OHIF-SAM2

Paper

Prerequisite

• Install Docker, NVIDIA Container Toolkit

• tested version: Docker: 27.3.1, NVIDIA Container Toolkit: 1.16.2, CUDA Version: 12.6

Getting Started

• Run bash start.sh
• Go to http://localhost:1026
• Upload all DICOM files in sample-data

Demo Video

• Prompts (Positive, Negative Points and Bounding Boxes) are supported.

• Two SAM2 buttons are available: SAM2_one (faster) for the slices where prompts are given, SAM2 for all slices via propagation.

• Multiple labels are supported - click nextObj button once it is done with the current label

Updates

• OHIF upgrade: 3.10-beta.48, cornerstone 2.0 -> Support partially 3D segmentation rendering

• Fixed flipping segmentation issue

Next steps

• Support SAM2 for any label anytime: Currently, if a user pass to the next Obj, the user cannot reuse SAM2 for the old label, only manually edit

• Integrate Text prompt

(potential) FAQ

• Q: Load library (libnvidia-ml.so) failed from NVIDIA Container Toolkit

• A: Run sudo apt-get install --reinstall docker-ce Reference

• Q: Failed to initialize NVML: Unknown Error Or `No CUDA available``

• A: Edit no-cgroups = falsein /etc/nvidia-container-runtime/config.toml Reference

How to Cite

@article{jcho2024ohifsam2,
  title={OHIF-SAM2: Accelerating Radiology Workflows with Segment Anything Model 2},
  author={Cho, Jaeyoung and Liu, Jingyu and Schlamp, Kai and Rastogi, Aditya and Vollmuth, Philipp},
  journal={TechRxiv preprint techrxiv.173387978.85520380},
  url={https://www.techrxiv.org/users/868002/articles/1248561-ohif-sam2-accelerating-radiology-workflows-with-segment-anything-model-2},
  year={2024}
}