between the presentation of the first and the second set of two items, we sometimes presented a cue: this cue indicated which of the four items would likely be tested at the end of the trial. Crucially, this cue could have either a prospective or a retrospective function, depending on whether it pointed to location where an item had already been presented (a retrospective cue, or retrocue), or to a location where a stimulus was yet to appear (a prospective cue, or precue). This allowed us to examine neural responses to attention-guiding cues that were identical with respect to everything but their forwards- or backwards-looking nature. See Figure 2 for a task schematic.

Figure 3. Results: retro-cueing and pre-cueing
trigger different attention-related ERPs.
[from Myers et al, 2014]
We found marked differences in event-related potential (ERP) profiles between the precue and retrocue conditions. We found evidence that precues primarily generate an anticipatory shift of attention toward the location of an upcoming item: potentials just before the expected appearance of the second set of stimuli reflected the location where volunteers were attending. These included the so-called early directing attention negativity (or 'EDAN') and the late directing attention-related positivity (or 'LDAP'; see Figure 3, middle panel; and see here for a review of attention-related ERPs). Retrocues elicited a different pattern of ERPs that was compatible with an early selection mechanism, but not with stimulus anticipation (i.e., no LDAP, see Figure 3, upper panel). The latter seems plausible, since the cued information was already in memory, and upcoming stimuli were therefore not deserving of attention. In contrast to the distinct ERP patterns, alpha band (8-14 Hz) lateralization was indistinguishable between cue types (reflecting, in both conditions, the location of the cued item; see Figure 4).

Figure 4. Results: retro-cueing and pre-cueing trigger similar patters
of de-synchronisation in low frequency activity (alpha band at ~10Hz).
[from Myers et al, 2014]
What did we learn from this study? Taken together with the ERP results, it seems that alpha-band lateralization can have two distinct roles: after a precue it likely enables anticipatory attention. After a retrocue, however, the alpha-band response may reflect the controlled retrieval of a recently memorized piece of information that has turned out to be more useful than expected, without influencing the brain’s response to upcoming stimuli.

It seems that our senses are capable of storing a limited amount of information on the off chance that it may suddenly become relevant. When this turns out to be the case, top-down control allows us to pick out the relevant information from among all the items quietly rumbling around in sensory brain regions.

Many interesting questions remain that we were not able to address in this study. For example, how do cortical areas responsible for top-down control activate in response to a retrocue, and how do they shuttle cued information into a state that can guide behaviour? 

Key Reference: 

Myers, Walther, Wallis, Stokes & Nobre (2014) Temporal Dynamics of Attention during Encoding versus Maintenance of Working Memory: Complementary Views from Event-related Potentials and Alpha-band Oscillations. Journal of Cognitive Neuroscience (Open Access)

Friday 10 April 2015

Research Briefing: Preferential encoding of behaviourally relevant predictions revealed by EEG

Figure 1. Accurate predictions help us prepare the best action
Statistical regularities in the environment allow us to generate predictions to guide perception and action. For example, consider the challenge facing a goal keeper during a penalty shoot-out. There is simply not enough time to act responsively. By the time the ball in hurtling along its path to some deep corner of the net, it is probably already too late to plan and execute the appropriate action to save the goal. Instead, the goal keeper must actively predict the likely trajectory of ball before it has even left the boot of the other player. The goal keeper must use the any subtle clues betrayed by the kicker, any reliable signal to help prepare for a dive in the correct direction.


Predictions are useful in many contexts, not just professional sport. In everyday life, your brain is constantly generating predictions that help you to interpret the world around you and plan appropriate behaviour. Hermann von Helmholtz described the importance of predictions derived from past experience for interpreting perceptual information (see previous post). More recently, theorists that argued that the brain is essentially a predictive machine - for example, the Free Energy Principle proposes that perception and action are best conceptualised as a dynamic interplay between predictions we make about our environment and how well these predictions explain future events.

Research Question

In any given context, some predictions might be useful for behaviour, but others less so. Here, we asked whether and how the brain learns relevant and/or irrelevant predictive relationships using electroencephalography (EEG).

Methods Summary

Participants in our experiment performed a simple target detection task (see Figure 2). On each experimental trial, they were presented with a visual stimulus drawn for a set of ten possible fractals images. At the start of the experiment, one image was assigned as the target. Participants were simply instructed to press a button as quickly as possible each time they detected the target image. Critically, unbeknownst to the participants. we also assigned specific roles for some of the other 'non-target' images. Firstly, we randomly assigned one of the stimuli to act as a task-relevant predictive cue. We rigged the presentation probabilities such that the target stimulus was more likely to follow the predictive cue than any other stimulus. We reasoned that participants should be able to implicitly learn this task-relevant predictive relationship to help prepare for the response to the target stimulus (faster response times). 

