help
help > RE: No unsmoothed data in analysis
Jul 26, 2015 05:07 AM | Alfonso Nieto-Castanon - Boston University
RE: No unsmoothed data in analysis
Dear Jan,
Typically when entering into CONN already preprocessed functional files you select smoothed normalized volumes (e.g. named sw*.nii). In this case the default setting for extracting ROI-level data (in the GUI this option is defined in the Setup.Functionals tab, in the bottom "functional data for ROI-level analysis" field) is to search first whether alternative volumes exist (in the same folder but named without the leading 's' character, e.g. named w*.nii), and if so use those volumes when extracting average BOLD-signals within your ROIs. The purpose of this is to avoid spillage from neighboring ROIs that could result from spatial smoothing prior to ROI BOLD signal extraction. If you get the "warning: unsmoothed data ..." warning, that means that CONN was not able to locate these files and it will revert to extracting BOLD-signals from the original/selected functional files.
If you don't have access to the original functional data prior to smoothing, or if you do not care too much about the precise boundaries of your ROIs (e.g. when using ROIs defined in a common space like MNI, since certain amount of inter-subject variability in the precise boundaries is expected any way), then extracting ROI data from smoothed functional volumes is a reasonable alternative. The biggest concern will probably be in the subject-specific White matter/CSF ROIs, since these are used for aCompCor to characterize the BOLD signal noise, and contamination from gray matter voxels can potentially introduce fluctuations of interest into this noise characterization (this danger is nevertheless minimized in CONN by the automatic erosion of the White matter and CSF masks).
So, summarizing, it likely does not have a big impact, but if you do have the unsmoothed data available it is still recommended to extract ROI-level BOLD signals from unsmoothed volumes as that will lead to more precisely defined ROI signals and possibly also slightly improved noise characterization in aCompCor.
Hope this helps
Alfonso
Originally posted by Jan Stasinski:
Typically when entering into CONN already preprocessed functional files you select smoothed normalized volumes (e.g. named sw*.nii). In this case the default setting for extracting ROI-level data (in the GUI this option is defined in the Setup.Functionals tab, in the bottom "functional data for ROI-level analysis" field) is to search first whether alternative volumes exist (in the same folder but named without the leading 's' character, e.g. named w*.nii), and if so use those volumes when extracting average BOLD-signals within your ROIs. The purpose of this is to avoid spillage from neighboring ROIs that could result from spatial smoothing prior to ROI BOLD signal extraction. If you get the "warning: unsmoothed data ..." warning, that means that CONN was not able to locate these files and it will revert to extracting BOLD-signals from the original/selected functional files.
If you don't have access to the original functional data prior to smoothing, or if you do not care too much about the precise boundaries of your ROIs (e.g. when using ROIs defined in a common space like MNI, since certain amount of inter-subject variability in the precise boundaries is expected any way), then extracting ROI data from smoothed functional volumes is a reasonable alternative. The biggest concern will probably be in the subject-specific White matter/CSF ROIs, since these are used for aCompCor to characterize the BOLD signal noise, and contamination from gray matter voxels can potentially introduce fluctuations of interest into this noise characterization (this danger is nevertheless minimized in CONN by the automatic erosion of the White matter and CSF masks).
So, summarizing, it likely does not have a big impact, but if you do have the unsmoothed data available it is still recommended to extract ROI-level BOLD signals from unsmoothed volumes as that will lead to more precisely defined ROI signals and possibly also slightly improved noise characterization in aCompCor.
Hope this helps
Alfonso
Originally posted by Jan Stasinski:
Dear CONNers,
could anyone tell me what is the cost of performing roi to roi and seed to voxel analysis despite that error message : "warning: unsmoothed data for subject 1 session 1 not found. Using original functional data instead for ROI extraction". Will the results be somehow less significant or devaluated? I have imported 1st level models from spm for my analysis.
i will be grateful for your help
Jan
could anyone tell me what is the cost of performing roi to roi and seed to voxel analysis despite that error message : "warning: unsmoothed data for subject 1 session 1 not found. Using original functional data instead for ROI extraction". Will the results be somehow less significant or devaluated? I have imported 1st level models from spm for my analysis.
i will be grateful for your help
Jan
Threaded View
| Title | Author | Date |
|---|---|---|
| Jan Stasinski | Jul 25, 2015 | |
| Alfonso Nieto-Castanon | Jul 26, 2015 | |
| Jan Stasinski | Jul 27, 2015 | |
