fMRI Flashcards

Question

Explain what happens when an RF tuned to a certain frequency is applied

Answer 1

The protons it is applied to spin out of their usual alignment ● When they relax back into their usual alignment they release energy which emits the magnetic resonance (MR) signal ○ How the pulses are applied (in what timing and pattern) determines the type of image that is produced (t1, t2, etc.)

Answer 2

angle by which the RF pulse disrupts the spin of the protons measured in degrees or radians and it can be adjusted by changing the RF pulse's strength or duration, to optimize the BOLD signal

Answer 3

**Physics perspective: **Time between RF pulses applied to the same slice. **fMRI perspective: **Time between the start of one whole-brain volume and the next.time between RF pulses; crucial time interval between each volume acquisition

Answer 4

The number of slices × the time to acquire each slice. More slices = longer TR unless

Answer 5

delay between that RF pulse and MR signal reading. It determins T2* sensitivity.

Answer 6

BOLD is a T2 effect; TE must be set near the T2 of tissue (~30 ms at 3T) to maximize sensitivity to oxygenation changes.

Answer 7

A complete 3D picture of the entire brain taken at a single moment, created by stacking up all the individual (horizontal) slices acquired during the specific moment in time, sequentially

Answer 8

A slice is a 2D cross-section of the brain. -when a fmri scanner takes a ‘picture’ it cannot capture the whole 3D brain at once, instead it takes a series of these flat 2D slides sequentially - two types of slice capturing: in order (ascending or descending) = sequential acquisition, or alternating = interleaved acquisition. The choice between one or the other depends largely on the type of scanner (e.g. Siemens vs Philips)

Answer 9

Slices are the 2D pictures, and a Volume is the complete 3D brain composed of all the slices collected over 1 TR

Answer 10

When maximizing reliability and consistency of most RSNs (attention, auditory, DMN). Eyes-open no-fixation is best for primary visual network.

Answer 11

**Short tr:** have lower signal to noise ratio, and are more suited for event-related designs. Are more susceptible to physiological noise. **Long tr:** More used in block designs; better signal to noise ratio, but more sensitive to motion.

Answer 12

**Short TE:** stronger signal, because the signal is collected before it decays :( less contrast = change in active/inactive regions not as clear yet (BOLD response takes time to develop) **- Long TE:** strong contrast :( weaker signal

Answer 13

Based on field strength (Tesla) and repetition time

Answer 14

The specific area of the participant’s body that is captured in the MRI image; it is the “window” captured by the scan, which has to be selected before the scan

Answer 15

larger area = larger pixels and lower resolution smaller area = smaller pixels and higher resolution)

Answer 16

having a larger FOV can increase scan time, because there are more voxels of data being sampled

Answer 17

because equal dimensions in x/y/z reduce bias in spatial normalization and smoothing

Answer 18

when a voxel contains more than one tissue type, blurring signal interpretation (e.g. gray matter and white matter)

Answer 19

1. slice timing correction 2. realignment/motion correction 3. coregistration 4. normalization 5. smoothing

Answer 20

Technique that acquires multiple slices simultaneously to reduce TR and increase temporal resolution.

Answer 21

Higher temporal resolution but increased leakage artifacts and potential noise amplification.

Answer 22

It reduces cross-talk between adjacent slices caused by RF excitation overlap.

Answer 23

The fact that in a typical fMRI acquisition the 2D slices of a volume are not acquired simultaneously within one TR. Instead, the 3D image made within one TR (eg. 2 seconds) is made of multiple 2D slices, those slices are made at varying times (eg. slice 1 at 0s, slice 5 at 1.5s, slice 10 at 2.0s).

Answer 24

solves the problem presented by interleaved slice acquisition (which is done to control cross-slice excitation), when djacent slices of the brain acquired at non-adjacent time points

Answer 25

- Reason: Help with temporal accuracy. Also depends on experiment type and setup (eg. TR length).

Answer 26

Because STC assumes that voxel locations are stable across time. Doin motion correction after STC avoids mixing motion artifacts with timing interpolation.

Answer 27

It estimates and corrects for head movements occurring during scanning by rigid-body aligning each volume to a reference volume (typically the first or mean image) using 6 parameters: x/y/z translation and pitch/roll.yaw rotation.

Answer 28

Because fMRI relies on consistent spatial alignment between voxels across time, even small movements can distort results and introduce false correlations. The whole idea of localizing brain structures in fMRI hinges on the assumption that each voxel is in the same position throughout the whole scan So we have to account for movement in the data (remove movement artifacts)

Answer 29

Motion creates spurious changes in voxel intensity that mimic task effects, introduces systematic differences between conditions, and correlates with group membership (children, patients), producing false activations.

Answer 30

The 6 estimated motion parameters (plus their derivatives/expanded terms) included in the GLM as nuisance regressors to model residual motion-related signal variation.

Answer 31

Realigning each brain image (volume) to a reference image - usually the mean, first, or last volume - using a rigid-body transformation. This accounts for three translations (x, y, z) and three rotations (pitch, yaw, roll). By aligning all the functional images together We quantify how much alignment is necessary - so how much each volume deferred from the reference We end up with movement parameters that we can incorporate in our analysis later on to control for movement And we can create a mean functional image to use later on in the coregistration

Answer 32

- quantify amount of movement in motion parameters to be used later on - create a mean functional image to use later on in the coregistration

Answer 33

Aligning the subject’s mean fMRI image (low resolution, T2) to their high-resolution anatomical T1 image. This ensures functional data overlays correctly on anatomy.

Answer 34

Because T1 and T2-weighted images have very different contrast (bright vs dark CSF, inverted tissue intensities), requiring intensity-based algorithms that maximize mutual information.

Answer 35

Warping each participant’s brain into a standard space (e.g., MNI) using nonlinear transformations, enabling voxelwise group comparisons.

Answer 36

To make it possible to compare data across participants by reducing intersubject variability and facilitate reporting in the form of standard (x,y,z) coordinates. This means brain activity or anatomy can be analyzed and averaged at the group level.

Answer 37

Group maps become anatomically misaligned, producing false positives/negatives because activation is smeared or misplaced across subjects.

Answer 38

Spatial averaging: replaces the signal values of each voxel with a weighted average of the surrounding voxels using a small filter (called a Gaussian kernel)

Answer 39

- Reduce random noise in the data, - Increase the signal-to-noise ratio (SNR) - Makes brain activity patterns more similar across participants (since everyone’s brain anatomy differs a bit). - Helps the data meet the statistical assumptions used later in analysis (like the General Linear Model). - Makes activity maps more comparable between participants

Answer 40

Increase SNR Increase normality of the data

Answer 41

This can blur important fine details, so the kernel size must be chosen carefully, usually around 4-10 mm, depending on voxel size and study goals

Answer 42

To model each voxel’s BOLD time series using the GLM and estimate how strongly each task condition explains the signal for that participant, producing β (beta) maps and contrast maps for group analysis.

Answer 43

A statistical model that explains each voxel’s BOLD timecourse as a combination of predicted responses to experimental events (after HRF convolution) plus noise. The GLM estimates a beta value for each predictor, and contrasts of these betas reveal which voxels respond more to one condition than another.

Answer 44

It predicts how much each experimental condition and nuisance factor contributes to the BOLD signal at each timepoint and estimates those contributions for every voxel independently.

Answer 45

The sequence of signal intensity values measured from the same voxel (or ROI) across consecutive time points during the scan. It reflects how the BOLD signal in that location changes over time in response to neural activity, noise, motion, and physiological fluctuations.

Answer 46

Y = Xβ + ε **Y**is the observed BOLD time series **X** is the design matrix (task regressors convolved with the HRF plus nuisance regressors to give predicted BOLD responses) **β** are the estimated parameter weights **ε** is the residual noise.

Answer 47

By multiplying events/stimuli with HRF to produce predicted "convolved" BOLD signal

Answer 48

Whether a voxel responds more strongly to one condition than another, revealing which brain regions are engaged by specific task conditions.

Answer 49

the specific way study conditions are contrasted/analyzed in order to accurately extract BOLD signal information to your variables of interest – model specification is how you analyze the data; which contrasts, baselines and parameters are chosen

Answer 50

One method of model specification (A-B) Shows the difference in brain activation between a target and control task

Answer 51

One type of model specification Contrasting multiple factors at different/multiple levels, so the conditions would be contrasts themselves. allows to test both main and interaction effects.

Answer 52

One type of model specification Used when a variable changes continuously (ex. increasing task difficulty). Tests whether brain activity increases or decreases linearly with that variable.

Answer 53

A matrix that encodes the predicted BOLD response for each task and nuisance regressor at every timepoint. It represents the experimental design (after HRF convolution) and forms the X in the GLM equation.ally showing what happened in the experiment and at what time. It is represented as the X in the GLM equation.

