import nilearn
from nilearn import datasets
from nilearn.input_data import NiftiMasker
from nilearn.datasets import load_mni152_template
from nilearn.image import resample_to_img
from nilearn.image import load_img
from nilearn.image import image

### Load MNI template.
template = load_mni152_template()
template_img = load_img(template)

### Fetch first subject's fMRI and ROI.
haxby_dataset = datasets.fetch_haxby(subjects=(1,))
fmri_filename = haxby_dataset.func[0]
mask_filename = haxby_dataset.mask_vt[0]

### Resample fMRI to MNI template.
fmri_resampled = resample_to_img(fmri_filename, template)

### Apply mask.
masker = NiftiMasker(mask_img=mask_filename, standardize=True)
fmri_masked = masker.fit_transform(fmri_resampled)
num_samples, num_voxels = np.shape(fmri_masked)

### Fetch binary mask data.
binary_mask = masker.mask_img_.get_data()
indices = np.where(binary_mask == 1)
# convert into x,y,z triplets
mask_coords = zip(indices[0], indices[1], indices[2])
mask_coords_mtx = np.reshape(mask_coords, (num_voxels, 3))
mask_coords_mtx_MNI = np.zeros_like(mask_coords_mtx)

# convert from pixel indices to MNI coordinates
for i, coords in enumerate(mask_coords_mtx):
    ix, iy, iz = coords
    x, y, z = nilearn.image.coord_transform(ix, iy, iz, template.affine)
    
    mask_coords_mtx_MNI[i] = [x, y, z]
    print "(x, y, z) = ({0}, {1}, {2})".format(x, y, z)