search

mwaskom / ziegler

fMRI Reporting Webapp
BSD 2-Clause "Simplified" License
3 stars 1 forks source link

Cluster size #10

Closed dgdillon closed 10 years ago

dgdillon commented 10 years ago

In the Activation Peak Table Ziegler generates--is "Prob" equal to number of voxels in the cluster?

mwaskom commented 10 years ago

It's the probability taken from FSL's Harvard-Oxford atlas, which is a probabilistic atlas. The cluster table shows you the maximum probability region at each peak and the probability associated with that. So something like "Middle Frontal Gyrus - Prob 45" means that the peak falls within the MFG for 45% of the subjects in the database, but it might be in the IFG for 30% of the subjects and in white matter for 25%. (You can use the atlas in FSL view and it will give you the full report for each voxel you click on, which is nice).

The Harvard Oxford atlas is easy to use programmatically, and having a probabilistic estimate based on macroanatomy is vastly preferable to something asinine like saying "this peak is in BA45". But the atlas primarily uses gyrus-based definitions and for a lot of the things I care about I've found sulcal-centered region definitions to be much more informative. So I'm not necessarily sure it's the best way to characterize patterns of activations, but it should still give you some idea of where the blobs are.

dgdillon commented 10 years ago

I wish I could retract that dumb question. Here's a better one: where can I find the number of voxels in each cluster?

dgdillon commented 10 years ago

Also, I completely agree on the merits of a probabilistic estimate for macroanatomy--I'm a big fan of the Harvard Oxford atlas in FSLView.

mwaskom commented 10 years ago

Hm I am not sure that information gets written out in text by FSL anywhere.

You could look in the zstat1_threshold_seg.nii.gz file which has the cluster index value at each voxel. But this is after the watershed thresholding process which tries to pull apart subregions of large contiguous clusters based on the local maxima, it won't actually correspond to the cluster size that was used for GRF-based thresholding.

dgdillon commented 10 years ago

Okay, I'll play around a bit and see what I find--will let you know if I discover anything particularly useful. Thanks for your detailed and helpful answers, they are much appreciated.