How to handle lesion masks, including multi-session data?

[danlurie]

This question came up in the context of how smriprep should handle multi-session data that has associated lesion masks (see https://github.com/poldracklab/smriprep/issues/58), but I think it raises a BIDS question too.

Currently, lesion masks are treated by BIDS as a special case of ROI masks, which are discussed in BEP003. As they are created by identifying damaged tissue from an existing image or set of images (e.g. T1, T2, FLAIR), they are by definition a derivative. That said, fmriprep currently looks them in the BIDS raw data directory (under /anat). This is a hack, and needs to get fixed at some point, but we did things this way because the lesion masks are used as inputs during structural preprocessing.

It would be great to get input from the community on the following:

• Since lesion masks are both a derivative and an input to preprocessing, where (i.e. which dataset, raw or derivatives) should they live?
• For some multi-session data sets (let's call this Case A), a single mask will be drawn from one of the sessions (usually the first), and then applied to all subsequent sessions. For other multi-session data sets (let's call this Case B), there will be a different mask drawn for each session. How should we handle Case A? Should we require a copy of the mask to be present in each session sub-directory, or would it be better to have it live in a higher level directory?

[leiliew]

Hi - Just wanted to ping this and see if with the new BIDS specifications, there was greater clarity on how to proceed. From what we can tell, the revised BIDS specification wants us to make a /manual_masks/sub/anat/sub-*-desc-lesion_mask.nii.gz file. But, it is unclear to me why this requires the initial /manual_masks/ directory at the top level, rather than just being in our regular study/sub/anat/ format. Any help understanding whether or not we need to use /manual_masks/ as the top level would be great. Thanks!

[yarikoptic]

The question/ambiguity where to store lesions common (e.g. in MNI space) to all subjects is remaining. Anyone reached peace of mind since original discussion on "how to do it" while "minimizing BIDS abuse"?