README.md

Back to Projects List

Spine Segmentation

Key Investigators

• Ron Alkalay (Beth Isreal Deaconess, Boston)
• Steve Pieper (Isomics)
• Andres Diaz-Pinto (KCL)
• Juan Ruiz (Ebatinca, ULPGC)
• YOU

Project Description

Investigate and implement methods to segment the human spine from CT scans. See last Project Week's page for background.

Objective

1. Ideal segmentation will independently segment and label the vertebral bodies.
2. We want the system to integrate with Slicer's segmentation infrastructure.
3. We think a deep learning approach using MONAILabel will be useful for this.

Approach and Plan

1. Learn as much as possible about MONAILabel
2. Investigate VerSe and if possible port it to Slicer/MONAI
3. Figure out if/how we can use spine CTs from IDC for training.

Progress and Next Steps

1. Held many productive discussions and worked on training with the VerSe public data
2. Exchanged notes with the other MONAI Label projects
3. Installing MONAI Label at BIDMC machines to train on cadeveric and patient spine scans
4. Plan to make single-vertebra models for faster training of high resolution models (tractable on smaller GPU memory footprint)

Illustrations

Current effort

Initial effort

Background and References

• https://github.com/anjany/verse
• https://projectweek.na-mic.org/PW35_2021_Virtual/Projects/SpineSegmentation/