fMRI Data Analysis using ANOVA
Analysis of Variance (ANOVA) can be used to detect activations in
fMRI data, without specifying the shape of the time course
expected. ANOVA is commonly used to analyse evoked potentials, and is
suited to the type of experiment where the same stimulus is presented
a number of times, and to which the shape of the response is the same
following each presentation, but this shape is not predictable.
The FMRIB MEDx implementation is described here.
Running MEDx ANOVA
-
For the script to successfully run you need to have a group page open in
MEDx, containing the images to be analysed.
-
From the FSL menu, select FMRI,
then Model-Free ANOVA.
-
There is a choice of two ways to run the ANOVA. The one you should
choose depends upon whether the stimuli were presented at regular or irregular
intervals.
-
For stimuli presented at regular intervals select the tab labelled Cyclic
ANOVA.
-
For stimuli presented at irregular intervals, select the tab labelled Triggered
ANOVA.
Cyclic ANOVA
-
Thresholding is done, such that F-scores are only calculated in regions
containing the brain. Enter a percentage of the maximum pixel intensity
in the first image of the volume that is to be used as a threshold (The
default is 10%).
-
If the first images of the group were acquired as saturation scans then
these should be ignored from the analysis. Enter the number of volumes
to ignore in the appropriate box.
-
Next enter the number of volumes per complete cycle. The number of
cycles in the data set will be computed from the total number of volumes
in the group.
-
Press OK or Apply to run the analysis.
-
Two images are produced by the analysis. The first is an atatistical
parametric map of the F-scores that result from the analysis. The
second is a transformation of this map to Z-scores. This is done
by calculating, on a voxel by voxel basis, the p-value associated with
the F-score, and the calculating the Z-score that corresponds to that p-value.
Triggered ANOVA
-
Thresholding is done, such that F-scores are only calculated in regions
containing the brain. Enter a percentage of the maximum pixel intensity
in the first image of the volume that is to be used as a threshold (The
default is 10%).
-
Enter the number of events in the appropriate box.
-
Enter the number of volumes following each event to include in the analysis.
This should cover the return to baseline after the activation.
-
Enter the name of a text file containing the volume number (first volume
is zero) of the start of the event.
-
For example, an experiment where upon a visual cue the subject presses
a button. The time to acquire a single volume is 3 seconds, and the
cues were given at 6, 21, 42, 60, 78 and 102 seconds. The text file
should look like
The number of events would be 6, and the volumes per event would be 5,
to include 15 seconds following the activation.
-
The text file can be edited by pressing the Edit button. The
text editor window must be closed before proceeding, for any changes
made to the file to take effect. Simply saving the text file is not
enough when editing the trigger file in this way.
-
Press OK or Apply to run the analysis.
-
Two images are produced by the analysis. The first is an atatistical
parametric map of the F-scores that result from the analysis. The
second is a transformation of this map to Z-scores. This is done
by calculating, on a voxel by voxel basis, the p-value associated with
the F-score, and the calculating the Z-score that corresponds to that p-value.
Theory behind using ANOVA for fMRI Analysis
For the theory behind ANOVA see section
6.3.5. of my PhD thesis, or
Clare, S., Humberstone, M., Hykin, J.,
Blumhardt, L., Bowtell, R. and Morris, P. (1999)
Detecting Activations in Event-Related fMRI Using Analysis of
Variance Magnetic Resonance in Medicine. 42 1117-1122.
Stuart Clare 18.2.00