--skull-strip-fixed-seed should be implicit when --random-seed is set

[rolandhaidari]

What would you like to see added in this software?

Summary
When a user explicitly sets --random-seed, the expectation is that the pipeline will produce reproducible, deterministic outputs. However, even when parallelism is fully disabled via --omp-nthreads 1 and --nprocs 1, PETPrep still contains a separate source of randomness in the skull-stripping step that is only pinned by the independent flag --skull-strip-fixed-seed. Without knowing about this interaction, a user who sets --random-seed will still get non-deterministic outputs.

Current behavior
Running with --random-seed 42 --omp-nthreads 1 --nprocs 1 does not guarantee identical outputs across runs. The skull-stripping step introduces additional randomness that varies between executions regardless of the seed and regardless of parallelism settings.

Proposed solution

Automatically enable --skull-strip-fixed-seed whenever --random-seed is explicitly provided. This could be implemented as a conditional default in the argument parser, with a note in the documentation explaining the interaction.

Do you have any interest in helping implement the feature?

No

Additional information / screenshots

No response

[mnoergaard]

Thanks for opening this issue @rolandhaidari. I am curious to hear what e.g. @effigies or @oesteban have to say to this? As far as I can see from e.g. fMRIPrep, the same pattern exists, so this is relevant both across fMRIPrep and PETPrep.