VL 4 Methods

• Spatial resolution —> how fine grained we can look at brain

• temporal resolution —> how frequently the brain's activity is sampled

• interference —> interfering with brain activity in certain areas temporarily -> more than mere correlation

• mobility

• EEG

• MEG

• ERP

& bad spatial resolution

• CT

• MRT

• PET

• fMRI

• Fmri immobile

• DTI immobile

• MEG less mobile

• EEG mobile

TMS

• main magnet —> constant magnetic field

  • along length of MRI scanner—> high field strength

• gradient coils —> spatial encoding

  • additional magnets with varying magnetic fields

  • along different spatial axes

• radio frequency Coil —> disturbs protons from alignment

• magnet influences protons in nucleus of hydrogen molecules

• main magnet aligns protons along the direction of magnetic fielt —> spin around direction of magnetic field

• gradient coil induces another short impulse aligning protons in antiparallel way

• return time differs for each type of matter (tissue)

• oxygen is natural contrast agent

• oxygenated hemoglobin —> blood less magnetic —> less distortions —> stronger signal

• desoxygenated hemoglobin —> blood more magnetig —> more distortion —> weaker signal

• hemoglobin transports oxygen to activated neural regions

Blood oxygenation level dependent contrast

• more activity in brain region —> more metabolism —> more oxygen needed —> more oxygenated blood delivered

• BOLD signal is INDIRECT —> does not measure activity directly, therefore delayed peak (4-6 seconds to maximum) and long time to return to rest level (up to 30 sec)

• signal is slow —> 2 seconds per slice in worst spatial resolution

• setting is challenging -> noise, nothing magnetic can go in, takes 20-40 min

• tradeoff between temporal & spatial resolution

• voxel sizes 3x3x3mm

• Outcome: large & complex 4D datasets

  • 100.000 voxels

  • 1 set (1 volume) takes about 2 sec

  • typically: more than 40 slices per volume

Problem: always multiple activated brain regions

—> how to identify the regions that are involved in specific mental process?

1. measurement under stimulation

2. measurement in control condition

3. generating a difference image

• originally subtraction

• today: statistics

problem: sizes of brains & anatomy of gyri and sulci differ between subjects

—> how to identify whether area activated across all participants

• spatial normalization to standard brain

• outdated: Atlas of Talairach & Tournoux 1988

  • basis: right hemisphere of an older lady —> much smaller than standard brain

• Montreal Neurological Institute MNI brain

  • average brain across 152

Problem: noisy data

—> how to conduct analyses across subjects in noisy data

• 3D spatial smoothing before statistics via 3D Gauss curve

• single voxel results —> sometimes single voxels seem to not be activated —> if that happens across multiple participants effect for whole area may not be significant

• therefore smoothing in order to be able to perform group statistics

—> Consequence: weakening of the signal but improves chance of finding group statistics

• motion correction —> realignment when participants head moves

• activation patterns in typical studies are not strengths per se but may be p-values —> other statistical values!

• borders are the threshholds of significance

y=x*ß+E

Y: Matrix of BOLD Signals (time x voxels)

X: Design Matrix (time x regressors) Experimental design & model of BOLD response

ß: Matrix Parameters (regressors x voxels) —> reflected in brain maps

E: Error Matrix (time x voxels)

• preprocessing

• 1st level analysis: GLM for every participant —> maps of statistical parameters = one ß-map per voxel

• 2nd level analysis: group level analysis

  • t-test, ANOVAs, regression

  • one test per voxel —> statistically problematic, might be up to 10.000 t-tests

• SPM Matlab based

• FSL

• AFNI

• Brain Voyager

• Python

• Blocked designs

• Event-related or single-trial designs

• repeated presentation of same stimulus / same type of stimulus

• two ways

  • 1 experimental condition multiple blocks

  • 1 experimental condition per block

• often relatively fast timing: 1 block typically 20-60 seconds

• statistical power because many repetitions

• greater efficiency —> more stimuli in shorter time

• randomized sequence of stimuli

• multiple experimental conditions

• variable intervals between trials

Advantages:

• psychologically better paradigm because subjects perform decision once —> evoked response by single trial instead of whole block

• trial-specific activation can be estimated

Disadvantage:

• Bold signals are slow, we need time between stimuli to separate different signals

—> can be solved by optimal randomization of trials

• efficiency vs. event-related activation

• psychological plausibility / adequateness

• duration

• behavioral correlates

• statistical power

• dead salmon —> without proper statistical correction (e.g. correction for multiple comparisons) fmri data can show false positives —> brain activity where there is none

• multiple comparisons problem in fmri —> each voxel compared to other brains —> false positives

• unexpectedly high correlations in social neuroscience studies with fmri

  • correlations of 1 unlikely —> brains responses are influenced by multitude of factors

  • problem of non-independence (circularity) —> hypothesis & data selection selecting significantly activated voxels and then correlations on them

• reverse inference not possible —> interpretation ist limited

  • just because amygdala is activated —> person is not necessarily scared

Resting state activity

• constant deactivation of same brain areas under task conditions in 9 PET studies

• Raichle proposal 2006: task positive and task negative network (Default Mode Netwirk)

  • task positive: active during task performance, deactivated during rest

• But: there are multiple resting state networks (7)

1. visual

2. somatomotor

3. dorsal attention

4. ventral attention

5. limbic

6. frontoparietal

7. default mode network