Try using image or SC mask

[naga-karthik]

The issues with convergence of mutual information with INR T1w and INR T2w could be improved by using an image mask or the spinal cord mask. Here's a snippet I used for the image mask:

import nibabel as nib
from skimage import filters

input_img_nii = nib.load(input_path)
input_img = input_img_nii.get_fdata()

# "Minimum" method from scipy.filters gives a decent SC image mask 
# cuts out part of the image where the contrast is too low but rest is fine
thr = filters.threshold_minimum(input_img, nbins=256)
mask = input_img > thr

# create nii image and save
mask_nii = nib.Nifti1Image(mask, input_img_nii.affine, input_img_nii.header)
nib.save(mask_nii, output_path)