in write_nifti an additional dimension is added to the output image

[michalhg]

In the write_nifti function that exports the 3D segmentation prediction to a nifty image file, and addition dimension is added (either for channel or batch) which makes it unreadable using ITK-SNAP or Slicer.
I propose the following solution

results_img = nib.Nifti1Image(data.squeeze().astype(output_dtype), to_affine_nd(3, target_affine)) # squeeze channel dim

for row 105 in row nifty_writer.py file.

[rijobro]

Hi @michalhg, have you looked into NiftiSaver, which has the save_batch method?

[rijobro]

Or if you also want to remove the channel dimension, how about Compose([SquezeDim, SaveImage])?

[michalhg]

I am using save_batch, but in the end it calls write_nifti without removing the channel dim.
regarding your solution, should I avoid save_batch, and just call Compose([SquezeDim, SaveImage]) ?
can you provide an example how to use it instead of save_batch?

[rijobro]

Hi, I looked into this a bit more today, and have suggested this PR, allowing the user to squeeze any trailing singleton dimensions, being the channel dimension in your case.

For the record, I could still load the 4D image (3D image with singleton channel dimension) in ITKSnap.

[wyli]

addressed by #1680