Mask image neuroimaging

, , ,
1

Hi !
please help !
i have a 262 nifti image [’ image1.nii’ , ‘image2.nii’…] (each of shape(182,218,182) and a mask mask.nii of shape (182,218,182) .
and i want to apply mask on image but i have error

Traceback (most recent call last):

File "<ipython-input-61-88ba8b5de2d2>", line 5, in <module>
masked_data = apply_mask(img, masker)

 File "C:\Users\moham\Anaconda3\lib\site-packages\nilearn\masking.py", line 707, in 
 apply_mask
 mask, mask_affine = _load_mask_img(mask_img)

 File "C:\Users\moham\Anaconda3\lib\site-packages\nilearn\masking.py", line 63, in 
 _load_mask_img
 % values)

 ValueError: Given mask is not made of 2 values: [ 0  1  2  3  4  5  6  7  8  9 10 11 12 
 13 14 15 16 17 18 19 20 21 22 23
 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47
48]. Cannot interpret as true or false

Please help me

2

Looks like your mask NIFTI is a collection of 48 regions?
The nilearn mask function only take mask images consist of 1 and 0.
If you want to use the whole 48 regions as a mask, you will need to make all values above 0 one (binarize the image).
If you want to mask each individual region, extract each region and binarize it before passing to the mask function.

3

How did you extract this mask image? mask.nii

4 
masker = nb.load('C:/Users/JHU-ICBM-labels-1mm.nii.gz')
mask = image.math_img("img>0", img=masker)
nb.save(mask, "mask.nii.gz")

And now i can use it :

masked =  'C:/Users/mask.nii.gz'
x = ['ze.nii.gz' , 'ke.nii.gz' , ....)
rena = Parcellations(method='rena', n_parcels=5000 ,mask=masked)
rena_fit =rena.fit_transform(x)

is it correct like this ?

5

Looks correct to me.

6

ok thank you very much !
@KamalakerD i have another question :
x =['aa.nii ',‘bb.nii’ ,‘gg.nii’ …] # we have 262 subjects

for i in range (100):

    x=resample(x)    # to have every time a diiferent distribution
    rena = Parcellations(method='rena', n_parcels=1000 ,mask=masked)
    rena_fit =rena.fit_transform(x)

So, when i run with 100 randomized parcellations (each parcellation with 1000 parcels) and 262 subjects , i have a problem on memory , when i reduce the number of subjects to 50 , it takes 5 hour to do this !
but i will do this with the 262 subjects :confused:
how can i do this please to perform the computational time .

7

ERROR :

File “C:\Users\Anaconda3\lib\site-packages\nilearn\regions\rena_clustering.py”, line 128, in _make_edges_and_weights
weights_unmasked = _compute_weights(X, mask_img)

File “C:\Users\Anaconda3\lib\site-packages\nilearn\regions\rena_clustering.py”, line 53, in _compute_weights
weights_deep = np.sum(np.diff(data, axis=2) ** 2, axis=-1).ravel()

File “C:\Users\Anaconda3\lib\site-packages\numpy\lib\function_base.py”, line 1273, in diff
a = op(a[slice1], a[slice2])

MemoryError

@KamalakerD

8

Out of 262, you choose 50 random subjects to fit 1000 parcels? This is repeated for 100 iterations?
That’s what you are trying to do?

What does this resample do?

9

x =['aa.nii ',‘bb.nii’ ,‘gg.nii’ …]
i do the resampling method (bootstrap) to change the order of subject to have at each parcellation a different distribution of parcels .

The goal is to do 100 randomized parcellation and each parcellation have 1000 parcels .
that’s why for each iteration i (each parcellation) :
1/ we resample x (to change order)
2/ then, we train parcellation with ReNA to do the clustering .

the problem is when i take a dataset with only 60 subjects , the method works very good (~5 hours) but when i take a dataset with 262 subjects , i have a problem of memory

10

when i take a dataset with 262 subjects , i have a problem of memory

Can you try by changing the data dtype of the images?

from nilearn._utils import check_niimg_4d
x = check_niimg_4d(x, dtype='int32')
11

yes , i try it and i have the same problem
my computer have RAM 8go , maybe this is the problem ?
because when i try it with a dataset of 60 subjects , it work but when i try it with 262 , i have an error of memory

12

yes , i try it and i have the same problem
my computer have RAM 8go , maybe this is the problem ?

When you fit ReNA on 262 images, you have a Memory Error? Irrespective of whether in the loop?
Thanks!

13

Yes ! and when i take a dataset with just 60 subjects , it works very goog without any error !
thank you

14

did you solve the memory error issue?

15

No :sleepy: @KamalakerD

16

Do you think you can upload data some where and share with me a link pointing to that data?
I will try to get my hand on it to see if I can propose a solution.
Thanks!

17

Otherwise, there is this code which exists in this nilearn sandbox repo: https://github.com/nilearn/nilearn_sandbox/blob/master/nilearn_sandbox/mass_univariate/rpbi.py

May be the repo has some memory efficient code.

If you use that code, please cite this article: B . Da Mota, V. Fritsch, G. Varoquaux, T. Banaschewski, G. J. Barker,A. L. Bokde, U. Bromberg, P. Conrod, J. Gallinat, H. Garavanet al.,“Randomized parcellation based inference,”NeuroImage, vol. 89, pp.203–215, 2014

18

okey thank you very much ! i will try it and if it work i will cite the article .

19

hello @KamalakerD
in this function :

 _build_parcellations(all_subjects_data, mask, n_parcellations=100,n_parcels=1000,

n_bootstrap_samples=None,random_state=None,memory=Memory(cachedir=None),n_jobs=1, verbose=False):

i don’t understand this part :

  draw = rng.randint(n_samples, size=n_samples * n_parcellations)
  draw = draw.reshape((n_parcellations, -1))

and in the function :

  _ward_fit_transform(all_subjects_data, fit_samples_indices,
                    connectivity, n_parcels, offset_labels):

why he use “data_fit” in the fit part and “all_subjects_data” in the transform part ?

Please explain to me please

20

That’s the randomization step. It generates random indices to build parcellations.

Ideally, parcellations should be built on a subsample of full data to avoid possible overfitting results.

I suggest that you read the paper and code documentation thoroughly.

Best,
Kamalakar