3D Cuboid Segmentation Aggregation

[goldenfire6]

Draft Status

Ready - team will start page creating immediately

Category

Segmentation / Classification / Landmarking

Presenter Location

In-person

Key Investigators

• Mike Jin (Brigham and Women's Hospital; Centaur Labs, USA)
• Tamas Ungi (Queen's University, Canada)
• Amene Asgari (Brigham and Women's Hospital, USA)
• Tina Kapur (Brigham and Women's Hospital, USA)

Project Description

Combining multiple segmentations from skilled annotators can improve accuracy when creating ground truth image segmentations.
The Centaur annotation platform has new support for collecting and displaying annotations on DICOM studies using OHIF viewer.
We are currently in the middle of a 1-month task of sourcing and creating segmentations for radiology findings from hundreds
of annotators on 30,000 brain and abdominal CTs.

Objective

We would like to extend an existing 2D clustering algorithm for combining multiple 2D segmentations to work
for 3D cuboid segmentations.

Approach and Plan

1. Create 3D representation objects for cuboid annotations submitted through OHIF viewer
2. Implement clustering algorithm using a 3D instead of 2D distance metric

Progress and Next Steps

1. Describe specific steps you have actually done.

Illustrations

https://www.dropbox.com/scl/fi/0clde5ufr468jccyijk0n/Screenshot-2024-06-24-at-12.35.17-PM.png?rlkey=hefwskhabqlhlkmd9v32c8ky8&dl=0

Background and References

No response

[github-actions]

Project Page Pull Request Creation

✅ COMPLETED: See #1218