Linking brain to behaviour
We commonly have detailed information about brain activation, but need new methods and converging evidence to link this activation with the behaviour it is thought to support. Only by drawing this link can we understand how the brain gives rise to cognition.
Our lab is developing three key approaches to make this inferential leap: studying how information coding changes when participants make mistakes, formally relating the details of multivariate patterns to latent variables describing our behaviour, and combining brain stimulation with neuroimaging for causal inference.
Analysis of errors
Using multivariate pattern analysis, we have shown that the multiple-demand (MD) regions prioritise task-relevant information. But is this critical for successful behaviour? To assess this we study what changes in the brain when participants perform a task correctly, compared to when they make mistakes. Surprisingly, the MD regions don’t just fail to code the critical information on error, but instead seem to systematically code the incorrect information, providing a tight link between MD activity and behaviour. In current work we are studying the timecourse of incorrect information coding in MEG, and studying whether the method could be used to predict and prevent errors in high risk situations.
Key Publications: Woolgar et al., 2019 BioXriv
Key People: Hamid Karimi-Rouzbahani, Amanda Robinson (University of Syndey), Anina N. Rich (Macquarie University)
Latent variables in decision-making
Evidence accumulation models are models of decision making in which evidence for a possible decision “accumulates” until a certain threshold is reached. These powerful models allow behavioural responses (like the time and accuracy of a button press) to be decomposed into latent (hidden) variables governing the decision making process, which are typically interpreted as reflecting specific cognitive properties. Work in this stream asks to what extent multivariate patterns of brain activation reflect these latent variables, providing a possible bridge between neural activation and human behaviour. Perceptual decision making can be conceptualised as a cascade of interrelated computations - about the nature of the percept, the application of a rule, the motor response, and so on. We use time-resolved MEG/EEG data to track these computations and relate them to the latent variables derived from evidence accumulation models.
Key People: Dorian Minors, Alessandro Tomassini (University of Cambridge), Tijl Grootswagers (University of Western Sydney)
Combining brain stimulation and neuroimaging
Our ability to attend to information that is currently relevant is
fundamental for cognition. The multiple-demand (MD) network is
thought to underlie this ability and bias processing across the rest
of the brain in favour of what is important to us. Using fMRI, we
have “read out” neural codes from these regions and found that
they do indeed encode the aspects of the task that are most
relevant to us. However, what we don’t know is whether the
neural codes that we observe are causally important – do they
drive processing elsewhere in the brain or result from it, and are
they actually critical for behaviour?
To investigate this, we use the powerful combination of
neurostimulation (TMS) and neuroimaging (fMRI). With the new
TMS-fMRI setup at the CBU, we can modulate a region’s function
while concurrently reading out the impact on neural activity
across the rest of the brain, and observing changes in behaviour.
Current projects focus on stimulating the dorsolateral prefrontal
cortex, a region of the MD network, to understand its specific
causal role in the selection of task-relevant stimuli and rules.
Key publications: Jackson et al. 2020 BioRxiv