Why does it matter whether a regressor matches the observed BOLD signal or not?

If BOLD signal matches the regressor, region is likely involved in the task, If there's no match, the region is probably not directly related to task

What is special about Brodman areas?

First distinct areas of cerebral cortex based on cytoarchitectonics

How is the "General linear model" used in fMRI in relation to BOLD signal?

Goal: to estimate β values + test if significantly different from zero, Statistically model+ separate task-related BOLD changes from unrelated flux

Standard brain (i.e. a brain template to which individual subject's brains can be aligned) is...

A step called normalization, Often done during preprocessing to make individual brains more comparable

Why are beta estimates important?

Indicate the relationship strength between regressors and BOLD signal, Help identify which brain regions are involved in a task, Provide data for group-level analysis and brain mapping, Used to compare different conditions (e.g., Faces vs. Objects)

Are orthogonalisation of regressors (1) and orthogonal task design (2) the same thing?

No, 1= after data collection (correction) 2= before collection (prevention)

What does the BOLD signal measure?

Changes in blood oxygenation + vascular responses to neural activity, Relative differences in brain activity between conditions (task vs rest)

What is associated with "Time-of-flight imaging"?

Repeated excitation of tissue resulting in reduced signal, Fresh blood = strong signal

Variance inflation factor (VIF) quantifies how much a regressor’s variance is explained by other regressors. Result is..

Very bad (95% of the variance is explained by other regressors)

HRF convolution = convolution (*) of HRF (predefined model) and the neural event(s), but why is it necessary?

Part of the process of creating the design matrix (X) for the GLM, All regressors have to be HRF convolved, since they represent neural events, Accounts for the delay and shape of hemodynamic response to neural event

What does the design matrix (X) in the GLM function entail?

Columns = individual regressors or predictors (e.g, task, baseline, motion), Rows= time points in the scan (expected neural responses for each predictor

What are some analytical considerations to reduce multicollinearity?

Regularized regression, Reduce model complexity (remove predictors that are not needed), Orthogonalization of regressors, Dimensionality reduction

Temporal autocorrelation is also known as...

Serial dependence - past time values can predict future time values

How to solve the problem of multicollinearity?

Compensate for the problem through analytical strategies, We can't, Avoid problem before it occurs through experimental design

How is the traditional view on brain areas shifting with the modern view?

Shifting from functional localisation to network dynamics, Heterarchy rather than hierarchy

Why is motion correction necessary?

Motion artifacts can lead to misleading interpretations of brain function, Significant errors in data compromises accuracy + reliability of images

What is functional connectivity?

Suggests a functional relationship but not necessarily direct communication, Temporal synchronization between brain regions

What is the difference between error term and residual error?

e can be observed and measured, ϵ can't be directly measured or observed

What is true for block designs?

Good for when you expect low and high activity, Usually high statistical power - compare two independent time points

What can we do with task-free paradigms (i.e rest) if there is no task-behaviour or stimuli?

Useful for clinical + developmental studies (participant unable to do task), Allows study of functional connectivity (e.g. how brain regions interact)

pRF models are encoding models, and the same analysis framework can be used to ....

map other non-visual domains to the brain (e.g. numerosity), map visual space to the brain (i.e location + size of pros)

What is true for behavioral tracking?

Behavior predicts brain activity - both cognitive and motor, Measured in fMRI: Microphones, eye tracking, joysticks, pneumographic belts, It is essential for task-based AND task-free approaches

Why would one use a "On-Off paradigm" (Task performance ("ON") and rest/baseline conditions ("OFF"))?

Reduces low-frequency noise by averaging over multiple ON-OFF cycles., Enhances signal detection/ Helps isolate task-related brain activation

What happens during "Transverse magnetization"?

There is proton phase coherence, The precession axis of the protons align orthogonally to B0 (flip 90*)

What is the difference between univariate vs. multivariate approaches?

Univariate = analysing each voxel separately (e.g. GLM), Multivariate = combine information of multiple voxels

How can we define regressors?

