Interpreting t1_dice_distance in QSIPrep

Software Support
1

Summary of what happened:

Hi all,

I’m reviewing the QSIPrep quality control outputs and have a question about the t1_dice_distance metric. From the documentation, it looks like this measures the overlap between the subject’s T1w-derived brain mask and the diffusion-derived brain mask, but I’m unsure how to interpret the actual values in practice:

  • What would be considered a “good” or “acceptable” t1_dice_distance?
  • What do the values correspond to (arbitrary value, pixels, mm…)?
  • Are there typical thresholds or ranges one should use to flag problematic scans?
  • How does this compare in usefulness to motion-related metrics like fd_mean or fd_max? Would it be sufficient to just rely on FD-based quality control, or is t1_dice_distance capturing something qualitatively different that I should pay attention to? After a quick analysis, I figured that it has a mild positive correlation with fd_mean and fd_max.

Any insight into how others use or interpret this metric in their quality control would be really helpful!

Thanks!

Command used (and if a helper script was used, a link to the helper script or the command generated):

Does not apply.

Version:

1.1.0

Environment (Docker, Singularity / Apptainer, custom installation):

Apptainer

Data formatted according to a validatable standard? Please provide the output of the validator:

bids_validator

Relevant log outputs (up to 20 lines):

Does not apply.

Screenshots / relevant information:

Does not apply.

2

Hi @catalinasaini and welcome to neurostars!

There is no standard threshold. I would say I would flag something over 0.05. Not necessarily exclude, see below.

It’s a proportion, so unitless.

I wouldn’t include dice distance as a covariate in any model, like motion typically is (although see this paper for why motion may not matter much when using simpler tensor models: Head Motion in Diffusion Magnetic Resonance Imaging: Quantification, Mitigation, and Structural Associations in Large, Cross-Sectional Datasets Across the Lifespan - PubMed. QSIPrep has other qc measures that may be better than motion in this context). However, is a brain mask is missing important parts of cortex (I guess that’s will depend on the analysis), then you would probably want to exclude that subject. The dice distance will flag people who likely have a deficient brain mask.

Makes sense, a high motion scan will likely have problems with spatial registration, which would lead to higher dice distance.

Best,
Steven

3

Thank you so much for the answers! I will keep this information in mind for quality control.