Performance of deep learning-based segmentation of soft tissue sarcoma by MRI sequence, tumor type and location

This repository provides the trained weights and usage guide for the segmentation model described in our paper.

Implementation

Our model is built and trained using the nnU-Net v1 framework.

To use our model, please:

1. Install the official nnunet (v1) library.
2. Follow the standard nnU-Net directory structure for environment variables.

---

Pretrained Model Weights

The best-performing 3D full-resolution model weights are available here:

• Google Drive: Download Weights

Usage: Place the downloaded folder into your $RESULTS_FOLDER/nnUNet/3d_fullres/ directory. You can then run inference using:

nnUNet_predict -i INPUT_FOLDER -o OUTPUT_FOLDER -t TaskXXX_XXX -m 3d_fullres

Citation

If you find this work useful in your research, please consider citing:

@article{peng2026performance,
  title={Performance of deep learning-based segmentation of soft tissue sarcoma by MRI sequence, tumor type and location},
  author={Peng, Linkai and Perronne, Laetitia and Gennaro, Nicol{\`o} and Rashidi, Ahmad Pour and Kobus, Zuzanna and Seo, Mirinae and Borhani, Amir A and Kelahan, Linda and Subedi, Kamal and Savas, Hatice and others},
  journal={Skeletal Radiology},
  pages={1--10},
  year={2026},
  publisher={Springer}
}