QSIPrep segmentation outputs

Hi all,

QSIPrep generates two segmentation maps as final outputs: *dseg.nii.gz and *aseg_dseg.nii.gz. I have two questions:

  1. Are these segmentation outputs used by QSIRecon?
  2. Are they used for any other downstream processes?

The reason I’m asking is that in *dseg.nii.gz, the white matter regions appear very restricted—some tracts are clearly missing. Could this be due to SynthSeg applying stricter criteria for white matter?

Additionally, I noticed several mislabels in *aseg_dseg.nii.gz, particularly around the brain’s outer layer. Is this expected? The major cortical areas seem correct, but the outer layer looks somewhat fragmented (looks like confetti!).

Are these results expected? My team is deciding whether to keep these segmentation maps or use our own existing methods. Any insights would be appreciated!

The aseg file is generated by SynthSeg and is just passed along to the output workflows. In QSIPrep, it doesn’t get used for anything except to generate the tissue-type dseg. The derived tissue-type dseg file is only used in QSIRecon for scalar plots, so no outputs actually depend on these maps. I do recommend keeping the dseg files just so that the QSIRecon scalar plots don’t error out.

1 Like

@tsalo Got it, thanks for the quick reply!