Connectivity Flashcards
Connectivity, what are we talking about?
Structure-function relationships!
What are the functional properties of certain structures in the brain
One structure in the brain will usually carry out one specific function!
True?
False?
False, most structures in the brain are related to a multitude of functions.
This poses a problem for studying connectivity, as it introduces noise.
Brain mapping challenges: The problem of context dependence
When showing participants stimuli on a screen, the activation patterns in the brain will differ based on factors such as attention and visual context.
Brain mapping challenges: The problem of connectivity
E.g. Amygdala-Hippocampus-FFA example
Activity in most brain areas is modulated by other brain areas, depending on the stimuli.
For example, patients with Hippocampus damage and amygdala damage show increased FFA activation when viewing faces compared to houses.
For face stimuli, however, patients with only hippocampus damage have increased FFA activation when seeing fearful faces vs neutral, which is not the case for people who also have amygdala damage. Thereby we see that amygdala modulates the activation of FFA based on the context.
Brain mapping challenges: Describe the problem of Co-activation vs Functional connectivity
In the brain, multiple voxels may correlate with the stimuli that is being investigated. However, it is hard to determine whether these regions constitute a network or whether they carry out discrete functions.
Because of this problem, it is very useful to go beyond mapping certain functions to certain regions, and investigate functional integration between brain regions.
Functional Conncetivity
The temporal correlation between spatially remote neurophysiological events
Describing the functional connectivity without looking at causation. It is “mechanism-free” because it simply describes the connection but not the mechanism.
For mechanisms use Effective Connectivity
Effective Connectivity
The influence that the elements of a neuronal system exert over another.
Tries to the describe the mechanisms of the brain by establishing causal links between brain regions.
Functional Connectivity Methods: Seed-voxel correlation analyses
Examines how brain activity in a specified region (seed region) is correlated to voxels in other regions.
Each voxel in the seed region is correlated to all other voxels in the brain. By itself, the correlations can’t tell us a lot, but if we introduce an experimental condition we can examine whether the correlation changes based on the condition. Similarly, between-subject studies can examine e.g. whether correlations differ between healthy participants and Alzheimer’s patients.
Limitation: You need to know which voxel is your seed voxel and have an a priori hypothesis. This limits exploratory analysis and you run the risk of sticking to false brain mappings
Functional Connectivity Methods: Principle Component Analysis (PCA)
Performed by recording the variance of individual voxel’s timecourses - giving us a matrix of variation across space and time.
PCA is an algorithm which allows us to examine which voxels explain the most variance in all the other voxels across space and time, and extract these. Thereby, the “principle components” are left, which are the statistically most important areas in the brain.
This is done using Singular Value Decomposition (next card).
Functional Connectivity Methods: Singular Value Decomposition (SVD)
Used in PCA
Singular Value Decomposition (SVD) takes in a matrix of voxel activation over time and space. This is used to create three new matrices:
- Eigenvariates: columns = components, row = variability over time (Measure of variability over time)
- Eigenimages: columns = components, row = variability over space (Measure of variability in space)
- Eigenvalues: columns = components, row = total amount of variance explained (measures of the amount of variance explained by a combination of Eigenvariates and Eigenimages)
By choosing the components with the highest Eigenvalues(explaining the most variance), we extract the “principle components” in the data.
In the picture, the top image is a plot of all 40 voxels, and on the bottom it is reconstructed using only three voxels which explain the most variance.
Effective Connectivity Methods: Psycho-physiological interactions (PPI)
PPI examines the interaction effects of brain activity and experimental conditions.
In the example from the lecture, a simple study shows that while V5 activity is modulated by the experimental condition (Attention), V1 is not. This is descriptive, but doesn’t provide definitive proof that it is actually attention that causes the difference in activation, as it could also be noise or other factor altering V5 activity.
In order to provide a mechanistic link between attention and V5, the timecourse activation of V1 is inserted into the GLM:
V5_activity ~ Attention + V1_activity + Attention*V1_activity
The interaction is thus “psycho-physiological”. In the experiment, the interaction between V1 activity and attention reliably explains V1 activity, which shows that the correlation between V1 and V5 is modulated by attention.
Method lies between functional and effective connectivity analysis because it is still correlation-based and very simplistic. Thus more powerful methods are needed to establish causal links.
Effective Connectivity Methods: Granger Causality
Stated as: Given two timeseries y1 and y2, y1 is considered to be caused by y2 if its dynamics can be predicted better using past values from y1 and y2 as opposed to using past values of y1 alone.
In the example from the lecture, participants had to press either a right or left button based on an auditory cue. They showed that they could predict time course in motor cortex better if they include supplementary motor cortex activation and also auditory cortex activation, which strongly suggests a causal link.
However, in neural terms it is hard to conclude Granger causality, as time series of different brain areas are based on BOLD responses resulting from neurovascular coupling which introduces a lot of noise in the measurements of time series.
Effective Connectivity Methods: Dynamic Causal Modelling (DCM)
In short, build a computational model of the brain system you want to investigate, and optimise it to fit fMRI data. Does the model reflect the brain?
DCM is used to infer causal relationships in brain systems. A brain system can e.g. be the visual cortex. When working with DCM, a “system model” is constructed based on previous research on the area about information flow in the system and potential modulators of the information flow and how the system state changes over time as a response to different inputs.
The system models generally consist of matrices that describe:
1. System connectivity (how different parts of the system are connected)
2. Modulation of connectivity (How the different connections are modulated)
3. System state (the current state of the system)
4. Input (what and where is the input)
All of these are described over time, so the model includes information about how the system state changes according to input.
When these models are defined, data from a real brain is gathered in an experimental setup that reveals the functioning of the “brain system” in question. The fMRI data is then extracted, and a “hemodynamic forward model” is applied to the DCM, in order to convert the system states into artificial BOLD signals. Thereby, we have the a timecourse of the DCM and of the brain, which can be compared. With this, parameters in the DCM can be optimised to fit the actual BOLD responses as well as possible, which allows us to gain insight about the causal structures in the brain system.
