help
help > RE: Intensity Normalization: Where and When?
Aug 17, 2021 09:08 AM | AmirHussein Abdolalizadeh - Tehran University of Medical Sciences
RE: Intensity Normalization: Where and When?
Dear Alfonso,
Thanks for your detailed response. So, you mean that there is no need for intensity normalization and I can proceed my analysis into denoise, and this intensity differences may not cause any issues in the statistical analysis we may follow?
Bests,
Amir
Originally posted by Alfonso Nieto-Castanon:
Thanks for your detailed response. So, you mean that there is no need for intensity normalization and I can proceed my analysis into denoise, and this intensity differences may not cause any issues in the statistical analysis we may follow?
Bests,
Amir
Originally posted by Alfonso Nieto-Castanon:
Dear
Amir,
That's an interesting question. There is no voxel-specific intensity normalization step that is typically applied to the functional data. There is a session-level scaling that is intended to re-scale the data to PSC units (this is controlled by the 'BOLD signal units' option in the Setup.Options tab) but this is simply a global scaling factor (i.e. it applies the same scaling factor to all voxels) so it will not change those distributions. That said, as most functional connectivity measures are scale-invariant (e.g. the correlation between two voxel BOLD timeseries will be independent of the scale of these timeseries) pre-scan normalization or other intensity-normalization procedures will have no effect on those measures (there are a few exceptions to this, mainly the subject-specific singular value decomposition used potentially for denoising in voxel-to-voxel measures, or the principal component decomposition used within White and CSF ROIs also during denoising, which are not scale-invariant and would be affected by the choice of using pre-scan normalization and/or other manual intensity normalization steps, but I do not know of prior studies looking at the efficiency of these steps in the context of different intensity-normalization procedures)
Hope this helps
Alfonso
Originally posted by AmirHussein Abdolalizadeh:
That's an interesting question. There is no voxel-specific intensity normalization step that is typically applied to the functional data. There is a session-level scaling that is intended to re-scale the data to PSC units (this is controlled by the 'BOLD signal units' option in the Setup.Options tab) but this is simply a global scaling factor (i.e. it applies the same scaling factor to all voxels) so it will not change those distributions. That said, as most functional connectivity measures are scale-invariant (e.g. the correlation between two voxel BOLD timeseries will be independent of the scale of these timeseries) pre-scan normalization or other intensity-normalization procedures will have no effect on those measures (there are a few exceptions to this, mainly the subject-specific singular value decomposition used potentially for denoising in voxel-to-voxel measures, or the principal component decomposition used within White and CSF ROIs also during denoising, which are not scale-invariant and would be affected by the choice of using pre-scan normalization and/or other manual intensity normalization steps, but I do not know of prior studies looking at the efficiency of these steps in the context of different intensity-normalization procedures)
Hope this helps
Alfonso
Originally posted by AmirHussein Abdolalizadeh:
Dear all,
After analyzing and visualizing the functional preprocessed data (montage; attached) of my subjects, I realized most subjects show dark-inside fMRI scans (hollow-resembling) while others do not. Checking their raw fMRI data showed that the "Prescan Normalization" was not active in their acquisition. I wonder whether there is any need for intensity normalization for the fMRI acquisitions, and whether it is done by "denoising" or not. I tried to do that on one of the subjects. Interestingly, before denoising there was a right-leaning skewness in the probability distribution which converts to a normal distribution after that.
Any help is appreciated.
Bests,
Amir
After analyzing and visualizing the functional preprocessed data (montage; attached) of my subjects, I realized most subjects show dark-inside fMRI scans (hollow-resembling) while others do not. Checking their raw fMRI data showed that the "Prescan Normalization" was not active in their acquisition. I wonder whether there is any need for intensity normalization for the fMRI acquisitions, and whether it is done by "denoising" or not. I tried to do that on one of the subjects. Interestingly, before denoising there was a right-leaning skewness in the probability distribution which converts to a normal distribution after that.
Any help is appreciated.
Bests,
Amir
Threaded View
| Title | Author | Date |
|---|---|---|
| AmirHussein Abdolalizadeh | Aug 8, 2021 | |
| Alfonso Nieto-Castanon | Aug 15, 2021 | |
| AmirHussein Abdolalizadeh | Aug 17, 2021 | |