Week 7: Connectivity analysis Flashcards
T/F: task-related changes make up for A LOT of brain-related energy consumption
False; they only make up for ~5% of brain-related energy consumption
Resting state fMRI
Looking at BOLD changes in the brain regions while subjects lay in a scanner without performing a task
Low-frequency fluctuations
In task-related fMRI, we usually want to get rid of low-frequency drift (through high-pass filtering and pre-whitening), but in the resting state case, the low frequencies are actually relevant because they may carry information about different brain regions working together.
Resting state fMRI cares about…
studying the correlation between spontaneuos BOLD signals to learn about its intrinsic functional connectivity
Pros of resting state
- removes the burden of experimental design and subject design
- good for studying clinical populations which may not be able to carry out tasks in the scanner
- easy to record
- allows easier comparison between labs
- experiments do not need to be synched
- allows for open-access data share to be easier
Resting state studies have revealed…
…large scales of correlation between brain regions during rest > resting state networks between brain regions during rest.
Resting state networks represent…
…the neuronal baseline activity of the brain. This was found both within and between subjects. These networks are thought to represent functional systems supporing perceptual and cognitive processes.
Regions that are co-activated during tasks…
…tend to show similar resting tate BOLD patterns
Most research on resting-state functional networks has focused on an ultra-low frequency spectrum of…
~ 0.01-0.1 Hz, which is separable (and lower) from respiratory and cardiovascular signal frequencies
According to Joe, resting state scans should last…
15 minutes or longer
Is there consensus whether subjects should be asleep/awake/eyes open/closed in the scan for resting state?
No
What is the main difference in pre-processing task and resting state scans?
Frequency filtering: in resting, we care about certain low frequencies and we do not care about high frequencies > band-passing to 0.01 and 0.08 Hz > anything below or above is filtered out.
Beware!
Non-neuronal physiological signals may interfere with resting state BOLD data
Removal of confounding signals will increase the quality of the data
Common practice in rsfMRI to monitor and correct for confounding signals
Non-neuronal physiological signals may interfere with resting state BOLD data
Removal of confounding signals will increase the quality of the data
Common practice in rsfMRI to monitor and correct for confounding signals
Needs more nuisance variables than task-based
Why?
Nuisance variables are those that may affect the measured result, but are not of primary interest
Anti-correlations
Spontaneous BOLD signals in two brain regions that have a negative Pearson cross-correlation coefficient. NFC mechanisms in the context of network physiology are less understood and have been a subject of debate. Several studies demonstrated that the NFC could be an artifact introduced by a global signal regression procedure, but this is still up for debate.
Adjacency matrix to represent connectivity between nodes
matrix withs 1s representing a connection between nodes and 0s elsewhere
Functional vs. effective connectivity
undirected vs. directed association between two time series
FC (functional connectivity)
- makes statements about the structure of relationship between brain regions (e.g., connections, networks, activation of pathways following a task)
- no assumptions about underlying biology
- region A is correlated with other 3 regions
EC (effective connectivity)
- directed association between two brain regions
- stronger conclusions
- causal effects
- based on biological assumptions
- activation in V1 leads to activation in V5, which leads to activation in PPC
- uses terms as: enhancing, inhibiting, etc. > suggest causality!
EC levels
- time
- trials
- subjects
- studies
Functional connectivity recap
Undirected association between two time series and / or physiological variables
Makes statements about the structure of relationships among brain regions
Makes no assumptions about the underlying biology
Methods to investigate FC
bivariate analysis
seed-based connectivity
beta series
partial correlation
inverse covariance
Seed-based methods
Correlation between time course of a seed region and voxels from other brain regions
We can find regions correlated with seed region
Seed can also be a physiological variable
e.g., heart rate as seed and look at correlation with other regions
Issue with seed-based methods
Hemodynamic lag may not match up between different regions
Time series between different regions may not match up even if the neural activity does
Seed-based methods
Activation-induced correlations
The challenge arises because, during a task, many brain regions may be activated simultaneously. This shared activation effect means that the activity levels in various regions might increase or decrease together simply because they are responding to the task, even if they do not have a direct functional connection.
Addressing activation-induced correlation issue
Beta series: estimate the size of the evoked response on each trial separately at each voxel and then examine the correlation of these responses, rather than the correlation in the entire timecourse.
Example from book:
Beta-series correlation estimates a response for each trial using a separate regressor, as shown in the design matrix in the middle of the figure. The original timeseries for the two regions plotted in the figure are highly correlated (r = 0.953), due to activation-related transients in the signal. When the beta series is estimated using the general linear model with the design matrix shown in the middle, the resulting estimates show that the trial-by-trial response of these two regions is actually uncorrelated (r =0.007). Thus, beta-series correlation can identify
functional connectivity without being sensitive to activation-induced effects.
We only care about amplitude across different trials!
Partial correlation
Correlation between two regions, after the effect of all other regions have been removed
Covariance vs. inverse covariance
Covariance matrix can represent relations between all variables while inverse covariance shows the relations of elements with their neighbors
PCA vs. ICA
PCA compresses information, while ICA separates information. The goal of PCA is reexpressing a dataset in terms of a set of components that are uncorrelated, or orthogonal to one another; the goal of ICA is to detect unknown signals in a dataset, sometimes called the blind source separation
problem
PCA is good for dimensionality reduction, while ICA is better for artifact removal!