FMRI Ian Cameron

25 important questions on FMRI Ian Cameron

What are the general steps/types of visualization/analysis you can do with your fMRI data?

Activation map - signal time course - signal magnitude in ROI - connectivity (correlation, interaction, influence)

What questions can you answer with the signal time course?

What did the fMRI response look like?
Was there a difference between conditions?
(Baseline needs to be defined)

What questions can you answer with the signal magnitude in ROI?

Was the brain region more/less active depending on different conditions/different subject groups/different brain regions?
  • Higher grades + faster learning
  • Never study anything twice
  • 100% sure, 100% understanding
Discover Study Smart

What is an advantage of analyzing fMRI data as signal magnitudes in ROIs?

It is convenient for standard statistics.
It avoids many pitfalls associated with false positives in fMRI.

Give for examples of sample artifacts.

Ghosts, metallic objects, hardware malfunctions and spikes.

As what will you find your data after the scan on the computer?

As a series of image volume files (DICOM) format. The signals you are looking for are very small changes in intensity (brightness) across scans, about 1-4% increases.

What is a DICOM?

Digital Imaging and communications in medicine.
A single file that has information about the data (name, image characteristics, properties of scan that created it) and multiple dimensions of image pixel data. For MRI this is gray scale pixel intensity.

What is the spatial and temporal resolution of fMRI data?

Relatively low spatially - ~2-4mm voxels.
Relaitvely high temporally (few seconds for whole brain, less than one sec with more modern sequences).

What are the general steps of/things you have to do as preprocessing?

Porbably account for slice-acquisition.
Must account for subjet movement.
Must account for drift in the signal and any other soures of noise that could affect your statistics.
Have to decide on how to display your data (native subject space?, standard/group averaged spaced?).
Have to decide on how to smooth you data.

Why are the first scans (dummy scans/disdaqs) discarded?

The first scans are discarded as the MR signal is not in steady state yet. (T1 does not recover completely, but reaches a steady state of recovery after some scans.)

What are two common orders of the preprocessing steps?

Image reconstruction (scanner) -> Distortion correction (scanner) -> Motion correction <-?> Slice timing correction -> Temporal filtering ->
FSL) Spatial smoothing -> Statistical analysis -> Spatial normalization
SPM) Spatial normalization -> Spatial smoothing -> Statistical analysis

What is the difference between the two ways to correct for the slice time?

Data is analyzed in time points corresponding to an image volume made up of slices. The acutal images can not be taken at the same time.
Therefore one either corrects for that at the preprocessing stage (SPM) or at the model stage (FSL). In preprocessing, data is interpolated as if every slice was taken at the same time. Otherwise slices acquired later in TR would appear to have a shofted (earlier) response. In the model approach, the model is shifted rather than changing the data. This requires a more complex model of the signal depending on the slice.

What are the two different descending slice time methods and their disadvantages?

Sequential descending and interleaved descending.
The disadvantage of the sequential approach is that the excitation of one slice may carry over to the next one (slices are not prefectly rectangular). The interleaved approach offers a gap between two neighbouring slices to avoid that problem. However, if the subject moves throughout the scan, a slice may capture another part of the brain than before. That part of the brain will be in a different level of excitation, causing bad data.

How is motion correction in general done?

Volumes are aligned with a target volume (the functional volume that is closest in time to the anatomical image) using three rotations and three translations as parameters in certain algorithms.

What are 3 examples of transformation/interpolation methods used for motion correction?

Nearest neighbour
Linear interpolation
Spline

What type of motion can motion correction algorithms correct for and what not?

Can
Head motion which leads to spurious activation
Brain regions of interest which move in position over time 

Cannot
Mass motion artifacts (moving large objects that create changes in magnetic field map --> can lead to signals around the edges.
Movement between and within slice acquisition
Non-linear distortions and drop-out due to inhomogeneity of the magnetic field
Interpolation artifacts due to resampling

What is additional motion correction (FSL:ICA AROMA)?

Uses independent component analysis (ICA) to decompose voxels*time data intio space*time course components.
These components can be identified (algorithm or manual) as
-task/signal: lower frequency, blob-like in space and occurs in gray matter
-noise/motion: higher frequency, spikes, ring-like in space and on the edge of brain or in the ventricles

What is normalization and how does it (in general) work?

Normalization warps your (anatomical or functional) image to match standard template brains using affine transformations (3 translations, 3 rotations, 3 shears and 3 scalings in general possible). This process can be helped by segmentation. Other images like gray/white matter or CSF masks or other scans like T2 weighted anatomicals can be used for that.

What is spatial smoothing and what does it do?

A Gaussian filter (kernel) is applied resulting in a weighted average across neighbouring voxel intensities. Therefore, high frequencies are removed and replaced with information in a larger spatial scale. Signal from individual voxels are thus replaced and those stretching across neighboruing voxels preserved. 
This helps to account for spatial variability cross subjects.

What are recommendations for spatial smoothing?

Usual recommendations are 1 t 2 times the voxel size (~4mm-8mm at full width half max FWHM) of kernel)

What is temporal filtering used for?

Sometimes there is a negative linear drift (the MR signal decreases over time) in the signal. Then a high pass filter is used to cut these low frequencies out. It is not common to use low pass filters.

What is global intensity normalization?

Image intensity values are arbitrary units. Interest lies in difference to a baseline value. So every voxel is normalized to some mean value, taken from the whole data set (session, subject).

Why is MRI data 4 dimensional?

A voxel is 3 dimensional and also a corresponding intensity, summign up to 4 dimensions.

How is MRI data be represented in matrix form?

Usually, every voxel represents a column in which the BOLD signal is given per time point. Time thus represents the rows.

How is a design matrix structured?

In a design matrix the columns are the regressors, the rows the time points and the values are the predicted hemodynamic responses.

The question on the page originate from the summary of the following study material:

  • A unique study and practice tool
  • Never study anything twice again
  • Get the grades you hope for
  • 100% sure, 100% understanding
Remember faster, study better. Scientifically proven.
Trustpilot Logo