Figure 2. Behavioural task to compare neural processing associated with task-relevant and irrelevant statistical relationships. Presentation probabilities were randomly assigned for each participant at the start of the experiment [from Stokes et al., 2014]
Critically, we also included a task-irrelevant predictive relationship to test whether learning is specific to task-relevant relationships, or do participants implicitly encode all the regularities that they experience. Because this manipulation was by definition task-irrelevant, there was no behavioural index of learning. However, we could look to the EEG data to compare the neural response to task relevant versus irrelevant predictive relationships.
Figure 3. Reaction times became faster for cued targets 
as participants presumably learned the predictive 
nature of the task-relevant cue [from Stokes et al., 2014]

Results Summary

Analysis of the reaction time data confirmed that participants learned the task-relevant statistical regularity that we introduced into the experiment. Reaction times were faster for cued targets relative to uncued targets (see Figure 3). By definition, there is no behavioural measure for task-irrelevant learning in this task, so we must turn to the EEG data (see Figure 4). Panel A shows the EEG response to cued targets relative to uncued targets as a function of learning (block number) in frontal, central and posterior scalp electrodes. The colour scale shows the difference in voltage between cued and uncued targets, i.e., the effect of the predictive cue on target processing. Towards the end of the experiment (blocks 7 & 8), a positive difference emerges at around 300ms after the presentation of the target. We also estimated the effect of learning by calculating the linear relationship between block number and the EEG response. In panel B, we can see the scalp distribution of this learning effect. In comparison to the robust learning effect of task-relevant predictions, we find no evidence for an effect of block (i.e., learning) on task-irrelevant predictions (Panel C & D). Finally, in Panel E we also plot the time-course of the learning effect for relevant (in blue) and irrelevant (in red) predictions for frontal, central and posterior electrodes, revealing a significant effect of learning relevant predictions (black significance bar, relative to baseline), but not irrelevant learning (relevant>irrelevant in grey significance bar). 

Figure 4. The EEG learning effect for cued vs. uncued targets and cued vs. uncued control non-targets. There was a robust effect of learning for predicted target stimuli (Panel A & B) relative to the task-irrelevant stimulus pairs (Panels C & D). In Panel E, we directly contrast the effect of learning task-relevant and irrelevant predictions in frontal, central and posterior electrodes, revealing a significant difference from around 250ms post-stimulus in central and posterior channels. Horizontal bars indicate significant regression slopes in the target learning condition compared to chance (in black; central: p = 0.053, cluster-corrected, dashed line, posterior: p = 0.0130, cluster-corrected, solid line), and directly compared to the control non-target condition (in grey; central: p = 0.026; posterior: p = 0.045, cluster corrected) [from Stokes et al., 2014]
Finally, we performed the same analysis, but time-locked to the cue stimulus (Figure 5). All the conventions were the same, expect now we are looking at the response to the predictive stimulus (rather than the predicted stimulus). Again, we observed a robust learning effect for the task relevant predictive cue (Panels A,B), but not for the task-irrelevant (Panels A,B and C for the direct comparison).   

Figure 5. Event-related potentials to predictive stimuli: target cue and control non-target cues. All the conventions are the same as Figure 4. Note that there is a significant learning effect of the task-relevant predictive cue (Panel E, in blue), but not the task-irrelevant cue (in red) [from Stokes et al., 2014]


This experiment shows that learning predictive relationships critically depends on the task relevance. In our experiment, participants were not explicitly informed about any of the statistical relationships between stimuli, but simply learned them through experience. Task-relevant predictions clearly benefited behaviour in the task. As participants learned the implicit statistics of the task, they responded more quickly to cued, relative to uncued targets. The learning effect was also clearly evident in EEG activity, consistent with differential processing of predictive, and predicted, task-relevant stimuli. In contrast, there was no evidence for a corresponding neural effect for task-irrelevant predictions, providing strong evidence that the brain prioritises which relationships to learn. Of course, our null effect does not mean that task-irrelevant predictions are never learned or represented, but rather highlights to importance of task-relevance in modulating the learning process. 

As a side note, this experiment also provides a nice example of how we can use EEG to probe cognitive variables without requiring a behavioural response. In many situations, we are interested in how the brain processes non-target information. This presents an obvious challenge for a behavioural experiment: how can we measure processing, without making the stimulus task-relevant? Here, we use EEG to measure the response to task-irrelevant input, thereby providing insights at both the neural and cognitive level (see relevant post here)

Key reference: 

Stokes, Myers, Turnbull & Nobre (2014). Preferential encoding of behaviourally relevant predictions revealed by EEG. Frontiers in Human Neuroscience, 8:687 [open access]

Thursday 9 April 2015

New arrival, keeping us all busy

It has been a while since I have posted anything new, but in the meantime this little guy has arrived in our lives: