fMRI Experimental Design

Question 1

What is the purpose of an experimental control?

Answer 1

Removing the effect of confounding variables
e.g., testing the effect of a drug: active drug vs placebo drug

Question 2

What are the types of statistical inference?

Answer 2

Between-group comparisons
Within-group comparisons
Linear regression/association
Mixed designs
Advanced methods e.g., multivariate analysis, real time fMRI

Question 3

What is the difference between a between-group comparison and a within-group comparison?

Answer 3

Between-group comparison ==
Independent-sample t-test
i.e., Two groups of pps are recruited (patients vs healthy adults)
Every subject experiences only 1 condition

Within-group comparison ==
Paired-sample t-test
i.e., same pps assigned to two conditions (experimental and control condition)

Question 4

What is a mixed design?

Answer 4

Two groups and two conditions e.g., baseline and follow up (within), active drug and placebo (between)

Question 5

What are the different types of fMRI experimental designs?

Answer 5

Task-based fMRI - BOLD signal during a certain task
Task-free fMRI - BOLD signal in the absence of a task

Block design
Event related design

Question 6

What is a block design?

Answer 6

Consists of several discrete epochs of on-off periods, with the ‘on’ representing a period of stimulus presentations, and the ‘off’ referring to a state of rest or baseline

These blocks are alternated throughout the experiment to ensure that signal variation from small changes in scanner sensitivity, subject movement or attention shifts have a similar effect on the signal responses associated with each of the different states

Question 7

What is event-related design?

Answer 7

Each task is presented individually for a short amount of time e.g., 3s
In this way, tasks can be more randomised, rather than being blocked together by condition

Question 8

What is the best control condition for a sub-vocal word reading experimental condition?

Answer 8

Sub-vocal pseudoword reading

Experimental condition = BRAIN, LATHE, HOURGLASS
Control condition = ARBIN, HETAL, SHOLGARUS

this is better than nonwords - they are pronounceable, and look like they could be real words

Question 9

What can we measure using fMRI task-free designs?

Answer 9

When we do ‘nothin’ we normally engage in forms of introspective and abstract forms of cognition which can be measured in fMRI
e.g., remembering, prospection (looking forward to the future), theory of mind

Question 10

What are fMRI task-free designs/resting-state fMRI used to study?

Answer 10

To study the correlations among regions

== Functional connectivity of brain networks

Question 11

What is region based functional connectivity?

Answer 11

Pick a region of interest
Calculate a mean time course of the entire area
What are the areas whose activity correlates with the average time-course of this area?
Cross-sectional/longitudinal/correlational studies

Question 12

What is a correlational design?

Answer 12

Investigates relationships between variables without the researcher controlling or manipulating any of them

A correlation reflects the strength and/or direction of the relationship between two variables

Question 13

Why do you need to be cautious with correlational studies?

Answer 13

Due to spurious correlations - where two variables are statistically related/correlated, but there is no causal link between them

Question 14

What causes spurious correlations?

Answer 14

Often caused by a third variable that is not being accounted for (confounding variable)

Question 15

What is a mixed design?

Answer 15

A study that combines features of both a between-subjects design and a within-subjects design

Question 16

What are the advantages of MRI?

Answer 16

• Versatility (structure and function)
• Simple statistics
• In vivo methodology
• Longitudinal designs - can repeat scans overtime due to lack of harm (like radiation in x-rays)
• Relatively safe and cheap compared to other techniques

Question 17

What are the limitations of MRI?

Answer 17

• MRI is rather devoid of biological information - techniques are rather distant from what is actually happening at the biological level
• Temporal resolution is ‘suboptimal’
• Counter-indications
  
  • Metal
  • Pregnancy

Question 18

What is real-time fMRI?

Answer 18

Real-time fMRI relies on very rapid data transfer and analysis within a few seconds of data collection

Question 19

What are the applications of real-time fMRI?

Answer 19

Used clinically for conditions such as locked in syndrome where the patient has full capacity but it paralysed
Novel brain-machine interface
Coupling of the contents of conscious experience and brain activation
Planning neurosurgical interventions
Neural feedback

Question 20

Using real-time fMRI what can the BOLD signal be used as?

Answer 20

Using the BOLD signal from the brain as feedback to control localised neural activation
- Nonpharmacological therapy e.g., for patients with affective disorders
- Overcomes limitations of other feedback methods e.g., skin conductance responses and EEG
- Directly targeting specific brain regions with superior spatial resolution

Question 21

What is a neurofeedback paradigm?

Answer 21

fMRI neurofeedback is a type of biofeedback in which real-time fMRI signals are used to self-regulate brain function

fMRI neurofeedback techniques use multivariate analysis of a particular brain region to induce a specific activation pattern in the targeted region

Question 22

How can neurofeedback paradigms be used as treatment for people with depersonalisation disorder?

Answer 22

Cannot read other people’s emotion
Have impairment in the anterior part of the insular and underactivation of this area
Can use neurofeedback to increase the activation of the anterior insular cortex
If we can feedback the information from the brain region to the patient in real-time, we can ask them to try two different mental strategies which will increase that brain region’s activation

Question 23

What is the setup of neurofeedback in fMRI?

Answer 23

1. First have to identify the experimental ROI (area where you want to increase activation)
2. Also choose a control region (somewhere we know is not involved with the task)

Question 24

Why is a control region needed for real-time feeback?

