Slide-SAM: Medical SAM meets sliding window

Quan Quan^1,2*, Fenghe Tang^3*, Zikang Xu³, Heqin Zhu³, S.Kevin Zhou^1,2,3

¹ Institute of Computing Technology, Chinese Academy of Sciences
² University of Chinese Academy of Sciences
³ School of Biomedical Engineering, University of Science and Technology of China
^* Equal Contribution

TODOs

Paper released
Code released
Slide-SAM-B weights released
Slide-SAM-H weights released

Models

Large scale Medical Image Pretrained Weights

name	resolution	Prompt	Weights
Slide-SAM-B	1024 x 1024	box & point	ckpt [code: 7be9]
Slide-SAM-H	1024 x 1024	box & point	ckpt [code: 05dy]

Getting Started

Install tutils tools

pip install trans-utils

Prepare datasets

We recommend you to convert the dataset into the nnUNet format.

00_custom_dataset
  imagesTr
    xxx_0000.nii.gz
    ...
  labelsTr
    xxx.nii.gz
    ...

Try to use the function organize in [nnunet-style](nnUNet/documentation/dataset_format.md at master · MIC-DKFZ/nnUNet (github.com)) or organize_by_names to prepare your custom datasets.

Then run :

python -m  datasets.generate_txt

A [example]_train.txt will be generated in ./datasets/dataset_list/

The content should be like below

01_BCV-Abdomen/Training/img/img0001.nii.gz	01_BCV-Abdomen/Training/label/label0001.nii.gz
01_BCV-Abdomen/Training/img/img0002.nii.gz	01_BCV-Abdomen/Training/label/label0002.nii.gz
01_BCV-Abdomen/Training/img/img0003.nii.gz	01_BCV-Abdomen/Training/label/label0003.nii.gz

Cache 3d volume into slices

After generating the [example]_train.txt file, check the config file configs/vit_b.yaml.

Update the params in dataset by yours. And the dataset_list should be the name of the generated txt file [example].

Then run

python -m datasets.cache_dataset3d

Start Training

Run training on multi-GPU

CUDA_VISIBLE_DEVICES=0,1,2,3 python -m core.ddp --tag debug

Sliding Inference and Test

python -m core.volume_predictor

Citation

If the code, paper and weights help your research, please cite:

@inproceedings{quan2024slide,
  title={Slide-SAM: Medical SAM Meets Sliding Window},
  author={Quan, Quan and Tang, Fenghe and Xu, Zikang and Zhu, Heqin and Zhou, S Kevin},
  booktitle={Medical Imaging with Deep Learning},
  year={2024}
}

License

This project is released under the Apache 2.0 license. Please see the LICENSE file for more information.

4.5 KiB Raw Blame History Unescape Escape