Updates facemask.nii.gz with better ear and mouth mask

[miykael]

I've updated the facemask.nii.gz so that it also cuts out the ears of a subject. As discussed on neurostars, cutting out the ears can be a necessary requirement for local ethic committees to consider an MRI image as "fully anonymized".

I've also cut out some more parts of the mouth area to make sure that the dental information are excluded from the defaced image.

The adaptation of the previous facemask was done manually and verified on a dataset of 12 subjects.

[chrisgorgo]

Thanks - could you post some screenshots of the mask pre and post adjustment here? This will help document the evolution of the package.

[miykael]

Of course. The following figure shows regions that were only in the old facemask in yellow, regions that are now new in the facemask in blue and regions that haven't changed in green. The template mean_reg2mean.nii.gz is depicted in the background.

As can be seen, the new mask cuts out the ears, removes some more regions of the jaw and adds again the region close to the frontal pole. I've added this last part again, because it lead to frontal pole removal in some of my subjects.

[chrisgorgo]

@robertoostenveld @jasmainak would such mask be still ok for MEG skull reconstruction and electrode localization?

[chrisgorgo]

Thanks for providing figures @miykael!

[jasmainak]

I’m not sure but I suspect coregistration would be difficult without the pre auricular points.

Maybe the accuracy of the surfaces will also affect source localisation accuracy.

I think @agramfort can give better opinion than me here

On Sat 28 Apr 2018 at 17:46, Chris Filo Gorgolewski < notifications@github.com> wrote:

Thanks for providing figures @miykael https://github.com/miykael!

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/poldracklab/pydeface/pull/22#issuecomment-385185460, or mute the thread https://github.com/notifications/unsubscribe-auth/APHiomLAbyowO_fTRQcPPNK_r9rQ_98xks5ttI7vgaJpZM4TrDUI .

-- Sent from my iPhone

[agramfort]

no indeed. For MEG/EEG coreg we need the preauricular points and the tragus and at least the upper half of the nose.

[robertoostenveld]

Agreed with Alex. After application of this mask re-doing the coreg would indeed be difficult (or very imprecise), if not impossible (depends on the coreg method and whether a headshape is available).

IMHO coregistration should be done prior to making the de-identified anatomical in BIDS format and coreg information should be shared in the AnatomicalLandmarkCoordinates field. As such I don't see the problem of this (or any other) mask.

Considering that after conversion of dicoms to nifti's it would also be difficult to determine MR parameters; I would say there is nothing new here: some parameters have to be determined and shared prior to or during conversion.

[chrisgorgo]

I understand the localization could be done before defacing, but it would be more convenient if we could come up with a mask that removes some ear/nose details, but keeps enough to make the localization possible. Or do you think that this is not possible?

[agramfort]

don't know what is ethically acceptable but I agree it would be great to still allow this.

[robertoostenveld]

it might work if you would remove the eyes and cheeks, but keep the nose (including the tip and nasal wings) in place. Also the front of the ear (including the details around the ear canal, but not the ear shell) should remain intact. I don’t know whether this could be done robustly enough to automate, or whether it requires visual inspection on every subject.

On 30 Apr 2018, at 18:34, Alexandre Gramfort notifications@github.com wrote:

don't know what is ethically acceptable but I agree it would be great to still allow this. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/poldracklab/pydeface/pull/22#issuecomment-385454002, or mute the thread https://github.com/notifications/unsubscribe-auth/AA2346oUfIxBcKiONkYnakvJgz3OQiUbks5ttz0GgaJpZM4TrDUI.

[Shotgunosine]

@chrisfilo What about including this mask as an option until a mask that works better with MEG can be developed?

[chrisgorgo]

Sounds good to me.

On Mon, Feb 4, 2019, 8:09 AM Dylan Nielson <notifications@github.com wrote:

@chrisfilo https://github.com/chrisfilo What about including this mask as an option until a mask that works better with MEG can be developed?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/poldracklab/pydeface/pull/22#issuecomment-460306332, or mute the thread https://github.com/notifications/unsubscribe-auth/AAOkp0MZltFUV5v5tA9UPUQeBGlxg83Bks5vKFssgaJpZM4TrDUI .

[miykael]

@Shotgunosine - Happy to adapt the changes. Should I just include an option like --remove_ears, which indicates that this template should be used?

[vsoch]

oh my. That sounds quite painful.

[Shotgunosine]

@miykael That sounds good. Just a heads up, I'm hoping to get #19 merged shortly, which might require some restructuring in this PR to be compatible.

[miykael]

@vsoch - I agree. But my current PR would overwrite the current facemask, which supposedly is not the goal :-)

@Shotgunosine - Good, then I will wait for https://github.com/poldracklab/pydeface/pull/19 to be merged.

[Shotgunosine]

@miykael #19 is merged.

[muschellij2]

It seems as though this facemask.nii.gz allows you to remove ears but leaves the eyes (while removing more mouth), which seems like it'd be good to create a series of structures to combine together as I'd like the original facemask sheared face, but also remove the ears, but the proposed mask didn't remove enough of the face for my type of data.