Answer 24

We don’t want people to increase or decrease their BOLD response globally across the brain e.g., holding their breath, we want to specifically change the experimental brain region
Therefore, we need to have a control region to look at the difference between the ROI and control region

INCREASE BLOCK=
ROIexp(BOLDincrease - BOLDdecrease) - ROIctrl(BOLDincrease - BOLDdecrease)

DECREASE BLOCK =
ROIexp(BOLDdecrease - BOLDincrease) - ROIctrl(BOLDdecrease - BOLDincrease)

Question 25

What cognitive strategies can be used to increase blood flow in neural feedback?

Answer 25

‘Recall and relive personal memories of situations and events in which you were very physically aroused. These could be pleasant or unpleasant’

‘Become aware of the tension of the muscles in your face, or limbs, the sensations created by the noise and vibration of the scanner, heartbeat or breathing rate’

Question 26

What cognitive strategies can be used to decrease blood flow using neural feedback?

Answer 26

‘Count back from 100 in 3s’

Question 27

What is multivariate pattern analysis?

Answer 27

A set of methods that analyse neural responses as patterns of activity - looking at the activation of different voxels as a multivariate pattern

Motivated by machine learning and AI - enables the advance machine learning techniques such as pattern classifiers

Has been proven to be more sensitive and more informative about the functional organisation of the cortex compared to univariate approaches

Question 28

What is the principle of multivariate pattern analysis?

Answer 28

In univariate analysis the research question asks will there be different areas of activation in response to two different types of stimuli (i.e., face compared to house)

In contrast, multivariate pattern analysis uses reverse inference - based on the activated neural pattern, we will be able to predict which visual stimuli the subject perceived

Question 29

Why is multivariate analysis better than univariate analysis?

Answer 29

Univariate voxel-wise analysis relates psychological or physical dimensions to the activation of single voxels and thus can fail to map the neural basis of experimental variables and conditions when these variables have a distributed multidimensional effect on activation

MVPA measures that take into account information from multiple voxels, may be necessary to answer whether a region codes for a particular psychological dimension or experimental condition

Question 30

Explain how MVPA works

Answer 30

fMRI data is recorded whilst the subject sees a large number of the stimuli to be discriminated

The acquired data is then split into a training data set and a testing data set

The data from the training set is entered into a pattern classifier which attempts to detect features in the neural patterns that distinguish the two types of trials from one another

Next, the classifier is presented with unlabelled data from the testing set and based on the patterns it detected from the training data set, it attributes the most likely label to each of the testing trials

For each stimulus the classifier guess is compared to the correct stimulus label and classifier performance is calculated as the percentage of correct guesses

The higher the percentage, the higher the indication of consistent differences between the patterns in that brain area (i.e., the two stimuli have been perceived differently)

Question 31

What is one consideration when using MVPA?

Answer 31

The two stimuli being compared in MVPA must be completely independent of each other in order to make any conclusions of generalisability

Question 32

What is RSA?

Answer 32

Representational similarity analysis
== multivariate method that can be used to extract information about distributed patterns of representations across the brain

Responses are placed into a representational similarity matrix (i.e., if you have 3 picture stimuli, they can be placed into a 3x3 matrix to compare each of the stimuli)

Pearson’s r correlation coefficient is calculated for each of the comparisons of stimuli and the resulting BOLD activation pattern

Question 33

What did Kreigeskorte et al. (2008) study?

Answer 33

Used representational similarity analysis to compare activation patterns of faces and houses

Computed representational dissimilarity matrices which characterise the information carried by a given representation in a brain or model

Pp given an experimental condition - either looking at pictures of a face or a house

Record the activity patterns in the brain

Compute dissimilarity to create a representational dissimilarity matrix

Question 34

What further studies did Kreigeskorte et al. (2008) do?

Answer 34

Created further representational dissimilarity matrices using the following categories: human (body/face), animate (not human, body/face), inanimate (natural/artificial)

Look at the BOLD responses and the pattern of activity between different objects presented and then correlated the pattern from every object to every other object

Recorded in the human inferior temporal cortex
Findings = naturally forms different clusters in the representational dissimilarity matrix
Inanimate objects most similar to inanimate objects in the matrix same with human objects most similar with human objects in matrix

Question 35

What did Kriegeskorte et al. (2008) find when they recorded from the early visual cortex instead of the inferior temporal lobe?

Answer 35

Found a very different organisation within the representational dissimilarity matrix - there are no clear clusters of correlation between the same objects

Question 36

What does the representational dissimilarity matrix in the early visual cortex tell us?

Answer 36

In the early stages of visual processing, the brain is not interested in object categories - it doesn’t differentiate objects coming from different categories e.g., animals or humans
Brain is more interested in low level features e.g., whether the object is round or if they are the same colour etc

Question 37

What did Kriegeskorte et al (2008) find when they did the same experiments in monkeys?

Answer 37

The representational dissimilarity matrix is very similar to the human matrix
Surprising considering the neural representation between different brains is very different

Question 38

How can representational similarity analysis be used in a clinical context?

Answer 38

Memory impairments in dementia

Decreased activation at medial temporal lobe compared to healthy controls
Increased activation at medial temporal lobe for high-risk individuals comparing to low risk controls during encoding but decreased activation during recall

Question 39

Explain set up using RSA to test activation of the MTL in Alzheimer’s patients

Answer 39

Scan areas involved in memory e.g., hippocampus
Present two different pictures to the pp during the scan e.g., two different types of houses
Record the activity pattern when the two houses are presented
Give a delay and then present only one of the houses again
Would expect to see an activity pattern more similar to the pattern seen when first presented the same house

Hypothesis for pps with dementia is that they would have less reliable episodic reinstatement meaning the activity patterns during the two phases will not be similar

Question 40

What were the findings of encoding-retrieval similarity?

Answer 40

Found no significant difference in recognition memory performance, but
Spatial location related episodic memory impairments in family history of dementia