Multi-channel vs multi-label rootlets segmentation

Just writing down here one idea that we came across during the 2024-07-25 "lumbar classification challenge meeting" meeting:

- explore multi-channel (4D nii image = one class per image) vs multi-label (3D nii image = one image with multiple classes, each class has a different int value) approach for rootlets segmentation

Currently, we use the multi-label (3D) approach.

Start by reviewing the current multi-label 3D NIfTI representation and segmentation workflow. Compare it with the proposed multi-channel 4D representation for rootlets segmentation, including how each class is encoded. Done means documenting the trade-offs and recommending whether to keep the current approach or explore the alternative.

Written by the indexing model from the issue text.