add merge ROIs feature

[marcobarilari]

Open questions:

• load_nii() is not happy with masks that, I guess, are not resliced in MNI space cause of distortions in the image. Solution? Seems that load_untouch.nii() ignores these distortions but not 100% sure about what it does.

• how naming the resulting *mask.nii from 2 or more masks? @Remi-Gau

• sum more than 1 matrix in an elegant way (no for loop)

[Remi-Gau]

Open questions:

* [ ]  `load_nii()` is not happy with masks that, I guess, are not resliced in MNI space cause of distortions in the image. Solution? Seems that `load_untouch.nii()` ignores these distortions but not 100% sure about what it does.

I would try spm_read_vols(spm_vol()) instead.

* [ ]  how naming the resulting `*mask.nii` from 2 or more masks? @Remi-Gau

* [ ]  sum more than 1 matrix in an elegant way (no `for loop`)

My 2 cents: load all images as a "4D" image and use any along the fourth dimension

images = {filenameA; filenameB};

mask = spm_read_vols(spm_vol(images));

mask = any(mask, 4);

This will give a logical true value to any voxel along the 4th dimension that has some non-zero values in it.

So assuming that you gave binary masks as inputs this should be fine.

Further assumptions we have here are the images have the same dimensions.

I think we can make this more flexible by making some sort of recusive call to createROI('mask') but let's worry about that later.

[Remi-Gau]

We are still developping this but I suspect we'll want to have the same way to use specification for all "cases":

so that it will be something like :

specification(1).mask = roi_filenameA
specification(2).mask = roi_filenameB

not necessary to have now but just to let you know that the "API" and the type and name of arguments for this function are still a moving target. :-)

[marcobarilari]

not sure if this replys to it but in a batch file you would have eg

roiImages = {fullfile(opt.ROI.roiDir, 'label-sphere5xMinus40yMinus50zMinus22_mask.nii'); ...
             fullfile(opt.ROI.roiDir, 'label-sphere5x36yMinus46zMinus17_mask.nii')};

[Remi-Gau]

More like this

specification(1).mask = fullfile(opt.ROI.roiDir, 'label-sphere5xMinus40yMinus50zMinus22_mask.nii');
specification(2).mask = fullfile(opt.ROI.roiDir, 'label-sphere5x36yMinus46zMinus17_mask.nii');

[Remi-Gau]

I am pretty sure that in this case you can just convert mask.img to a logical array:

mask.img = logical(mask.img);

[marcobarilari]

I think that using

      % merge the logical masks into one
      mask = any(mask, 4);

will save us from having values >1

although, maybe for the future, we would like to have different values for each single single ROI merged

[Remi-Gau]

If you mean having a single image that combines different masks has a different "label" for each ROI: I agree but I would have that as an entirely different function.
And I think that MarsBar must have some functions to do that.

[marcobarilari]

Thanks Remi :)

My 2 cents: load all images as a "4D" image and use any along the fourth dimension

images = {filenameA; filenameB};

mask = spm_read_vols(spm_vol(images));

mask = any(mask, 4);

I am not sure that spm_read_vols(spm_vol(images)) can read multiple files at once.

This will give a logical true value to any voxel along the 4th dimension that has some non-zero values in it.

So assuming that you gave binary masks as inputs this should be fine.

ok great!

Further assumptions we have here are the images have the same dimensions.

Yes indeed, would you add a check for that?

I think we can make this more flexible by making some sort of recusive call to createROI('mask') but let's worry about that later.

[marcobarilari]

last but not least, what about the name? Tricky one if we don't want to end up with a very long one eg:

mask1 = 'label-sphere5xMinus40yMinus50zMinus22_mask.nii'; 

mask2 = 'label-sphere5x36yMinus46zMinus17_mask.nii';

maskMerged = 'label-sphere5xMinus40yMinus50zMinus22PLUSsphere5x36yMinus46zMinus17_mask.nii'

[marcobarilari]

also (this one can become an issue by itself), do you think it would be possible to have a more human-readable label like

label-sphere5leftMT_mask.nii

instead of

label-sphere5xMinus40yMinus50zMinus22_mask.nii,

this way we can have a nicer name for the merged masks like

label-sphere5leftMTsphere5rightMT_mask.nii

[Remi-Gau]

last but not least, what about the name? Tricky one if we don't want to end up with a very long one eg:

mask1 = 'label-sphere5xMinus40yMinus50zMinus22_mask.nii'; 

mask2 = 'label-sphere5x36yMinus46zMinus17_mask.nii';

maskMerged = 'label-sphere5xMinus40yMinus50zMinus22PLUSsphere5x36yMinus46zMinus17_mask.nii'

I mean I think that last one rolls off the tongue quite easily...

[Remi-Gau]

I am not sure that spm_read_vols(spm_vol(images)) can read multiple files at once.

I think that if you sue spm_select to create your list of files it will work.

[Remi-Gau]

also (this one can become an issue by itself), do you think it would be possible to have a more human-readable label like

label-sphere5leftMT_mask.nii

instead of

label-sphere5xMinus40yMinus50zMinus22_mask.nii,

this way we can have a nicer name for the merged masks like

label-sphere5leftMTsphere5rightMT_mask.nii

I would use the renameFile function for that:

roiImage = `label-sphere5xMinus40yMinus50zMinus22_mask.nii`;

roiImage = renameFile(roiImage, ...
                          struct('label', 'sphere 5 left MT', ...
                                 'hs', 'lh'));

See the other_demo.m

[CerenB]

I am not sure that spm_read_vols(spm_vol(images)) can read multiple files at once.

I think that if you sue spm_select to create your list of files it will work.

Yes, I "grew up" using load_nii and load_untouch_nii and I investigated it a bit these differences recently. I used load_nii
with MNI space images and load_untouch_nii for native space images/ROIs . And we started using these two because this is what cosmo-mvpa uses. If spm is better (i have only used it recently...) then yes, let's switch to spm for dealing with nii files!

I'm adding @mohmdrezk here in case he has a preference for nifti toolbox some reason that I do not know!