- GFNet with random data augmentation using Pytorch
- GFNet with adversarial contrastive learning (coming soon)
Code developed and tested in Python 3.8.12 using PyTorch 1.10.0. Please refer to their official websites for installation and setup.
Some major requirements are given below
nibabel @ file:///home/conda/feedstock_root/build_artifacts/nibabel_1673318073381/work
numpy==1.24.2
pandas==1.5.3
torchio==0.18.90
torchvision==0.14.1+cu116
- ADNI : https://adni.loni.usc.edu/data-samples/access-data/
- AIBL :https://adni.loni.usc.edu/aibl-australian-imaging-biomarkers-and-lifestyle-study-of-ageing-18-month-data-now-released/
- OASIS:https://oasis-brains.org/
*** Notice*** : The data should be concluded in a .csv
file, which is described as follows: (Longitude data is also included)
Filename | Status (label) | Age | Gender | MMSE | Apoe |
---|---|---|---|---|---|
ADNI_002_S_0295 | 0 | 85 | 1 | 28 | 1 |
ADNI_002_S_0413 | 0 | 77 | 2 | 29 | 0 |
... | ... | ... | ... | ... | ... |
ADNI_141_S_1137 | 1 | 81 | 2 | 24 | 0 |
- Registration using FSL FLIRT function for one case
bash registation.sh original_data_floder/ADNI_002_S_0295.nii processed_data_folder/
- Also, we could register all the data in the folder,
python3 registration.py ADNI /home1/zhangsj/AD_class/output_file
Notice : the output file must be in a absolute path
- Step 2: conduct z-score normalization
- Step 3: clip the intensity within a range
numpy.clip(X, -1,2.5)
- Step 4: conduct the background removel (We use
bet2
in FSL to refine the skull stripping)
python3 back_remove.py folder_output/ folder_before_final/
To perform GF-Net with data augmentation on ADNI1 using a 1-gpu machine, run:
nohup python3 main_sl.py \
--name ADNI1 \
--seed 123 \
--batch-size 10 \
--size (32,32,32) \
--out_channel 1 \
--gf_depth 4 \
--gfopc 2 \
--l_lr 0.0005 \
--optim Adam \
--prob 1.0 \
--reg True \
--epochs 400 \
--epochs 100
If you have any question about the implementation of GFNet or data pre-processing, please contact me through
zsjxll@gmail.com
If you find our work beneficial to your work, please cite our paper
@inproceedings{zhang20223d,
title={3D Global Fourier Network for Alzheimer’s Disease Diagnosis Using Structural MRI},
author={Zhang, Shengjie and Chen, Xiang and Ren, Bohan and Yang, Haibo and Yu, Ziqi and Zhang, Xiao-Yong and Zhou, Yuan},
booktitle={Medical Image Computing and Computer Assisted Intervention--MICCAI 2022: 25th International Conference, Singapore, September 18--22, 2022, Proceedings, Part I},
pages={34--43},
year={2022},
organization={Springer}
}