Dear FSL experts,
I have a question regarding using prethreshold maksing with randomise and TFCE options in FEAT GUI.
I tried to do a group analysis with randomise with 5000 permutations with TFCE, without and with a bilateral nucleus accumbens mask.
I understand that by using the prethreshold masking, the null-distribution for FWE correction will be different from that without this prethreshold masking, so a voxel's FWE corrected p-value will change depending on maksing or not.
What about the TFCE value itself for a voxel in the mask?
Does the pre-threshold maksing change the "cluster support" for a voxel's TFCE value?
What I mean is when using prethreshold masking with TFCE, does a voxel from outside the mask still contribute to the TFCE value for a voxel within a mask?
Or should the TFCE value of a voxel within the mask only be contributed by other voxels within the mask?
I am assuming the former is the case, as the masking is "prethreshold", so the voxel statistics would be calculated in the same way before the masking anyway.
And I think this is why the thresh_pstat*.nii.gz I get from with and without masking are exactly the same because they are uncorrected p-values?
The result of one particular contrast showed without masking is as followed:
Cluster Index Voxels P -log10(P) Z-MAX Z-MAX X (mm) Z-MAX Y (mm) Z-MAX Z (mm) Z-COG X (mm) Z-COG Y (mm) Z-COG Z (mm) COPE-MAX COPE-MAX X (mm) COPE-MAX Y (mm) COPE-MAX Z (mm) COPE-MEAN
9 45409 0.0142 1.85 3.54 8 -28 -36 -1.68 -26.5 12.8 59.2 2 -38 2 16.1
8 153 0.0378 1.42 3.24 -24 -52 44 -27.3 -52.4 47.6 23.9 -32 -56 52 13.5
7 148 0.0388 1.41 3.24 20 42 -14 22.1 47.5 -13.9 12.5 30 46 -16 8.44
6 79 0.0438 1.36 3.35 50 -18 -26 46.9 -16 -24 12.2 44 -16 -24 9.1
5 50 0.0468 1.33 2.54 42 -52 -12 46.1 -51.3 -14.3 14.5 44 -52 -14 12.9
4 26 0.0462 1.34 2.79 -24 56 -12 -24.5 56.9 -11.4 10.5 -24 60 -10 8.8
3 19 0.049 1.31 2.44 44 -10 -16 41.3 -9.07 -14.3 15.2 42 -10 -14 13.2
2 11 0.0492 1.31 2.58 -24 -46 -6 -24 -46.9 -6.17 16.3 -24 -48 -8 13.2
1 4 0.0494 1.31 2.45 -26 -64 -6 -26.5 -64.5 -6.5 11.9 -26 -64 -8 11.4
The result with the same contrast with maksing is as followed:
Cluster Index Voxels P -log10(P) Z-MAX Z-MAX X (mm) Z-MAX Y (mm) Z-MAX Z (mm) Z-COG X (mm) Z-COG Y (mm) Z-COG Z (mm) COPE-MAX COPE-MAX X (mm) COPE-MAX Y (mm) COPE-MAX Z (mm) COPE-MEAN
2 45409 0.0026 2.59 3.54 -14 10 -10 -0.212 12.6 -4.7 59.2 2 -38 2 16.1
1 153 0.047 1.33 0 -24 -54 40 nan nan nan 23.9 -32 -56 52 13.5
Obviously, cluster 9 and 8 without masking correspond to cluster 2 and 1 with masking.
From the result, I can see that with the same Zmax value of 3.54 (although in different voxel coordinate because of the maksing), the corresponding p-value has changed presumably due to a different null-distribution for FWE related to the masking.
But if the previous is the case, that the null-distribution has changed, then I don't understand why the voxels in cluster 9 without masking and cluster 2 with masking are still the same at 45409?
Especially, in the first place, why does the result with masking still showed a cluster with 45409 voxels which is by itself much larger than the accumbens mask?
Why does voxels outside the mask still being formed into the cluster in the report?
Thank you very much!
Best wishes,
Chen-chia
########################################################################
To unsubscribe from the FSL list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1
|