Hello,
I have a problem with the denoising of my resting state images.
So i have [b]one subject[/b] that i want to compare to [b]36 other healthy subjects[/b].
I have [b]2 groups of healthy controls[/b], with [u]different RT[/u] and [u]different acquisition modalities[/u] (one had multi echo).
My issue is that when I preprocess and denoise these resting state in conn, the denoising step is not good, especially for the [b]CSF compound[/b] : for example here is 1 subject (see attached file), the denoising is really bad, and when i raise the threshold, there is explained variance by the CSF almost everywhere except in the ventricules... I specify that my ROI images for grey/white matter and CSF are good enough, and aligned with MNI space for the "bad" subjects".
I will now explain how I manage the preprocessing:
- As one group had a multi echo acquisition, the echo fusion is made by MEICA via AFNI, and it does the slice timing correction and realignment before the fusion; the other group and my subject of interest have simple echo acquisition and then had realignment and STC on SPM.
- I have tried to put the functional scans once they were realigned and STC in conn, so conn only had to make segmentation and normalization, but there were 2 situations regarding the anat:[list][*]I give conn raw anat scan and it does the whole preprocessing on it: in this case, there was no bias correction made, and the segmentation was pretty bad for the CSF and grey/white matter, so bad denoising; plus, there was an issue of « NON MATCHED DIMENSIONS » with the functional scans, I couldn't find what did this mean...[*]I give conn already preprocessed anat scan (already normalized and bias corrected), but then it does the segmentation from functional scan, therefore it was worst than previous situation...[/list]—> so I decided to put preprocessed functional scans in conn (but without ART outlier detection, normalization, nor smoothing), and completely preprocessed anat scans. I even put the different « ROI » resulting from the segmentation i.e. white matter, grey matter and CSF, already normalized. This is the way I do my preprocessing on conn.
And eventually, when I have to start the denoising, I have this kind of images, for some subjects it seems that the variance explained by CSF is mainly in the ventricles, but for some other (almost half I guess), this variance is mainly outside the ventricles, some times exclusively in the white or grey matter...
When I look at 1st level analyses: some of the subjects don't even have a correct default mode network activations...
So my point is:
- Is this an issue or totally normal? Should I improve the denoising or is it okay to take a look at the analyses with this denoising result?
- If an issue, how should I do to get over it?
I don't know if I'm clear, let me know if not...
Thanks a lot for your time and answer,
Best,
Fabien Hauw.
I have a problem with the denoising of my resting state images.
So i have [b]one subject[/b] that i want to compare to [b]36 other healthy subjects[/b].
I have [b]2 groups of healthy controls[/b], with [u]different RT[/u] and [u]different acquisition modalities[/u] (one had multi echo).
My issue is that when I preprocess and denoise these resting state in conn, the denoising step is not good, especially for the [b]CSF compound[/b] : for example here is 1 subject (see attached file), the denoising is really bad, and when i raise the threshold, there is explained variance by the CSF almost everywhere except in the ventricules... I specify that my ROI images for grey/white matter and CSF are good enough, and aligned with MNI space for the "bad" subjects".
I will now explain how I manage the preprocessing:
- As one group had a multi echo acquisition, the echo fusion is made by MEICA via AFNI, and it does the slice timing correction and realignment before the fusion; the other group and my subject of interest have simple echo acquisition and then had realignment and STC on SPM.
- I have tried to put the functional scans once they were realigned and STC in conn, so conn only had to make segmentation and normalization, but there were 2 situations regarding the anat:[list][*]I give conn raw anat scan and it does the whole preprocessing on it: in this case, there was no bias correction made, and the segmentation was pretty bad for the CSF and grey/white matter, so bad denoising; plus, there was an issue of « NON MATCHED DIMENSIONS » with the functional scans, I couldn't find what did this mean...[*]I give conn already preprocessed anat scan (already normalized and bias corrected), but then it does the segmentation from functional scan, therefore it was worst than previous situation...[/list]—> so I decided to put preprocessed functional scans in conn (but without ART outlier detection, normalization, nor smoothing), and completely preprocessed anat scans. I even put the different « ROI » resulting from the segmentation i.e. white matter, grey matter and CSF, already normalized. This is the way I do my preprocessing on conn.
And eventually, when I have to start the denoising, I have this kind of images, for some subjects it seems that the variance explained by CSF is mainly in the ventricles, but for some other (almost half I guess), this variance is mainly outside the ventricles, some times exclusively in the white or grey matter...
When I look at 1st level analyses: some of the subjects don't even have a correct default mode network activations...
So my point is:
- Is this an issue or totally normal? Should I improve the denoising or is it okay to take a look at the analyses with this denoising result?
- If an issue, how should I do to get over it?
I don't know if I'm clear, let me know if not...
Thanks a lot for your time and answer,
Best,
Fabien Hauw.