Connectivity: Dynamic Causal Modelling (DCM) for fMRI Flashcards
What are the different type of brain connectivity?
- Anatomical/structural connectivity
- Functional connectivity
- Effective connectivity
What is used to measure axonal connections in animals?
Tract-tracing technique
In humans: only post-mortem processing of the tissue or Diffusion weighted imaging
What is functional connectivity an umbrella term for?
An approach that gives us an estimate of the statistical dependencies between regional time series
- It could be a correlation or derived from ICA
What is the difference between function and effective connectivity?
Functional connectivity is just descriptive - completely ignorant about how the dependencies arises
Effective connectivity - make a claim about the directed influences between neurons or neuronal populations
What is anatomical/structural connectivity?
Presence of axonal connections
What is the idea about Dynamic Casual Modelling (DCM)?
Dealing with a system where we cannot observe the state of interest directly e.g. neuronal activity be it in terms of membrane potential firing rate - it is hidden from us, we can only access it indirectly in terms of some observed measurements that could be a BOLD signal or electromagnetic potential distribution that is measured on the scalp level
What can be done in DCM?
Describe the hidden dynamics of a system in terms of differential equation that are parameterised e.g. theta
Incorporate some knowledge about design perturbation of the system e.g. sensory inputs that is administered as an experimental list
Also describe mathematically how any particular neuronal state translates into a measurement in terms of BOLD or electrophysiology
If we describe such a model for a given measurement –> invert the model, fit the model to the data, estimate the parameters and reconstruct their posterior distribution
What is an example of experiment done on the DCM?
Present visual stimuli and presented them in the visual hemi-field either left or right while the subject is fixated centrally
We are interested in a small system comprising the lingual gyrus on the left and right and the fusiform gyrus also on both hemispheres
We can model the neuronal dynamics in that system using a few simple assumptions and also knowledge we have about the brain
What is the anatomy of the visual system?
If we present the visual system in the periphery of the visual field it will arrive in the contralateral visual cortex
A stimulus in the right visual field will first be received by the left lingual gyrus
The regions are connected ti each other reciprocally within and across hemispheres
What are the two assumptions of DCM?
- All that matters that we are interested in can be summarised by a single number per region e.g. the mean activity in that region [population synaptic activity across neurons in the brain]
- Everything is linear
- Write down equations for each of the areas
- Each area is represented by one number
- Describe the change in activity in area x1 as a linear combination of influences
- The influence that x1 exerts onto itself [self-connection]
- The influence that the second area exerts onto the first
What can the re-arrangement of equations give?
compact form by arranging terms into matrices and vectors
x = Ax + Cu
The matrices
1, Grouped the state changes into a vector [vector of dynamics]
- Grouped the coupling coefficients into a matrix which is the effective connectivity [endogenous connectivity]
- Have vector of system state
- Input parameters - describe how strongly the stimuli affect activity in the primary visual cortices
- Input functions ‘u’ - defined our stimuli
x= Ax + Cu
How can you model the changes in coupling strength by?
Slightly augmenting equations - how the context variable changes the two connections from the right to the left hemisphere
- Mathematically express that by slightly augmenting the previous equations
- if the contextual variable is off = 0 - everything is exactly the same as before
- We are modelling additive changes in coupling strength as a function of our contextual variable
What does DCM model?
Additive changes in connection strength as a function of some controlled variable - some controlled task or context variable
What does DCM allow you to do?
Define a model that describes how several populations of neurons or regions interacts
- First define the region of interest
- Specify the connections between the regions based on understanding of the system
- Specify where in that system perturbations enter [driving input] e.g. visual stimulations
- Induces activity which propagates along connections defined
- Specify which of these connections modulated in time by some controlled variables e.g. attention task
These are the 3 ingredients that map onto the 3 matrices
What are the 3 neural state equations?
A matrix - endogenous connectivity - connection strength per se
B matrices - one for each contextual variables that allow you to change temporally coupling strength
C matrices - encode strength of the driving or input perturbations
What happens if you integrate the equations?
Get time series for x - compare that to the real BOLD signal and based on the discrepancies you can make a choice of how you can update parameters and iterate through that procedure until you cannot further optimise the predictions
What gives an exponential function?
Integration of a first-order linear differential equation
What do Neural models enable us to make?
Inferences about brain circuitry using downstream measurements such as functional magnetic resonance imaging (fMRI)
What can neural models capture?
The mean activity of large numbers of neurons in a patch of brain tissue (Deco et al., 2008)
What is a common application of these models in Neuroimaging?
Assess effective connectivity - the directed casual influences among brain regions - or more simply the effect of one region on another
What is DCM?
A framework for specifying models of effective connectivity among brain regions, estimating parameters and testing hypothesis
What is DCM forward (generative) model?
Conceptualised as a procedure that generates neuroimaging timeseries from the underlying causes (e.g. neural fluctuations and connection strengths)
What does the generated timeseries depend on?
the model’s parameters, which generally have some useful interpretation; for example, a parameter may represent the strength of a particular neural connection
What happens after you have specified a forward model?
One can then stimulate data under different models (e.g. with different connectivity architectures) and ask which stimulation best characterises the observed data
- model inversion (i.e. estimation) - process of finding the parameters that offer the best trade-off between accuracy and complexity of the model
- Hypothesis are tested by comparing the evidence for different models (e.g. with different network architectures) either at the single-subject or the group level
How can you evaluate the evidence for a model?
One needs to average over the unknown parameters, which means model inversion is usually needed prior to model comparison