Represents an experimental condition or expected brain response over time, Linear combination of hypothetical time-series

Multicollinearity (when two or more predictors/regressors in the GLM are highly correlated) is a problem because ...

β coefficients become unstable (small change in data -> large change in β), Correlated βs reflect relationship with each other, not regressors to data, It makes it hard to tell which one is actually influencing the BOLD signal, It inflates errors, weakens stats, and distorts results

How to reduce the bouncing beta effect?

Check variance inflation factor (VIF) (help detect multicollinearity early), Use regularization techniques (Ridge regression or Bayesian approaches), Combine highly collinear regressors (model as single regressor), Use orthogonal task design (jittering events, randomized condition order)

Field strength (Tesla)

Tesla is the unit for measuring magnetic field strength, Higher field strength leads to a stronger signal in MRI

What is true about the process known as parcellation?

Dividing brain into smaller, functional, or anatomical regions for analysis, Based on anatomical features, functional patterns, or both

What is the second step to turn an encoding model into a decoding model?

Do a mathematical trick called "inversion of the weights"

What is residual error (e)?

Remaining variation unexplained by the modelled predictors (regressors), Difference between the observed values (y) and the fitted values (y-hat)

Another GLM assumption, Variance of residuals is constant across all levels of the data (e.g. time)

Exite protons and flip them into transverse plane (+ a higher energy state), Protons relax + emit energy (detectable RF signal) when pulse is turned off

How do we read contrast angiography & venography?

Use tracers (e.g Gadolinium) to visualize blood flow (arteries --> veins), Different flow rate, wall thickness, and timing of contrast enhancement

Why is unwarping necessary?

The magnetic field is never perfectly homogenous - distorts the picture, Misalignment of brain regions + inaccurate activation maps

Why do we check the residuals (difference between observed BOLD signal and predicted BOLD signal) for normality?

The statistical tests we use (like t-tests and F-tests) assume normality

What can we get if utilising seed-based connectivity?

Brain-wide map showing regions that are functionally connected to the seed, Correlated activity between predefined brain region (seed) + rest of brain

What is associated with non-modular brain theories?

Higher-order cognition, learning, neuroplasticity + complex decision-making, Brain functions emerge from distributed and interconnected neural networks

What is the general idea of modular brain theories?

Brain is organized into specialized, independent processing units (modules), Brain modules operate somewhat autonomously and have domain-specific roles

Multivariate pattern analysis

Combines information across multiple voxels, Set of computational techniques to analyse patterns of brain activity

What is the B0 magnetic field?

A strong magnetic field that aligns the precession axis of body's protons, A strong magnetic field produced from the main magnet of MRI scanner

Contains hundreds of thousands of both neural and non-neural cells, Is on average 8 mm^3, Has on average 500.000 neurons per voxel

How does vasculature affect the BOLD signal?

Spatial localization= "venous bias", deoxygenated hemoglobin most in veins, Temporal dynamic = vessels dilate + affect blood flow slowly -> HRF is slow

What is the first step to turn an encoding model into a decoding model?

Fit the encoding model (to get beta weights)

studbud

89 of the key terms

89 termsStudBud Administration5 plays

Imported from external platform

Flashcards89 cardsFPractice quizPHost liveL

Questions 89

Answers

1
Why does it matter whether a regressor matches the observed BOLD signal or not?
If BOLD signal matches the regressor, region is likely involved in the task
The regressors are used to incline statistical inferences
If there's no match, the region is probably not directly related to task
It doesn't matter as much as the strength of beta estimates
2
What is special about Brodman areas?
The only parcellation map that is not colorised
They are impossible to remember
They are the only fully mapped out parcellation so far
First distinct areas of cerebral cortex based on cytoarchitectonics
3
How is the "General linear model" used in fMRI in relation to BOLD signal?
Goal: to estimate β values + test if significantly different from zero
To make the exam more painful to prepare for
To interpret the signals from the MRI scanner to the computer voxels
Statistically model+ separate task-related BOLD changes from unrelated flux
4
Standard brain (i.e. a brain template to which individual subject's brains can be aligned) is...
A step called normalization
When the anatomical differences are too large to fit the data
Often done during preprocessing to make individual brains more comparable
A comparative model of the human brain, used for generalization
5
Why are beta estimates important?
Indicate the relationship strength between regressors and BOLD signal
Help identify which brain regions are involved in a task
Provide data for group-level analysis and brain mapping
Used to compare different conditions (e.g., Faces vs. Objects)
6
The beta coefficient represents how strongly each regressor is related to the observed BOLD signal for each time point
False
True
7
Are orthogonalisation of regressors (1) and orthogonal task design (2) the same thing?
No, 1= after data collection (correction) 2= before collection (prevention)
Yes, they both combat multicollinearity
Yes, but they are used at different stages
No, but they are interchangeable in cleaning data
8
What does the BOLD signal measure?
Changes in blood oxygenation + vascular responses to neural activity
Individual neuronal spikes + excitatory vs. inhibitory activity
Relative differences in brain activity between conditions (task vs rest)
Absolute neural activity levels (same as LFPs)
9
What is associated with "Time-of-flight imaging"?
Repeated excitation of tissue resulting in stronger signal
Repeated excitation of tissue resulting in reduced signal
High blood flow = stronger signal
Fresh blood = strong signal
10
Variance inflation factor (VIF) quantifies how much a regressor’s variance is explained by other regressors. Result is..
Very good! (R^2 is almost 1 (0.95))
Good (only 5% of the variance is explained by other regressors)
Very bad (95% of the variance is explained by other regressors)
Not good, but not bad (VIF = 20, 20% of the variance overlap)
11
HRF convolution = convolution (*) of HRF (predefined model) and the neural event(s), but why is it necessary?
Part of the process of creating the design matrix (X) for the GLM
All regressors have to be HRF convolved, since they represent neural events
It's part of processing the signal activity from the fMRI for visualisation
Accounts for the delay and shape of hemodynamic response to neural event
12
Any encoding model can be turned into a decoding model, but ....
it is not recommended
not vice versa
it is not commonly done
it is more tricky the other way around
13
Which model do we need to apply to model the raw fMRI data to the following statistical output?
GLM
Ridge regression
MVPA
GLMM
14
What does the design matrix (X) in the GLM function entail?
Conditions across 1 second time intervals
Columns = individual regressors or predictors (e.g, task, baseline, motion)
Rows= time points in the scan (expected neural responses for each predictor
Most of the data from the BOLD signal
15
What are some analytical considerations to reduce multicollinearity?
Regularized regression
Reduce model complexity (remove predictors that are not needed)
Orthogonalization of regressors
Dimensionality reduction
16
Which process is completed with this recipe?
Multicollinearity
Prewhitening
Normalization
Temporal autocorrection
17
Temporal autocorrelation is also known as...
Serial dependence - past time values can predict future time values
Regression correction of temporal voxel activity
Correction for temporal resolution noise
A counterweight to spatial voxel shifts
18
How to solve the problem of multicollinearity?
Compensate for the problem through analytical strategies
We can't
Just don't do experiments
Avoid problem before it occurs through experimental design
19
