Print

Print
Hi Daniel,


On 14 May 2014 07:51, Yang, Daniel <[log in to unmask]> wrote:
Dear FSL group,

Under the Post-stats, when we choose Cluster thresholding, we can use a Z statistic threshold to define contiguous clusters (http://fsl.fmrib.ox.ac.uk/fsl/fsl4.0/feat5/detail.html).


Yes
 
I also see that the Z statistic we choose determines the significance of the voxel, as seen in the color ramp bar, automatically generated. 

No no. If you choose cluster thresholding what happens is that the z-map is thresholded at the level you choose (say, 3.1), then a p-value is calculated for every cluster. This new map of p-values is thresholded at, say, 0.05 (so that only the significant ones survive), and used to mask the original z-stat image. The ramp bar shows the z-stats, and because all z-stats below 3.1 were thresholded out in the first step, then there is nothing below that level to be shown.


So I am a little bit confused here, wondering what Cluster thresholding is, and how it is different from Voxel thresholding. How is the Z-statistic being used to define contiguous clusters?


For the significance at the voxel level (i.e., not using clusters, but still considering the dependencies between the voxels using the random field theory), there is no z-threshold to be specified by the user, only a p-threshold, but that's used in an entirely different manner.
 
All the best,

Anderson


 

Thank you!
Daniel

-- 
Daniel (Yung-Jui) Yang, Ph.D.
Postdoctoral Researcher
Yale Child Study Center
New Haven, CT