Sparse Dynamic Volume TransUNet with multi-level edge fusion for brain tumor segmentation

Highlights

•Propose a novel encoder-decoder-based 3D brain tumor segmentation network.

•Combine voxel information, inter-layer feature connection and intra-axis information.

•Dynamically extract 3D spatial features and capture fine-grained edge details.

•Introduce a local-to-global feature modeling strategy for accurate segmentation.

Datasets

The dataset used in this paper is BraTS2019 and 2020. The datasets can be found in kaggle, using the training set part.

Requirement

This article is implemented by Pytorch.

1. PyTorch 1.9.0
2. Some other libraries

Cite

@article{zhu2024sparse, title={Sparse Dynamic Volume TransUNet with multi-level edge fusion for brain tumor segmentation},

author={Zhu, Zhiqin and Sun, Mengwei and Qi, Guanqiu and Li, Yuanyuan and Gao, Xinbo and Liu, Yu},

journal={Computers in Biology and Medicine},

pages={108284},

year={2024},

publisher={Elsevier}

}