How is the traditional view on brain areas shifting with the modern view?
From heterarchy to hierarchy
Shifting from functional localisation to network dynamics
Merging functional localisation and networking
Heterarchy rather than hierarchy
20
Why is motion correction necessary?
Motion artifacts can lead to misleading interpretations of brain function
Because individuals don't have the same sized head
Because the MRI scanner moves due to the high velocity
Significant errors in data compromises accuracy + reliability of images
21
What is functional connectivity?
The communication between two neighbouring voxels
Suggests a functional relationship but not necessarily direct communication
The output from resting state fMRIs
Temporal synchronization between brain regions
22
What is the difference between error term and residual error?
ϵ can be observed and measured, e can't be directly measured or observed
e is the only one we care about
They are the same thing
e can be observed and measured, ϵ can't be directly measured or observed
23
What is true for block designs?
Good for when you expect low and high activity
The most used design to date
Usually high statistical power - compare two independent time points
They are useless with multicollinearity
24
What can we do with task-free paradigms (i.e rest) if there is no task-behaviour or stimuli?
Look at it
Useful for clinical + developmental studies (participant unable to do task)
Turn up the magnetisation so people's hair starts to fry from the heat
Allows study of functional connectivity (e.g. how brain regions interact)
25
pRF models are encoding models, and the same analysis framework can be used to ....
map other non-visual domains to the brain (e.g. numerosity)
map regressors across different participant voxels
map visual space to the brain (i.e location + size of pros)
map decoding models to encoding models
26
What is true for behavioral tracking?
Behavior predicts brain activity - both cognitive and motor
Measured in fMRI: Microphones, eye tracking, joysticks, pneumographic belts
It is only possible as long as the participant is cooperating
It is essential for task-based AND task-free approaches
27
Why would one use a "On-Off paradigm" (Task performance ("ON") and rest/baseline conditions ("OFF"))?
To minimize complexities
Reduces signal detection and thus improving statistical inferences
Reduces low-frequency noise by averaging over multiple ON-OFF cycles.
Enhances signal detection/ Helps isolate task-related brain activation
28
What happens during "Transverse magnetization"?
The protons spin really fast
There is proton phase coherence
They get banned from the US
The precession axis of the protons align orthogonally to B0 (flip 90*)
29
What is the difference between univariate vs. multivariate approaches?
Univariate = combine information of multiple voxels
Univariate = analysing each voxel separately (e.g. GLM)
Multivariate = analysing each voxel separately (e.g. GLM)
Multivariate = combine information of multiple voxels
30
How can we define regressors?
The main part of the design matrix
Represents an experimental condition or expected brain response over time
The result of the HRF modulations
Linear combination of hypothetical time-series
31
Multicollinearity (when two or more predictors/regressors in the GLM are highly correlated) is a problem because ...
β coefficients become unstable (small change in data -> large change in β)
Correlated βs reflect relationship with each other, not regressors to data
It makes it hard to tell which one is actually influencing the BOLD signal
It inflates errors, weakens stats, and distorts results
32
How to reduce the bouncing beta effect?
Check variance inflation factor (VIF) (help detect multicollinearity early)
Use regularization techniques (Ridge regression or Bayesian approaches)
Combine highly collinear regressors (model as single regressor)
Use orthogonal task design (jittering events, randomized condition order)
33
The GLM assumption "independence of observations" is the only assumption that is NEVER violated in fMRI GLMs
True
False
34
Field strength (Tesla)
Tesla is a bran we need to boycott
Tesla is the unit for measuring magnetic field strength
Low field strength leads to stable spatial resolution
Higher field strength leads to a stronger signal in MRI
35
What is true about the process known as parcellation?
It is the closest we have come to a natural representative brain
Dividing brain into smaller, functional, or anatomical regions for analysis
Based on anatomical features, functional patterns, or both
It sounds like a vegetable
36
What is the second step to turn an encoding model into a decoding model?
Average the variance to get the weighted beta regression
Do a mathematical trick called "inversion of the weights"
Remove the residual error
Convert the Y-Hat regressors into a inverse function
