fMRIVolume Related Pull Request — Use of a specific BOLD mask #141
Conversation
sivcek
changed the title
| Compliance="HCPStyleData" | ||
| ComplianceMsg="" | ||
|
|
||
| if [ ! "${Mask}" = 'T1' ]; then |
There was a problem hiding this comment.
I think the purpose of the mask is more to reduce file sizes by setting non-brain to exact 0s, so I'm not sure this needs to be restricted to legacy data mode.
There was a problem hiding this comment.
Oh right, for intensity normalization...depending on how we do that, it may be important to use a tight mask.
There was a problem hiding this comment.
I'm not sure if there is a way to compute the intensity normalization using a tighter mask and then apply it on a wider one.
There was a problem hiding this comment.
I guess I don't really like this PR. If you need to, just dilate the existing brain mask (which will be maximally accurate to the given brain). I don't see any reason to invoke the MNI space mask or recompute it from the BOLD data (potentially masking out dropout regions). It probably would be possible to replicate the intensity normalization using wb_commands and gain control of the masking. I agree that intensity normalization would be influenced by dilation of the mask. Also, I am not sure I understand how in a rigid registration the BOLD image would expand, so I am unclear on the need for the PR in any case. If there is a strong feeling to keep it, then the options would be T1w, dilated T1w, or NONE.
|
@glasserm "expand" is not the right term. With lack of DC, expanded mask allows to see how badly aligned the image is. |
|
Let's implement as I suggested above |
|
The purpose as stated in the PR was to "help in checking the quality of bold registration". If that's the goal, I think there are better, more explicit ways we can check the registration than to expand the mask (which seems like a crude solution) -- e.g., we can make sure that we make available a non-masked version of the registered Scout, so that you can easily visualize what is occurring "beyond the borders defined by the T1w mask". Also, @sivcek, this is going to interact with PR #138 which I'm hoping to merge into the master tomorrow. |
|
Well I also agree with this. |
|
@glasserm @mharms The value goes beyond just QC checking. If somebody want's to stay for some reason in volume-land, they might want to make sure that the BOLD was not cut and would still proceed with suboptimal registration, that would be prohibitive for surface analysis, but might be usable for volume-based analyses. If it's ok with you. I'll wait for @mharms to pull in his PR and I'll see how to re-introduce the functionality to satisfy previous comments. |
|
The PR was merged last night, so check out the current master branch, and proceed from there. |
|
Ok. |
|
If we are keeping it, we have the default tight brain mask, a dilated version of this mask, and no mask as reasonable options. |
|
I have now added the following options:
I've also implemented checks for valid options. |
|
FINAL is the mask created by code that @mharms added. |
|
So final is a combination of the T1w mask and the FOV mask, right? |
|
Okay, lets call it T1_fMRI_FOV |
There was a problem hiding this comment.
So, IMO, the "FOV mask" (${InputfMRI}_mask) should always be applied, regardless of whether you want a more generous T1-related mask -- under what situation is it valid to include the voxels that are outside of the "FOV mask" (i.e., that don't have proper spatial coverage across all time points)?
If the FOV mask is always applied, then the T1 option becomes irrelevant.
|
|
||
| if [ "${UseMask}" = "DILATEDT1" ] ; then | ||
| # -- create a dilated version of T1 mask | ||
| ${FSLDIR}/bin/fslmaths ${BrainMask} -dilF ${BrainMask}_dil |
There was a problem hiding this comment.
Is a single round of dilation enough to accomplish your purposes with the DILATEDT1 mask? Seems like you'd want more than that...
There was a problem hiding this comment.
The options now will be:
T1_fMRI_FOV: combined T1w brain mask and FOV masks (the default),T1_DILATED_fMRI_FOV: a once dilated T1w brain based mask combined with fMRI FOVT1_DILATED2x_fMRI_FOV: a twice dilated T1w brain based mask combined with fMRI FOV,fMRI_FOV: a fMRI FOV mask
There was a problem hiding this comment.
I have also changed the flow of the code so that the FinalMask is the mask specified. And that the statistics are reported both on the FOV of the T1w brain as well as the actual mask used, with the mask used also listed in the ${FinalMask}.stats.txt
…DILATED_fMRI_FOV, T1_DILATED2_fMRI_FOV, fMRI_FOV, and expanded statistics report
|
Joined with other fMRIVolume related PRs into PR #156 |
Adds the (optional) ability to specify the mask to use in the final Intensity Normalisation step. The options are:
T1 - the default option
BOLD - a mask created by bet on the mean bold image across the timeseries
DILATED - use the MNI152_T1_2mm_brain_mask_dil.nii.gz mask
NONE - do not use a mask
In our experience using a wider mask than T1 (DILATED or NONE) helped in checking the quality of bold registration, as it was possible to see if the BOLD image was expanded beyond the borders defined by the T1w mask.