Explore the Two-Phase Approach for the instance segmentation

We are now thinking about training a single model segmenting both cervical and lumbar rootlets. As the training of such a model might be problematic (due to the number of classes: cervical + lumbar rootlets), we could explore the Two-Phase Approach as done in [SPINEPS](https://github.com/Hendrik-code/spineps). First, we would do binary segmentation of the rootlets, followed by instance segmentation (i.e., labeling of C2, C3, ..., L1, L2, ...). Alternatively, the first step could include two classes (aka groups: cervical and lumbar rootlets).

No files, tests, or entry points are named in the issue. Start by locating the current training and segmentation pipeline in this repository, then review the linked SPINEPS two-phase approach; done would require an agreed implementation direction for rootlet grouping and instance labeling.

Written by the indexing model from the issue text.