Answer 54

A value estimating how strongly a regressor explains a voxel’s BOLD time series. Larger β = stronger relationship between that condition and the voxel’s activation.

Answer 55

Whether a voxel responds more to one condition than another (e.g., β_faces > β_houses), revealing condition-specific neural activity.

Answer 56

A linear combination of beta weights (e.g., [1 –1]) that tests a specific hypothesis such as Condition A > Condition B. Produces a contrast map for each subject.

Answer 57

Voxelwise values showing the estimated difference between conditions for that participant (e.g., Faces – Houses). These maps are inputs to second-level analysis.

Answer 58

Regressors included to model noise, such as motion parameters, physiological signals, session drifts, and outlier volumes, so they do not confound task-related betas.

Answer 59

When two regressors overlap too much in predicted time courses, making it difficult to distinguish their contributions (unstable β estimates).

Answer 60

Each column represents a predictor (regressor), such as a task condition, convolved with the HRF, or a nuisance/control variable (e.g., motion). Each row represents a time point (one TR), corresponding to each acquired brain volume.

Answer 61

The numbers in each column of the design matrix represent the predicted BOLD response for a condition at each timepoint, showing how much that condition is expected to contribute to the voxel’s signal at that moment.

Answer 62

A regressor added to the GLM that models how the BOLD response varies with a continuous trial-by-trial variable (e.g., intensity, rating, reaction time).

Answer 63

Whether the voxel’s activation increases, decreases, or remains constant as the parametric value increases across trials.

Answer 64

Adjacent timepoints are not independent because the BOLD response is slow. Prewhitening or temporal filtering corrects this violation of GLM assumptions.

Answer 65

Including too many regressors (especially correlated ones), limiting the model’s ability to estimate β reliably and increasing noise.

Answer 66

The temporal correlation of different BOLD signals within different areas of the brain.

Answer 67

FC shows intrinsic brain networks. E.g. sensory networks, attention.

Answer 68

FCs shows that the relationship between different regions of the brain changes based on the task.

Answer 69

A method for assessing how functional connectivity between two brain regions changes depending on the psychological context or task condition. Testing whether functional coupling between areas is stronger or weakter depending on different psychological contexts or task blocks.

Answer 70

Context-dependent changes in connectivity between the seed region and other regions across the brain.

Answer 71

A model-based method that estimates the directed (causal) influences that brain regions exert on each other and how these effective connections are modulated by experimental conditions. ## Footnote DCM is about effective connectivity (directed causal influence), not functional connectivity (correlation).

Answer 72

Resting State Networks (RSNs) are characterized by large-scale patterns of functional connectivity that are consistently observed across individuals, sessions, scanners, and analytic approaches when no explicit task is being performed.

Answer 73

Participants are instructed to keep their eyes closed and not think about anything or fall asleep. Alternatively, participants keep their eyes open and fixate or don’t fixate on objects in the visual field.

Answer 74

Eye fixated condition preferred (exception: when monitoring primary visual network connectivity is more reliable with eyes open without fixation.)

Answer 75

Eyes open fixated or non fixated condition to be used

Answer 76

the probability of correctly rejecting the null hypothesis when a true effect exists (i.e., 1 – Type II error).

Answer 77

Sample size (n) True effect size Statistical threshold / Type I error rate (α)

Answer 78

combines the first-level results from all participants to test whether an effect is reliable at the group level.

Answer 79

First-level analysis estimates the GLM separately for each participant to determine how each voxel’s BOLD time series relates to the experimental conditions.

Answer 80

fMRI analyzes tens of thousands of voxels simultaneously. If you run a statistical test on each voxel independently, most “significant activations” will be false positives simply due to chance.

Answer 81

A very strict correction for the multiple comparison issue. It is used to ensure the entire brain map has a less than .1% chance of containing false positives (given a p-value of .001). It is often over conservative and may prevent true results from being detected. ## Footnote Voxel-based thresholding

Answer 82

A less strict correction for the multiple comparison issue. FDR controls the proportion of false positives among all voxels declared significant, making it a less conservative multiple-comparisons correction than FWE and more sensitive to true effects. ## Footnote Voxel-based thresholding

Answer 83

On the one hand, FWE is better when looking to have a high confidence in the results. On the other hand, FDR is used more when wanting to explore brain areas and detecting broader patterns. ## Footnote Voxel-based thresholding

fMRI Flashcards

(111 cards)