This repository contains the official implementations of
- 1 Fourier-Net: Fast Image Registration with Band-limited Deformation (AAAI-2023 oral presentation);
- 2 Fourier-Net+: Leveraging Band-Limited Representation for Efficient 3D Medical Image Registration.
Feb 2024. Our Fourier-Net won 2nd prize on the Oncoreg 2024 challenge.
- Solely FourierNet with NCC Loss / MIND Loss + Dice Loss + Diffusion Smoothness.
Oct 2023. Our Fourier-Net won 2nd prize on the Learn2Reg 2023 challenge.
- 1 NLST: our submitted displacement is the algorithmic average of LKU-Net and Cascaded Fourier-Net+. Best TRE_LM!
- 2 ThoraxCBCT: our submitted displacement is solely from Fourier-Net. Best TRE_LM!
| team | Task | TRE LM | TRE LM30 | TRE nodules | Dice | HD59 | SDLogJacDet |
|---|---|---|---|---|---|---|---|
| Birmingham | NLST | 1.425 | 1.926 | 1.135 | N/A | N/A | 0.045 |
| Birmingham | ThoraxCBCT | 4.0342 | 3.8522 | N/A | 0.5691 | 56.4998 | 0.0728 |
# Train
CUDA_VISIBLE_DEVICES=0 python train.py --start_channel 8 --using_l2 2 --smth_labda 1.0 --lr 1e-4 --trainingset 4 --checkpoint 403 --iteration 403001
# Test
CUDA_VISIBLE_DEVICES=0 python infer.py --start_channel 8 --using_l2 2 --smth_labda 1.0 --lr 1e-4 --trainingset 4 --checkpoint 403 --iteration 403001
# Test with Bilinear Interplotation for Mask
CUDA_VISIBLE_DEVICES=0 python infer_bilinear.py --start_channel 8 --using_l2 2 --smth_labda 1.0 --lr 1e-4 --trainingset 4 --checkpoint 403 --iteration 403001
# Report Results
python compute_dsc_jet_from_quantiResult.py
If you find the code helpful, please consider citing our work:
@inproceedings{jia2023fourier,
title={Fourier-Net: Fast Image Registration with Band-Limited Deformation},
author={Jia, Xi and Bartlett, Joseph and Chen, Wei and Song, Siyang and Zhang, Tianyang and Cheng, Xinxing and Lu, Wenqi and Qiu, Zhaowen and Duan, Jinming},
booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
volume={37},
number={1},
pages={1015--1023},
year={2023}
}
@article{jia2023fourierplus,
title={Fourier-Net+: Leveraging Band-Limited Representation for Efficient 3D Medical Image Registration},
author={Jia, Xi and Thorley, Alexander and Gomez, Alberto and Lu, Wenqi and Kotecha, Dipak and Duan, Jinming},
journal={arXiv preprint arXiv:2307.02997},
year={2023}
}
- Update 2D Fourier-Net pre-trained models. Nov 29 2022.
- Update 2D Fourier-Net training code. Mar 23 2023.
- Update 2D Fourier-Net-Diff training code and pre-trained models. Mar 23 2023.
- Update 3D Fourier-Net training code and pre-trained models. May 1 2023.
- Update 3D Fourier-Net-Diff training code and pre-trained models. May 1 2023.
- Update 2D Fourier-Net+ model. July 6 2023.
- Update 3D Fourier-Net+ model. July 6 2023.
- Update 2D Fourier-Net+ training/testing/pre-trained model. Nov 4 2023.
- Update 3D Fourier-Net+ training/testing/pretrained model. Nov 4 2023.
We would like to acknowledge the IC-Net, SYM-Net, and TransMorph projects, from which we have adopted some of the code used in our work.