37
What is residual error (e)?
The remaining data that is modelled by the ridge regression
A theoretical illustration of noise
Remaining variation unexplained by the modelled predictors (regressors)
Difference between the observed values (y) and the fitted values (y-hat)
38
Homoscedascticity
Another GLM assumption
A linear regression of the relationship between regressors and HRF
Is a technical term for dissimilar voxel variance
Variance of residuals is constant across all levels of the data (e.g. time)
39
Radiofrequency pulse
The only thing we can measure in an MRI
Starts the recording of the fMRI image
Exite protons and flip them into transverse plane (+ a higher energy state)
Protons relax + emit energy (detectable RF signal) when pulse is turned off
40
How do we read contrast angiography & venography?
Increasing contrast vectors of a normal fMRI
Use tracers (e.g Gadolinium) to visualize blood flow (arteries --> veins)
Different flow rate, wall thickness, and timing of contrast enhancement
Injecting magnetised ink dyes
41
Why is unwarping necessary?
The magnetic field is never perfectly homogenous - distorts the picture
It is not generally, only when the participant has a movement disorder
Misalignment of brain regions + inaccurate activation maps
Because the transfer from MRI receptor to the computer warps the RF signal
42
Which regression squares betas (ensuring all values are positive, not 0), so large betas are penalized more heavily?
Lasso regression
Ridge regression
OLS regression
Regularized regression
43
Why do we check the residuals (difference between observed BOLD signal and predicted BOLD signal) for normality?
Due avoid invalidity of the observed BOLD signal
To ensure that the beta coefficients represent the relationship
The statistical tests we use (like t-tests and F-tests) assume normality
It is not always necessary, but if the errors are large we need to check
44
What can we get if utilising seed-based connectivity?
Brain-wide map showing regions that are functionally connected to the seed
Correlated activity between predefined brain region (seed) + rest of brain
The anatomical functionality between voxels
Some very good numbers
45
What is associated with non-modular brain theories?
Brain activity measured mainly with naturalistic paradigms
Higher-order cognition, learning, neuroplasticity + complex decision-making
Brain functions emerge from distributed and interconnected neural networks
The previously established modular brain areas are completely ignored
46
What is the general idea of modular brain theories?
Brain is organized into specialized, independent processing units (modules)
A computational representation of the brain
The brain is a set of modules (3D voxels) that vary in activity from tasks
Brain modules operate somewhat autonomously and have domain-specific roles
47
Multivariate pattern analysis
Analyses individual voxels in mass, but separately
Combines information across multiple voxels
Utilise distribution of activity across voxels with lasso regression
Set of computational techniques to analyse patterns of brain activity
48
What is the B0 magnetic field?
A strong magnetic field that aligns the precession axis of body's protons
The second magnetic field
The phase coherence field
A strong magnetic field produced from the main magnet of MRI scanner
49
A fMRI voxel...
Contains hundreds of thousands of both neural and non-neural cells
Is the raw data of the fMRI signal
Is on average 8 mm^3
Has on average 500.000 neurons per voxel
50
How does vasculature affect the BOLD signal?
Spatial localization= "venous bias", deoxygenated hemoglobin most in veins
Temporal dynamic = vessels dilate + affect blood flow slowly -> HRF is slow
Affects the metabolism of neurons and thus the neural activity we measure
It creates noise we need to correct for during preprocessing
51
What is the first step to turn an encoding model into a decoding model?
Convolute the regressors with the HRF
Fit the encoding model (to get beta weights)
Get the raw data
Align the beta estimates to the regressors
52
What is the "Hemodynamic response function" (HRF)?
Describes how blood flow changes in response to neural activity
The event/ experiment we convolute with the regressors
Math model - represents relationship between neural activity + BOLD signal
The BOLD signal
53
Hierarchy vs. heterarchy
Hierarchy is a concept in the animal kingdom
Heterarchy= regions communicate bidirectionally and adjust based on context
Heterarchy modulates lower areas with a top-down system from higher areas
Hierarchy = brain processes information in a top-down or bottom-up manner
54
How has brain areas' function typically been defined?
Studies show local lesions cause selective cognitive deficits
Anatomically
By white old men
