Brain Imaging (CD) Flashcards
BRAIN IMAGINING TECHNIQUES
STRUCTURAL - CT - MRI FUNCTIONAL - PET - fMRI
CT
- moderately invasive via X-rays (ionising radiation)
- inexpensive
- widely available
- low spatial resolution (useful clinically NOT scientifically)
MRI
- non-invasive/innocuous via RF fields (radio frequency) to acquire images
- extremely high spatial resolution (1mm^3)
- no1 structural brain imaging choice in neuro research
PET
- moderately invasive via radioactivity
- measures indirect metabolic correlates of neural activity (blood flow/glucose metabolism); can also measure direct synaptic transmission (ie. via labelling receptors)
- high spatial resolution; extremely low temporal resolution
- extremely expensive
fMRI (FUNCTIONAL MAGNETIC RESONANCE IMAGING)
- non-invasive/innocuous
- measures indirect metabolic correlates of neural activity (blood flow/oxygen consumption)
- high spatial resolution (3mm^3)
- low temporal precision as measures slow processes
- moderately expensive
MRI SCANNERS
- originally designed for structural imaging; functional recently
- incredibly strong magnet (1.5-7 tesla)
- 1.5 tesla = 15k gauss
- Earth’s magnetic field strength = 0.5 gauss
- no metal included
BOLD (BLOOD OXYGEN LEVEL DEPENDENT) SIGNAL
- active neurons = blood flow to brain to provide oxygen to fuel the cells
- hemoglobin (iron-containing oxygen transporting blood protein) difs in response to magnetic fields depending on if it has bound oxygen molecule
- MRI scanner (giant magnet) detects these small changes in magnetic field
- fMRI = measures magnetic properties of oxygenated VS deoxygenated blood (brains don’t just “light up”)
FMRI EXPERIMENTAL DESIGN
- Design task to be used in scanner.
- Collect data.
- Pre-process data.
- Analyse data.
- Interpret results.
FED 1: THE BASELINE
- BOLD signal = arbitrary; no stable baseline
- most important aspect of any fMRI exp = provide both experimental/baseline condition
KANWISHER et al (1997) - good baseline = difs from exp condition only via process of interest (ie. face processing)
FED 1: BLOCK VS EVENT-RELATED DESIGNS
- block design = grouping trials together
- event design = dif condition trials randomly intermixed; occur close together in time (allow more complex/novel exps)
- BOLD signal = slow (peaks 4.5s post stimulus onset; takes 16s to return to baseline)
- all fMRI exps originally employed block designs (long periods of alternating task/baseline performance)
FED 1: BLOCK DESIGN LIMITS
HIGHLY PREDICTABLE STIMULI OCCURANCE
- subjects know what’s coming; may alter strategies accordingly (not always pro)
INFLEXIBLE FOR COMPLEX TASKS
- “oddball” stimuli impact OR stimuli/events occurring uncontrollably?
ECOLOGICAL VALIDITY
- does blocking trials change targeted psych process?
CAN’T SEPERATE TRIALS VIA PERFORMANCE
- (ie. to look at activation associated w/correct/incorrect response)
FED 1: EVENT DESIGN POSITIVES
FLEXIBILITY/RANDOMISATION
- eliminate block predictability
- avoid practice effects
POST HOC SORTING
- (ie. correct/aware/remembered/fast VS incorrect/unaware/forgotten/slow RTs)
CAN LOOK AT NOVELY/PRIMING
- (ie. P300)
CAN LOOK AT TEMPORAL DYNAMICS OF RESPONSE
- dissociation of motion artifacts via activation
- dissociate components of delay tasks
- mental chronometry
FED 2: COLLECTING DATA
- 2-3s to collect single volume
- to ref each point in brain in 3D space we divide image into cubes/voxels
- typical voxel = 3x3x3mm (refer w/x, y, z coordinates)
FED 3: PRE-PROCESSING STEPS
- correcting for non-task related variability in exp data (getting rid of “noise”)
1. HIGH PASS FILTERING
2. MOTION CORRELATION
3. SLICE TIME CORRELATION
4. COREGISTRATION
5. NORMALISATION
6. SPATIAL SMOOTHING
FED 3: HIGH PASS FILTERING
- HIGH PASS FILTERING
- remove low frequency oscillations ie. scanner drift that introduce data noise
- standard low pass filter = approx 120s
FED 3: NORMALISATION
- NORMALISATION
- MNI (Montreal Neurological Institute) Space = combo of 352 MRI scans on normal controls
- done to compare activation across subjects for group analyses
FED 3: SPATIAL SMOOTHING
- SPATIAL SMOOTHING
- application of Gaussian kernal
- why? as neurons don’t fire in isolation; if one fires, close ones tend to too
- smoothing attempts remodelling of data according to neuron property
FED 4: ANALYSIS
- multiple regression determines effect of many IVs/conditions on single DV/brain activation
- for each voxel we use multiple regressions to estimate how closely BOLD signal correlates w/time-course of each condition
- finally perform contrast (simple t-test comparing beta condition 1 beta values to condition 2 beta values)
FED 4: T-TEST CALC
- calc of t-statistic p/130k brain voxels = “in which regions is activation greater in c2>c1?”
- raw t-map indicates t-test magnitude via colour scale (yellow = > t-values)
- applied threshold defines STATSIG acceptance in activation between 2 conditions
- essentially arbitrary (as in any stat procedure)
FED 4: THRESHOLD EFFECT ON BRAIN ACTIVATION
- where is the line drawn?
- standard alpha level in psych research = p<0.05 (ie. we accept 5% false positive rate)
FED 4: MULTIPLE COMPARISON CORRECTION
- brain imaged divided to 130k voxels = 130k individual t-tests
- type 1/false positive error chance ^ w/each performed (5%)
- w/100 independent t-tests we have 99.4% T1 chance
- 130k tests + p<0.05 = STATSIG voxels guarantee simply via chance
- v important to correct for multiple comparisons (adjust alpha level/p-value)
FED 4: WHOLE BRAIN ANALYSIS
- examining effects on voxel by whole brain voxel basis
POSITIVES - requires no prior hypotheses about involved areas
- includes entire brain
LIMITS - can lose spatial resolution w/inter-subject averaging
- can produce meaningless “laundry list of areas” difficult to interpret
- depends highly on stats/selected threshold
- multiple comparison problems
FED 4: ROI (REGIONS OF INTEREST) ANALYSIS
- restrict out analysis to particular brain region
POSITIVES - hypothesis driven; avoids meaningless lists of activated regions
- avoids multiple comparisons problem; data summarised in single number p/subject reflecting mean activation across ROI voxels
- simple; exportable data treated as any; no special software for further analysis
- generalisable; easily comparable data across studies (ie. meta analysis)
LIMITS - easy to miss things elsewhere in brain
- not always simple how to define ROIs
FMRI LIMITS
- correlative data so can’t say region activated during task/function = essential for it; need converging TMS/neuro-psych evidence for stronger causation inferences)
- low temporal fMRI resolution; need converging EEG/TMS evidence for finer grained temporal info
FMRI PROCESS
- High pass filtering.
- Slice time correction.
- Motion correction.
- Coregistration.
- Normalisation.
- Smoothing.
- 1st level data analysis.
- 2nd level data analysis.
- Thresholding.
- Overlay on anatomical template.
- Data interpretation.
