MEDx Frequently Asked Questions - Statistics
Q: How do I compute the proper statistical threshold?
The Critical Threshold operation under the Functional menu can be used
to determine the z-score threshold that corresponds to a given probability.
This function allows you to specify a statistical image (such as a z-socre
or probability image) and optionally a mask image. The Uncorrected correction
type allows you to specify an uncorrected probability (e.g. 0.05). If
you then hit the return key it will show you the corresponding z-score
for that probability. Clicking on the OK or Apply buttons will apply the
given threshold to the statistical image.
The Bonferoni and Resel corrections can be used to deal with the multiple
comparisons problem. In any voxel-by-voxel statistical analysis a large
number of simultaneous statistical comparsion will be made (one at each
voxel). The Type I error, the probability that the analysis will result
in the classification of a voxel as activated when in fact it is not,
increases dramatically.
The above two corrections are designed to account for this type of error.
Of the two the Bonferoni is much more conservative. It determines the
number of voxels in the mask and divides the uncorrected probability by
this number to arrive at the corrected probability and its corresponding
z-score. The Resel correction takes into account the smoothness of the
data and determines the number of resolvable elements (resels). This method
takes into account the fact that the effective resolution of the data
reduces the total number of independent statistical tests performed in
the analysis. In this correction, the uncorrected probability is divided
by the number of resels to determine the corrected probability. When you
use either the Bonferoni or Resel corrections, the corrected probability
and its corresponding z-score are used to threshold the statistical image.
You can create a mask image by using the "Image Mask" option in the
Intensity Normalization dialog box found under the Functional menu. For
the resel correction you need to enter the FWHM which corresponds to the
the smoothness of the image. Typically, you will use the FWHM of the Gaussian
kernel that you used when smoothing your data for these inputs.
Other FAQ Topics
If you have a question that you would like to see addressed in our list
of Frequently Asked Questions, please contact
customer support.
Back to FAQ Home
|