fMRI blobs suggesting localised activity
55
What is meant by high ecological validity?
Studies measuring tasks outside of lab, or giving "normal" tasks
Animals and birds and trees and all that
That the data can be applied directly to real world scenarios
Findings are applicable to everyday life, outside of controlled environment
56
When you do an fMRI experiment (e.g tapping your finger/ looking at images), your brain activity changes. What's true?
GLM = Tries to explain how much of that activity is related to the task
Beta estimate = Tells us how strongly a brain region responds to that task
It creates noise
Regressors = Model the expected brain response based on task or experiment
57
What is true for event-related designs?
They are Jan's favourite thing in the world
They don't include free dinner and drinks
More flexibility compared to block design - answer broad range of questions
Usually less statistical power - you need more data
58
Anatomical connectivity
The interaction between brain regions
Only applicable to the cortex!
How fluid moves across the brain regions (time and speed)
Physical, structural connections between different brain regions
59
Heteroscedasciticy = variance in residuals may change over time + standard errors differ between conditions
True
False
60
The BOLD signal is a waveform, yet linearity is a central term in fMRI GLMs. Which assumption is true?
Without HRF* of regressors, the neural-BOLD relationship wouldn't be linear
It refers to the linear relationship between regressors and beta estimates
Relationship between neural events and BOLD signal can be modeled linearly
It is the linear relationship between Y and Y-Hat
61
What is true about nuclear resonance?
fMRI relies on this principle of nuclear resonance to measure BOLD signal
It is the interaction between nuclear spins + external magnetic field
It is irrelevant without a GLM
It is the spin speed of the MRI scanner
62
What is true for T1-weighted (structural) images?
Tissues with short T1's recover more quickly than those with long T1's
Long T1 materials are dark on T1-weighted images, while short are bright
Have high anatomical resolution, high contrast and fewer artefacts
They are preferred for fluid imaging
63
What is true for naturalistic conditions?
They involve walking in the scanner
Result interpretation more difficult + analytical challenges
Discovery of new phenomena more likely, but risk unconstrained exploration
Ecological validity + better generalisability of results
64
What is the core of the Atlas condordance problem?
People don't agree on anything
The instruments are not the same across studies/ parcellations
No ground-truth parcellation of the brain, yet many different brain atlases
They are incomplete for the most part, so overlapping is difficult
65
Preprocessing (cleaning up the signal) includes...
Normalization (anatomical differences --> align with same template brain)
Unwarping (image is distorted --> map + correct magnetic field distortions)
Deleting regressors
Motion correction (head movements shift voxels --> realignment)
66
What are some disadvantages of studying resting state (rs-fMRI)?
It is useless for anatomical connectivity
No control of what participants experience + often difficult to interpret
Imaging artefacts strongly affect connectivity results
It is boring
67
How would you define brain atlases?
Set of regions or parcels for the brain, based on standardized templates
The functional connectivity of fMRI activity seen as a GLM function
I would not, but thank you
The result of parcellation
68
What is true for T2-weighted (functional) images?
They are preferred for anatomical scans
Have superior fluid contrast, but lower resolution
Tissues with short T2 appear darker, while those with long T2 appear bright
A short T2 means that the signal decays very rapidly
69
Many fancy models nowadays are a mixture of encoding and decoding
True
False
70
How can we visualize arteries & veins?
Contrast-enhanced angio- & venography (visualises blood flow)
Susceptibility-Weighted Imaging (SWI) (primarily veins)
Injecting pen ink into the veins and hoping for the best
Time-of-flight imaging (visualises veins, but mostly arteries)
71
What is meant by functional localization?
The idea that the cortical and subcortical areas differ in function
An output of an fMRI modulation process
The idea that different brain areas specialize in certain tasks
The idea that you can draw limbs on the brain
72
What do all of these have in common: orthogonal task design, time-separate conditions, counterbalance, block designs
Only applicable to certain research questions & highly controlled tasks
Experimental considerations to reduce multicollinearity
They are integrated paradigms
They are task-based paradigms
73
Orthogonalization of regressors is done by mathematically adjusting one regressor to remove any shared variance. Why?
