This repository is the official implementation of Unsupervised Domain Adaptation for Cross-modality Abdominal Organ Segmentation via Organ Attention Style Transfer and Dual-stage Pseudo Label Filtering of Team hilab on FLARE 2025 task 3 challenge.
We use OrganAttenCycleGAN to translate images from CTs to MR/PETs while preserving anatomical structures.
The preprocessing pipeline for converting 3D images to 2D slices before image translation can be found in the folder flare25_styletranslation/data, and the training script is located in OrganAttenCycleGAN.
The dual-stage segmentation framework is based on the previous winning solution.
We introduce a dual-stage pseudo-label filtering implementation, which can be found in lab_filter.ipynb.
Our method achieves the following performance on FLARE2025
| Dataset Name | MR DSC(%) | MR NSD(%) | PET DSC(%) | PET NSD(%) |
|---|---|---|---|---|
| Validation Dataset | 81.21% | 88.54% | 81.43% | 71.94% |
| Test Dataset | (?) | (?) | (?) | (?) |
A Dockerfile is provided, and the official Docker image is available on Hugging Face.
docker load -i hilab.tar.gz
docker container run --gpus "device=1" -m 28G --name hilab --rm -v $PWD/FLARE_Test/:/workspace/inputs/ -v $PWD/hilab_outputs/:/workspace/outputs/ hilab:latest /bin/bash -c "sh predict.sh"
We would like to thank the contributors of the FLARE25 dataset and the authors of the Champion Solution for FLARE24-Task3 for their valuable resources and efforts.
@inproceedings{wang2025unsupervised,
title={Unsupervised Domain Adaptation for Cross-modality Abdominal Organ Segmentation via Organ Attention Style Transfer and Dual-stage Pseudo Label Filtering},
author={Wang, Huamin and Wu, Jianghao and Wang, Guotai and Zhou, Xianhao and He, Jinlong},
booktitle={MICCAI 2025 FLARE Challenge}
}
@incollection{li20243d,
title={A 3d unsupervised domain adaptation framework combining style translation and self-training for abdominal organs segmentation},
author={Li, Jiaxi and Chen, Qiang and Ding, Haoyu and Liu, Hongying and Wan, Liang},
booktitle={MICCAI Challenge on Fast and Low-Resource Semi-supervised Abdominal Organ Segmentation},
pages={209--224},
year={2024},
publisher={Springer}
}
@incollection{wu2024unsupervised,
title={Unsupervised domain adaptation for abdominal organ segmentation using pseudo labels and organ attention cyclegan},
author={Wu, Jianghao and Zhang, Guoning and Qi, Xiaoran and Wang, Huamin and Liu, Xinya and Wang, Guotai},
booktitle={MICCAI Challenge on Fast and Low-Resource Semi-supervised Abdominal Organ Segmentation},
pages={225--242},
year={2024},
publisher={Springer}
}