create_sphere uses MNI152_2mm template even when given mask

[jnecus]

The resulting mask shape does not match that of my provided mask (in this case t1_image)

Here is my code:

from nltools.mask import create_sphere
mask=create_sphere([123,201,102],radius=10,mask=t1_image)

Results:

t1_image.shape (180,256,256)
mask.shape (91,109,91)

[ljchang]

@jnecus: do you mind sharing your t1_image so we can try to fix this bug. Maybe a dropbox link, or on neurovault?

[jnecus]

Sure, here's a link
https://www.dropbox.com/sh/kgl7bglupmedetw/AACAcE2c4LW0qF6Y3JL-d4U6a?dl=0

[ljchang]

Thanks @jnecus. You definitely helped us identify a couple of previously undetected bugs. We mostly use our tools on normalized data.

I just pushed a fix for this bug and think it should work now #123. Though you'll have to install from github. One important thing to note is that the mask really does need to be a mask not just any nibabel instance. For your use case, you need to create a mask of your T1. Personally, I would create a mask using the skull stripped data. Here is example code using bet from fsl bet T1_TFE_1_0_Iso_SENSE.nii.gz T1_TFE_1_0_Iso_SENSE_stripped -m. Let us know if you have more problems with anything else.

[jnecus]

Thanks @ljchang , I downloaded the latest version from github and it works fine.

Rather than using the skull stripped image I used a mask image which was the same size as my t1_w, but filled exclusively with 1's. This is because I wanted to create a spherical roi which falls outside of the brain.

On that note, could you explain to be the need to include a mask in a function of which the role is ultimately to generate a mask. I cannot seems to grasp the use for this mask option?

[ljchang]

@jnecus: you need to pass a mask in so the function knows the space to build the rois. You're right though, that it actually doesn't need to be a mask. It could have been your original T1. It only uses the mask to get the dimensions of the image.