More accurate beta → each regressor explains a distinct part of signal
Better interpretability → attribute brain activity to specific conditions
Reduces redundant variance → Avoids inflated standard errors
Because of noise
74
What is done during counterbalancing?
A statistical inference to ontrol for inter-subject variance
Randomizing participant order
Each condition precedes each other condition and itself equally often
Alternating the signal contrasts between participants
75
What is "Longitudinal magnetization"?
When there is no phase coherence
A fleeting pulse
Magnetization that never changes
When the precession axis of the body's protons are aligned to B0/ z-axis
76
What is the error term (ϵ)?
Part of the signal that is modelled by the regressors
Not really useful implication of data, as it is theoretical
Noise, physiological artifacts + factors not captured by the model
Part of the signal that the model fails to explain
77
Why is normalization necessary?
It is a screening to register all images so the computer can analyse them
To compare and analyze across individuals, groups, or different sessions.
High inter-subject variability hinders group statistics and comparisons
Because people have weird heads
78
What is true for task-based paradigms?
Usually involves stimuli and/or participants has to do something
Task-evoked responses often explain shockingly little variance
They can also be task-free
Investigates brain activity associated with assumed cognitive processes
79
Why is venous bias a problem?
Because we are only interested in the arteries
Distorts the temporal profile of BOLD signal (timing of neural activity)
Difficult to distinguish true neural responses and influence of large veins
False positives in areas with large vascular networks
80
Repetition time (TR) is the interval between two successive radiofrequency (RF) pulses in an MRI sequence, and ...
Has a limited number of possible repetitions
Influences the temporal resolution, signal quality, and spatial coverage
Short TR = higher temporal resolution, but reduces spatial coverage
Longer TR = higher SNR, can sample more of the full hemodynamic response
81
Default mode network (DMN)
Activity of the DMN is low during both tasks and rest
Activity of DMN is stronger during tasks, and decrease during rest
Activity of the DMN is high during both tasks and rest
Activity of DMN is stronger during rest, and decreased during tasks
82
What is meant by activity covariation between voxels?
Suggest similar activity patterns, but not necessarily direct communication
The anatomical AND functional connectivity between voxels
Can suggest shared influences but does not imply interaction
The neighbouring activity between two voxels
83
What is precession?
Change in orientation of spin/ rotational axis due to an external influence
Something that predates something else
The magnetisation energy of the proton
Precision*
84
Where does the "bouncing" part of the term "bouncing beta effect" come from?
Beta values "bounce" unpredictably from high to low, each model run
Beta values "bounce" unpredictably due to multicollinearity
When you correct one beta coefficient one becomes very big and another high
Comes from the bounce regression used to calculate them
85
What is the Y-Hat (predicted BOLD signal) used for in context of Y (actual BOLD signal)
Ŷ is what we use to make statistical inferences about our data, y= raw data
A very cute hat for a sensitive guy
If (Ŷ) closely matches (Y) it means our model explains brain activity well
Best estimate of BOLD signal based on the task and regressors, but w/o (ε)
86
What is true for experimental control?
Clearer result interpretation + easier to standardise and replicate
It is limited to max 2 stimuli per task
It is the most widely used design paradigm
Constrained & artificial conditions -> reduced generalisation to real world
87
Inter-subject correlation (ISC) is a type of integrated approach, which can...
Ensure you get to watch movies and get paid 30 euros
Illuminate "general principles" between people's brain activity
Correlate voxel time across participants to find shared brain activity
Help smooth out statistical models
88
Integrated paradigms
Perform resting state and task-based paradigms at the exact same time
Examine both evoked (task-driven) and intrinsic (spontaneous) activity
Combine elements of both task-based fMRI (tb-fMRI) and resting-state fMRI
Involves boats
89
In regularised regression, the penalty term lambda (λ) is added to the GLM to do what?
Control the strength of regularization (acts on the beta estimates)
To bring the beta estimates close to zero
Make the course more agonising to learn
Prevent overfitting and reduce multicollinearity

89 of the key terms Flashcards | StudBud