Lecture 4: Neuroimaging Flashcards
What is an MRI scanner?
originally developed for structural imaging, more recently also used for functional brain imaging (functional Magnetic Resonance Imaging – fMRI)
What are the benefits of MRI scanners?
- Non-invasive
- High spatial resolution
- Reasonably affordable
What are the magnetic details of an MRI?
- Very strong magnet: usually 1.5 – 7 tesla
- 1.5 tesla = 15,000 gauss
- Earth’s magnetic field strength = 0.5 gauss
- Exeter MRI scanner = 30,000 X earth’s magnetic field strength
- NO METAL IN THE SCANNER!!!
What is a Blood Oxygen level Dependent (BOLD) signal?
- When neurons become active, blood flows to the part of the brain to provide oxygen to fuel the cells.
- Hemoglobin (the iron-containing oxygen transporting protein present in blood) differs in how it responds to magnetic fields, depending on whether it has a bound oxygen molecule
What does an MRI scanner detect?
Small changes in the magnetic field
What does an fMRI do?
Measure blood flow
- the magnetic properties of oxygenated vs deoxygenated blood
What are the steps from raw data to functional brain activation maps?
- Design a task to be used in the scanner
- Collect some data
- Preprocess the data
- Data analysis
- Interpret your results
What is an experimental design of an fMRI?
- BOLD signal is arbitrary: It has no stable baseline.
- Therefore, the most important aspect of any fMRI experiment is that we have to provide both a) an experimental condition and b) a baseline condition.
What is a good baseline condition for an fMRI?
- One that differs from the experimental condition only by the process of interest
- E.g. Face processing.
- Baseline = scrambled faces
What are block vs event-related designs?
- BOLD signal is slow – peaks 4-5 seconds after stimulus onset and takes around 16 seconds to return to baseline
- All fMRI experiments originally employed block designs – long periods of alternating task/baseline performance
What are the disadvantages of block designs?
- Block designs often group together lots of trials
- Limitations (Psychological)
- Highly predictable occurrence of stimuli: subjects know what is coming and may alter strategies accordingly (not always a pro)
- Inflexible for more complex tasks: impact of oddball stimuli? or stimuli or events that occur uncontrollably?
- Ecological validity. Does blocking trials change the psychological process you are interested in?
- Can’t separate trials by performance – e.g. to look at activation associated with correct vs incorrect response
What is an alternative to a block design?
Event related design
What is an event related design?
Trials of different conditions are randomly intermixed and occur close together in time
What are the advantages of event -related designs?
In freeing us from the necessity of block designs, event-related fMRI enables us to design more complex and novel experiments.
- Flexibility and randomisation
- eliminate predictability of block designs
- avoid practice effects
- Post hoc sorting
- (e.g., correct vs. incorrect, aware vs. unaware, remembered vs. forgotten items, fast vs. slow RTs)
- Can look at novelty and priming - rare or unpredictable events can be measured
- e.g., P300
- Can look at temporal dynamics of response
- Dissociation of motion artifacts from activation
- Dissociate components of delay tasks
- Mental chronometry
How is best to collect data in an fMRI?
To reference each point in the brain in 3D space we divide image into cubes or ‘voxels’
- Typical voxel size = 3x3x3 mm
- Refer to each voxel with coordinates (x,y,z)
How to preprocess the data in an fMRI?
Correcting for non-task-related variability in experimental data
- Getting rid of the ‘noise’
What are the steps of preprocessing the data of an fMRI?
- High pass filtering
- Remove low frequency oscillations such as scanner drift that introduce noise into your data
- Standard low pass filter is approx 120 secs
- Motion correction
- Keep track of how the head moves through relaxation versus stress
- Slice time correction
- Temporal difference between slices collected so need to shift back into alingment
- Coregistration
- align functional data with structural data
- Normalisation
- Standardised Space- Montreal Neurological Institute (MNI) space
- Combination of 352 MRI scans on normal controls
- Why normalize? So that we can compare activation across subjects and do group analyses.
- Montreal Neurological Institute (MNI) space
- Spatial smoothing
- Application of Gaussian kernel
- Why smooth?
- Because neurons do not fire in isolation.
- If a neuron fires, neurons close to it will tend to fire as well.
- Smoothing attempts to model the data according to this property of neurons.
How best to analyse data in an fMRI?
- Multiple regression is used to determine the effect of a number of independent variables (our conditions) on a single dependent variable (brain activation)
- For each voxel, we use multiple regression to estimate how closely the BOLD signal correlates with the timecourse of each condition – resulting beta values tell us how close this correlation is.
- Finally perform a contrast – simply a t-test comparing beta condition 1 beta values to condition 2 beta values
- Calculating the t-statistic for each of the 130,000 voxels in the brain is equivalent to asking “in which regions is activation greater in condition 2 than in condition 1?”
- Raw t-map indicates the magnitude of the t-statistic using a colour scale - Yellow = bigger t values
- The threshold we apply defines what we accept as a significant difference in activation between our 2 conditions
- Essentially arbitrary (as it is in any statistical procedure)
What are the effects of difference thresholds on brain activation maps?
- Where do we ‘draw the line’?
- Standard alpha level in psychological research is p < 0.05, i.e. we accept a 5% rate of false positives
What is the correlation for multiple comparisons?
- Brain images divided into up to 130,000 voxels= 130,000 individual t-tests
- Chance of a type 1 error (false positive) increases with every test performed
- With an alpha level of p < 0.05 (standard used in psychological research) with one t-test we have a 5% chance of type 1 error.
- With 100 independent t-tests, we have a 99.4 % chance of type 1 error
- In other words, with 130,000 t-tests and an alpha level of 0.05 we are guaranteed to find some significant voxels simply by chance.
- Very important to correct for multiple comparisons – adjust alpha level (p-value)
What are the two difference approaches to data analysis?
- Whole brain analysis - as we have just seen, examine effects on a voxel by voxel basis across the whole brian
- Region of interest (ROI) analysis - restrict out anaylsis to a particular brain region
What are the advantages of a whole brain analysis?
- Requires no prior hypotheses about areas involved
- Includes entire brain
What are the disadvantages of a while brain analysis?
- Can lose spatial resolution with inter-subject averaging
- Can produce meaningless “laundry lists of areas” that are difficult to interpret
- Depends highly on statistics and threshold selected
- Multiple comparisons problem
What are the advantages of a region of interest analysis?
- Hypothesis driven – avoids meaningless ‘laundry lists’ of activated regions.
- Avoids multiple comparisons problem – data summarised in a single number per subject reflecting mean activation across voxels in ROI.
- Simple – data can be exported and treated as any other type of data, requiring no special software for further analysis
- Generalisable – data can easily be compared across studies, e.g. for meta analysis.