help
help > RE: Normalized ALFF import values / Z-map display
Feb 7, 2023 10:02 PM | Alfonso Nieto-Castanon - Boston University
RE: Normalized ALFF import values / Z-map display
Hi Jeff,
The values imported as 2nd-level covariates when using the "import values" option in a second-level analysis of normalized ALFF measures will be simply normalized ALFF scores for each subject at each cluster/area (just the z scores associated with the raw ALFF value at each voxel, averaged across all voxels within each cluster; note: these z scores are just normalized scores computed directly from ALFF values, but despite the misleading name they are different from Fisher-transformed Z values which are computed directly from r values -see https://www.nitrc.org/forum/message.php?... for details-).
Regarding the normalized vs. raw ALFF values, that is a very interesting observation. I agree with you that this is likely at least partly influenced by the denoising procedure, which regresses aCompCor components derived from White Matter and CSF areas, likely producing a greater reduction in the overall BOLD variance in these areas compared to Gray Matter areas (although perhaps a portion of that difference in variance between gray matter and other areas exists even before denoising or when using a different denoising strategies? not sure about this). In any case, yes, that difference indicates that the raw ALFF values are consistently higher in gray matter areas compared to white matter and CSF areas for each subject, and since there is roughly the same volume in GM compared to WM+CSF, the zero-value of the normalized ALFF measure seems to then fall right at the boundary between those areas. I imagine that this difference should also be clear/visible in the raw ALFF measures if you display the effect-sizes (e.g. average ALFF values across all subjects), as the images you were showing are focusing on T-values, which may probably be somewhat confounded by differences in variability across subjects. Just my two cents. Let me know what you find out.
Best
Alfonso
Originally posted by Jeff Browndyke:
The values imported as 2nd-level covariates when using the "import values" option in a second-level analysis of normalized ALFF measures will be simply normalized ALFF scores for each subject at each cluster/area (just the z scores associated with the raw ALFF value at each voxel, averaged across all voxels within each cluster; note: these z scores are just normalized scores computed directly from ALFF values, but despite the misleading name they are different from Fisher-transformed Z values which are computed directly from r values -see https://www.nitrc.org/forum/message.php?... for details-).
Regarding the normalized vs. raw ALFF values, that is a very interesting observation. I agree with you that this is likely at least partly influenced by the denoising procedure, which regresses aCompCor components derived from White Matter and CSF areas, likely producing a greater reduction in the overall BOLD variance in these areas compared to Gray Matter areas (although perhaps a portion of that difference in variance between gray matter and other areas exists even before denoising or when using a different denoising strategies? not sure about this). In any case, yes, that difference indicates that the raw ALFF values are consistently higher in gray matter areas compared to white matter and CSF areas for each subject, and since there is roughly the same volume in GM compared to WM+CSF, the zero-value of the normalized ALFF measure seems to then fall right at the boundary between those areas. I imagine that this difference should also be clear/visible in the raw ALFF measures if you display the effect-sizes (e.g. average ALFF values across all subjects), as the images you were showing are focusing on T-values, which may probably be somewhat confounded by differences in variability across subjects. Just my two cents. Let me know what you find out.
Best
Alfonso
Originally posted by Jeff Browndyke:
Are the values imported as 2nd level covariates
from a simple "All Subjects" by condition (e.g., baseline TP) model
of normalized ALFF signal, Fisher Z or beta values? If beta
values, how would one obtain the Fisher Z values?
Lastly, I was comparing the raw ALFF map montage to the normalized ALFF montage (see attachment) and it looks like the gaussian distribution for the normalized ALFF signal is largely centered at the grey/white matter junction? Is this due to how the grey, white and CSF masks are handled in the noise reduction preprocessing? What accounts for the pretty clear demarcation relative to the raw ALFF map?
Thx,
Jeff
Lastly, I was comparing the raw ALFF map montage to the normalized ALFF montage (see attachment) and it looks like the gaussian distribution for the normalized ALFF signal is largely centered at the grey/white matter junction? Is this due to how the grey, white and CSF masks are handled in the noise reduction preprocessing? What accounts for the pretty clear demarcation relative to the raw ALFF map?
Thx,
Jeff
Threaded View
| Title | Author | Date |
|---|---|---|
| Jeff Browndyke | Jan 7, 2023 | |
| Alfonso Nieto-Castanon | Feb 7, 2023 | |