Implementation for ISBI 2024 paper Segmentation of Tiny Intracranial Hemorrhage via Learning-to-Rank Local Feature Enhancement by Shizhan Gong, Yuan Zhong, Yuqi Gong, Nga Yan Chan, Wenao Ma, Calvin Hoi-Kwan Mak, Jill Abrigo, and Qi Dou.
Our code is based on the framework of nnU-Net. The training data preprocessing and training protocol is exactly the same as the original nnU-Net.
Specifically, our modified variant can be found in the script.
For installing nnU-Net, please refer to the instructions.
As the original repo is continuing upgrading, which may make our modification fail. Please install the historical version we provided in our repo.
cd nnUNet
pip install -e .For data preprocessing, please refer to the instructions.
For training and predicting nnU-Net, please refer to the instructions.
To train with our learning-to-rank variant, change the command as
nnUNetv2_train DATASET_NAME_OR_ID 3d_fullres FOLD -tr nnUNetTrainer_rankTo predict with our learning-to-rank variant, change the command as
nnUNetv2_predict -i INPUT_FOLDER -o OUTPUT_FOLDER -d DATASET_NAME_OR_ID -c 3d_fullres -tr nnUNetTrainer_rankWe use the data stored in .nii.gz format, two sample cases can be found in the sample_data.
We provide several pre-trained checkpoints trained on our dataset correponding to different folds. You can download the checkpoint here.
If you find this work helpful, you can cite our paper as follows:
@INPROCEEDINGS{gong2024segmentation,
author={Gong, Shizhan and Zhong, Yuan and Gong, Yuqi and Chan, Nga Yan and Ma, Wenao and Mak, Calvin Hoi-Kwan and Abrigo, Jill and Dou, Qi},
booktitle={2024 IEEE 21th International Symposium on Biomedical Imaging (ISBI)},
title={Segmentation of Tiny Intracranial Hemorrhage via Learning-to-Rank Local Feature Enhancement},
year={2024}
}
Our code is based on nnU-Net.
For any questions, please contact [email protected]