FastSAM-3DSlicer: A 3D-Slicer Extension for 3D Volumetric Segment Anything Model with Uncertainty Quantification

paper
FastSAM3D code
video
ARCADE lab, Johns Hopkins University

Table of contents

• Fast Segment-anything model in 3D medical image (FastSAM3D): A 3D Slicer extension to FastSAM3D
• Table of contents
  • Introduction
  • Before You Try
  • How to Use
    • Features
  • Citation

Introduction

What are SAM, SAMMed3D, FastSAM3D and FastSAM3D_slicer?

• SAM is the vision foundation model developed by Meta, Segment Anything.
• SAMMed3D is the 3D version of SAM based on medical image.
• FastSAM3D is the faster version of SAMMed3D.
• FastSAM3D_slicer is the 3D slicer extension based on FastSAM3D and SAMMed3D.
  Why FastSAM3D and FastSAM3D_slicer?
• FastSAM3D is about 10 times faster compare with SAMMed3D, with small accuracy lose.
• FastSAM3D_slicer provide an interface for user to do segmentation based on FastSAM3D for medical image intutively.
• And it's really fast!

Before You Try

Make sure you have more than 3GB storage to download model weights and install pytorch. Don't forget to use the provided resample.py file to do resample for medical image.

How to Use

step 1: Download the file and compress it.

step 2: open the 3D slicer and open the extension manager, download the pytorch extension, and restart the slicer. step 3: open the extension wizard in the 3D slicer. step 4: click the select extension and choose the compressed file in step 1. step 5: the extension will now be available here

Features

• 3 View Inference
• Data type
  • NIFTI file
  • volume
• models
  • FastSAM3D
  • SAMMed3D
• interactions
  • include and exclude points

Citation

If you use FastSAM3D_slicer in your research, please consider use the following BibTeX entry.

@misc{shen2024fastsam3d,
      title={FastSAM3D: An Efficient Segment Anything Model for 3D Volumetric Medical Images}, 
      author={Yiqing Shen and Jingxing Li and Xinyuan Shao and Blanca Inigo Romillo and Ankush Jindal and David Dreizin and Mathias Unberath},
      year={2024},
      eprint={2403.09827},
      archivePrefix={arXiv},
      primaryClass={eess.IV}
}

@article{shen2024fastsam,
  title={FastSAM-3DSlicer: A 3D-Slicer Extension for 3D Volumetric Segment Anything Model with Uncertainty Quantification},
  author={Shen, Yiqing and Shao, Xinyuan and Romillo, Blanca Inigo and Dreizin, David and Unberath, Mathias},
  journal={arXiv preprint arXiv:2407.12658},
  year={2024}
}