NA-MIC Project Weeks
Back to Projects List
MorphoDepot: Collaborative segmentation projects
Key Investigators
- Murat Maga (Seattle Children's, USA)
- Steve Pieper (Isomics, Inc., USA)
- Andrey Fedorov (Brigham and Women's Hospital, USA)
Project Description
We are developing tools for segmentation of biological specimens (e.g. 3D microCT of fish or snakes). The idea is that a Lab Director will define a project, such as what scan to segment, what anatomical structures to segment, the terminologies to use, etc. Students or lab members would be assigned to segment subsets of the data. We want to leverage existing data management tools, such as github for organizing issues and contributions, and jetstream2 for hosting data and computation.
There is a presentation about the ideas here: https://morphocloud.github.io/MorphoDepotDocs/
And there is an existing extension here: https://github.com/MorphoCloud/SlicerMorphoDepot
We are interested in facilitating collaborative segmentation, including dividing a whole project into tasks, managing allocation to tasks to segmenters, managing/merging contributions, etc.
If you are interested in similar topics, please join our project!
Objective
- Networking: we would like to know how this fits with anyone else’s projects and possibly collaborate
- Talk with developers who are improving the terminologies and color modules about how to better structure our segmentations
- Harden the infrastructure for Mac/Windows (install gh cli for users).
- Work on MorphoDepotAccession module to make it easier to create segmentation task repositories (design ideas are here: https://github.com/MorphoCloud/SlicerMorphoDepot/issues/10)
- Explore how to improve the GH tasks efficiency as querying through tags to find repos, issues and PRs can be slow and may not scale.
- Bonus: if time brainstorm tools for comparing segmentations / reviewing and merging segmentations from team members more effectively.
- Extra Bonus: Come up with plans on how to use collaboratively segmented datasets to train AI models (and iteratively refine them)
Approach and Plan
- Define/create a JSON schema for the mandatory metadata for each MorphoDepot archive. These should include specimen metadata, and some key imaging metadata.
- Explore how to create a staging area for data donors to upload their scenes (volume + segmentation, if exists) to be reviewed by MorphoDepot team (quality control, metadata check, etc).
Progress and Next Steps
- Extension is fully functional and has been used in classroom with multiple users for basic segmentation tasks. (e.g., go to https://github.com/muratmaga/pinecone/issues and open an issue for yourself to test).
- Implement the MorphoDepotAccession
Illustrations
MorphoDepot module lists pending issues assigned to this user and allows you to load/segment/commit them and then request review.
MorphoDepotReview module lists pending pull requests and allows PI to accept edits or request changes.
Background and References
Previous Project Week work: