Lecture fMRI data analysis Flashcards
From scans to blobs
Terminology - Subjects
Participants that participate in the experiment
Terminology - Session
Measurement
Terminology - Run
Image you receive of the measurement
Terminology - Volume, slices & voxels
The brain (image) is made out of a volume, that is made up of slices, that is made out of voxels (3D pixel = volume element)
Field of View (FOX)
Field of the image that you see, you can see different slices inside the FOV.
Thinner slices -> higher resolution
fMRI signal - Functional images (T2*)
The grey value in a voxel is the BOLD signal. There are about 20-25.000 voxels.
The spatial resolution is about 2-4 mm and the temporal resolution is seconds (time series; every few seconds)
fMRI signal - Anatomical image (T1)
High(er) resolution (~1 mm) and be able to distinguish different tissue types
Used for normalization, localization (analysis)
EEG signal
Signal from an electrode on the scalp (compared to a voxel in fMRI); 32/64/128 electrodes
Low spatial resolution (cm) and time series with high temporal resolution (ms)
fMRI analysis
Preprocessing:
- Registration
- Normalize, smooth
Model specification/estimation
Statistical inference
fMRI analysis - Preprocessing
- (Slice-timing correction)
- Motion correction (realignment)
- Co-registration anatomy-functional scans
- Normalisation to standard brain (MNI space)
- (spatial) Smoothing
→Data ready for analysis
fMRI analysis - Preprocessing: slice-timing correction
If you want to correct the analysis to the fact that the signal of the slices is not acquired at the same time (this is only a problem if you have a quick design)
fMRI analysis - Preprocessing 1: registration
Realignment aka registration
- Motion correction (within the same image)
- Intrasubject registration (within subject)
- Correct with 3 translations (x, y, z) and with 3 rotations (pitch, roll, yaw)
-> Problematic when movement is bigger than voxel (displacement: rotation around anterior commissure)
Coregistration
- Intrasubject, intermodality registration
- Coregister anatomical MRI with functional MRI (different images with different contrasts)
fMRI analysis - Preprocessing 2: normalize, smooth
Spatial normalization: bring all participants into the same space
- Intersubject registration (between subjects)
- Register subject anatomy to atlas space:
MNI- or Talairach space
- Localization
Spatial smoothing:
- Blur data with Gaussian kernel (6-10 mm
FWHM)
- To satisfy random field theory assumptions
- Averages out noise/fluctuations (>SNR)
- For intersubject analyses
fMRI analysis - Model specification: design matrix
Design matrix: a mathematical description of your experiment
E.g. ‘stimulus on = 1’ , ‘stimulus off = 0’ : a
simple ‘block design’
Visual depiction of what the task looked like. We use a linear model for each voxel. Use expected shape of BOLD response.
Model: Y = X * B + E, in which Y is the observation, X the predictor, B(eta) is a parameter estimate and E is the error
fMRI analysis - Model specification
fMRI data are voxel time series (3D-time=4D).
You should reflect the hemodynamic response function (HRF) induced by neuronal events. The HRF is convolved with the design matrix, so this is an estimate.
Estimate for each voxel how much variance of the
fMRI signal (BOLD response) our convolved
parameters can explain.
-> Express in a Statistical Parametric Map (SPM)