How to check & fix automatically generated segmentations in large data sets

[katjaq]

We can start by browsing several brains and take note of segmentation errors (i.e. where either tissue has been classified brain which is actually not part of the brain; or vice versa chunks of tissue might be missing from the segmentation).
A good starting example might be the ABIDE I dataset (Autism Brain Initiative Data Exchange) which you can find on BrainBox here.

We could then choose one brain as an example, and

• build a pool of screenshots of errors which we spot
• find patterns in which regions the algorithms seem to struggle most
• and build our tutorial based on those, for instance
  • 'Deleting the meninges from the cortex classification',
  • adding screenshots and
  • describing how we fix the errors
    • which slice view worked best for you (coronal/ sagittal/ axial),
    • which pencil size/eraser,
    • starting from one end or rather the middle of the brain and everything which you will find useful to add
  • and maybe make a video/screencast to show the collaborative & real-time aspect of process

Please join us! Everybody is more than welcome! 🌞