swinunetr-btcv-tiny

Model Overview

This repository contains the code for Swin UNETR [1,2]. Swin UNETR is the state-of-the-art on Medical Segmentation Decathlon (MSD) and Beyond the Cranial Vault (BTCV) Segmentation Challenge dataset. In [1], a novel methodology is devised for pre-training Swin UNETR backbone in a self-supervised manner. We provide the option for training Swin UNETR by fine-tuning from pre-trained self-supervised weights or from scratch.

The source repository for the training of these models can be found here.

Installing Dependencies

Dependencies for training and inference can be installed using the model requirements :

pip install -r requirements.txt

Intended uses & limitations

You can use the raw model for dicom segmentation, but it's mostly intended to be fine-tuned on a downstream task.

Note that this model is primarily aimed at being fine-tuned on tasks which segment CAT scans or MRIs on images in dicom format. Dicom meta data mostly differs across medical facilities, so if applying to a new dataset, the model should be finetuned.

How to use

To install necessary dependencies, run the below in bash.

git clone https://github.com/darraghdog/Project-MONAI-research-contributions pmrc
pip install -r pmrc/requirements.txt
cd pmrc/SwinUNETR/BTCV 

To load the model from the hub.

>>> from swinunetr import SwinUnetrModelForInference
>>> model = SwinUnetrModelForInference.from_pretrained('darragh/swinunetr-btcv-tiny')

Limitations and bias

The training data used for this model is specific to CAT scans from certain health facilities and machines. Data from other facilities may difffer in image distributions, and may require finetuning of the models for best performance.

Evaluation results

We provide several pre-trained models on BTCV dataset in the following.

<table> <tr> <th>Name</th> <th>Dice (overlap=0.7)</th> <th>Dice (overlap=0.5)</th> <th>Feature Size</th> <th># params (M)</th> <th>Self-Supervised Pre-trained </th> </tr> <tr> <td>Swin UNETR/Base</td> <td>82.25</td> <td>81.86</td> <td>48</td> <td>62.1</td> <td>Yes</td> </tr> <tr> <td>Swin UNETR/Small</td> <td>79.79</td> <td>79.34</td> <td>24</td> <td>15.7</td> <td>No</td> </tr> <tr> <td>Swin UNETR/Tiny</td> <td>72.05</td> <td>70.35</td> <td>12</td> <td>4.0</td> <td>No</td> </tr> </table>

Data Preparation

The training data is from the BTCV challenge dataset.

• Target: 13 abdominal organs including 1. Spleen 2. Right Kidney 3. Left Kideny 4.Gallbladder 5.Esophagus 6. Liver 7. Stomach 8.Aorta 9. IVC 10. Portal and Splenic Veins 11. Pancreas 12.Right adrenal gland 13.Left adrenal gland.
• Task: Segmentation
• Modality: CT
• Size: 30 3D volumes (24 Training + 6 Testing)

Training

See the source repository here for information on training.

BibTeX entry and citation info

If you find this repository useful, please consider citing the following papers:

@inproceedings{tang2022self,
  title={Self-supervised pre-training of swin transformers for 3d medical image analysis},
  author={Tang, Yucheng and Yang, Dong and Li, Wenqi and Roth, Holger R and Landman, Bennett and Xu, Daguang and Nath, Vishwesh and Hatamizadeh, Ali},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={20730--20740},
  year={2022}
}

@article{hatamizadeh2022swin,
  title={Swin UNETR: Swin Transformers for Semantic Segmentation of Brain Tumors in MRI Images},
  author={Hatamizadeh, Ali and Nath, Vishwesh and Tang, Yucheng and Yang, Dong and Roth, Holger and Xu, Daguang},
  journal={arXiv preprint arXiv:2201.01266},
  year={2022}
}

References

[1]: Tang, Y., Yang, D., Li, W., Roth, H.R., Landman, B., Xu, D., Nath, V. and Hatamizadeh, A., 2022. Self-supervised pre-training of swin transformers for 3d medical image analysis. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (pp. 20730-20740).

[2]: Hatamizadeh, A., Nath, V., Tang, Y., Yang, D., Roth, H. and Xu, D., 2022. Swin UNETR: Swin Transformers for Semantic Segmentation of Brain Tumors in MRI Images. arXiv preprint arXiv:2201.01266.