An attentive‑based generative model for medical image synthesis
pip install -r ADC-cycleGAN_requirements.txt
pip install -r cluster_requirements.txt
I recommend you use two environments to install. Maybe some packages will conflict when you install them into one environment.
Run the “cluster.ipynb” to implement the cluster. Before that, you should install the jupyter on your computer and nb_conda as well. Please use your computer path instead of my path.
Then, you will get the cluster dataset under “cluster_path” root.
python -u ADC-cycleGAN.py --G_rate=5 --lambdaG=10 --date='CBAM12RS' --t_cluster=4 --n_cluster=1 --loop_number=1 --out_dir=/home/jiayuan/ADC-cycleGAN/result --save_dir=/home/jiayuan/ADC-cycleGAN/result/model --dataset_path=/home/jiayuan/ADC-cycleGAN/dataset/cluster
For “--date” parameters, you can set any word you want, it is will control the save file name and used to distinguish the different files.
For “--t_cluster”, set the total cluster. You needn’t change, because we use the number of clusters 4 for experiments. If you want to reproduce our ablation study, you can change it to one number of 2-5, but you should go back to step 2 and change some code to generate the number of clusters 2-3 and 5 datasets.
For “--n_cluster”, set the number of cluster. you should change it between 1-4 for each loop.
For “--loop_number”, set the number of loop. We have 5 times experiments for each method, so this parameter should be run from 1 to 5.
Finally, you should run the code with the following parameters:
| Loop_number | t_cluster | n_cluster |
|---|---|---|
| 1 | 4 | 1 |
| 1 | 4 | 2 |
| 1 | 4 | 3 |
| 1 | 4 | 4 |
| 2 | 4 | 1 |
| 2 | 4 | 2 |
| 2 | 4 | 3 |
| 2 | 4 | 4 |
| ... | ... | ... |
| 5 | 4 | 1 |
| 5 | 4 | 2 |
| 5 | 4 | 3 |
| 5 | 4 | 4 |
I strongly recommend you write a shell file to run automatically.
Totally you should run 20 times for “ADC-cycleGAN.py”. You will get 40 models because, for each training, you will get two directions model. Each training required 10 hours in GTX 1080 Ti GPU.
After you get the models, you should evaluate the model one by one. Please run "evaluate.ipynb" and change "dataset_path", "weight_path", and "save_path" to your path. All the results will save in “evaluate.txt”. Then you can use excel to analyze the results.
If you use this code for your research, please cite our paper:
@article{wang2023attentive,
title={An attentive-based generative model for medical image synthesis},
author={Wang, Jiayuan and Wu, QM Jonathan and Pourpanah, Farhad},
journal={International Journal of Machine Learning and Cybernetics},
pages={3897–3910},
year={2023},
publisher={Springer} }