Machine Learning Project - Questions about integrating Slicer into workflow

It sounds like an interesting project; hopefully others will chime in after the weekend, but to touch on one point here:

If you plan to have a Slicer-centric workflow, your best bet is to work in Slicer’s native “NRRD” format. There is very little advantage – and some disadvantage – to using Nifti with Slicer instead of NRRD unless you have to work with other software that only supports Nifti (but Elastix will work fine with NRRD, for example). Basically using Nifti will add another conversion step in various places, and there are occasional interpretation mismatches with other software.

Keep in mind that most Nifti-related software is focused on neuroimaging (of course there is some applicability cross-over for tools like registration).

As for DICOM: Slicer has robust support for interacting with DICOM at the beginning (import) and end (export) of studies, but I don’t think it is really recommended to try a DICOM-native workflow yet. Various tools using the CLI interface won’t work without without conversion to NRRD anyway (as far as I know – maybe this has changed or been automated recently).

1 Like