help > Distribution of voxel-to-voxel connectivity values looks very similar pre- and post-denoising
Showing 1-2 of 2 posts
Sep 19, 2019 06:09 PM | Willem le Duc
Distribution of voxel-to-voxel connectivity values looks very similar pre- and post-denoising
Hi all,
I've run my preprocessing steps and about to move on to denoising, but I've noticed something strange. The distribution of connectivity values in the original looks very similar to the distribution predicted for after denoising. Whereas in the tutorial videos, the distributions are hugely skewed before denoising, the distributions of my original data already all have roughly a mean value of zero before denoising (see attached plots).
Does this indicated an error in the preprocessing steps somewhere, or a problem with my scans? Or is it nothing to worry about?
Note: I didn't use the default preprocessing pipeline. My analysis is volume-based but calls for subject-space, unsmoothed data, with confounds covaried out. We're preprocessing and denoising in CONN but using FSL and R for the rest. My pipeline was as follows:
Best,
Willem
I've run my preprocessing steps and about to move on to denoising, but I've noticed something strange. The distribution of connectivity values in the original looks very similar to the distribution predicted for after denoising. Whereas in the tutorial videos, the distributions are hugely skewed before denoising, the distributions of my original data already all have roughly a mean value of zero before denoising (see attached plots).
Does this indicated an error in the preprocessing steps somewhere, or a problem with my scans? Or is it nothing to worry about?
Note: I didn't use the default preprocessing pipeline. My analysis is volume-based but calls for subject-space, unsmoothed data, with confounds covaried out. We're preprocessing and denoising in CONN but using FSL and R for the rest. My pipeline was as follows:
- Functional realignment and unwarp (subject motion estimation and correction)
- Functional slice-timing correction
- Functional outlier detection (ART-based identification of outlier scans for scrubbing)
- Functional Direct Coregistration to structural without reslicing (rigid body transformation)
- Functional Segmentation (Grey/White/CSF segmentation)
- Structural Segmentation (Grey/White/CSF estimation)
Best,
Willem
Sep 26, 2019 03:09 AM | Stephen L. - Coma Science Group, GIGA-Consciousness, Hospital & University of Liege
RE: Distribution of voxel-to-voxel connectivity values looks very similar pre- and post-denoising
Dear Willem,
The denoising looks fine to me. I guess you are working with healthy volunteers, and if you look at the machine parameters there might be some filters set (such as prescan normalize or hamming filter) which are partially correcting already a bit the connectivity, hence why you observe this. The filters might hide some of the connectivity, hence it's advised to disable these (except prescan normalize which allows for more homogenous subcortical voxels intensity and thus analysis).
So if you want to tweak further, you might want to have a look at your MRI machine's parameters, but otherwise I think the analysis should be fine as it is.
Hope this helps,
Best regards,
Stephen
The denoising looks fine to me. I guess you are working with healthy volunteers, and if you look at the machine parameters there might be some filters set (such as prescan normalize or hamming filter) which are partially correcting already a bit the connectivity, hence why you observe this. The filters might hide some of the connectivity, hence it's advised to disable these (except prescan normalize which allows for more homogenous subcortical voxels intensity and thus analysis).
So if you want to tweak further, you might want to have a look at your MRI machine's parameters, but otherwise I think the analysis should be fine as it is.
Hope this helps,
Best regards,
Stephen