Important:

ICASSP 2024 Poster. Paper: A Parameterized Generative Adversarial Network Using Cyclic Projection for Explainable Medical Image Classification.

Prerequisites

Linux
Python 3.8
NVIDIA GPU + CUDA CuDNN

anaconda virtual environment

conda install pytorch torchvision torchaudio pytorch-cuda=11.7 -c pytorch -c nvidia  

conda install tqdm  

conda install matplotlib==3.3.4  

conda install seaborn  

conda install scikit-learn

Dataset Discription

-covid_ct：

├─train├─X
│      │  001non-covid19.png
│      │  002non-covid19.png
│      ├─Y
│      │  001covid19.png
│      │  002covid19.png
├─valid├─X
│      │  001non-covid19.png
│      │  002non-covid19.png
│      ├─Y
│      │  001covid19.png
│      │  002covid19.png
├─test ├─X
│      │  001non-covid19.png
│      │  002non-covid19.png
│      ├─Y
│      │  001covid19.png
│      │  002covid19.png

Pretrain the binary classification networks

classification_hingeloss_preaugment

python train.py --dataroot icassp2024/augmented_covid --dataset_name covid --num_classes 1 \
--project_name convnext_tiny --model_name convnext_tiny --gpu_ids 0,1

python test.py --dataroot icassp2024/augmented_covid --dataset_name covid --num_classes 1 \
--classifier checkpoints/covid/convnext_tiny/convnext_tiny_best_netC.pth \
--project_name convnext_tiny --model_name convnext_tiny --gpu_ids 0

classification_cross-entropy_preaugment

python train.py --dataroot icassp2024/augmented_covid --dataset_name covid --num_classes 2 \
--project_name convnext_tiny --model_name convnext_tiny --gpu_ids 0,1

python test.py --dataroot icassp2024/augmented_covid --dataset_name covid --num_classes 2 \
--classifier checkpoints/covid/convnext_tiny/convnext_tiny_best_netC.pth \
--project_name convnext_tiny --model_name convnext_tiny --gpu_ids 0

ACGAN

python train.py --dataroot icassp2024/augmented_covid --dataset_name covid --num_classes 2 \
--project_name acgan_convnext_tiny --model_name convnext_tiny --gpu_ids 0,1

python test.py --dataroot icassp2024/augmented_covid --dataset_name covid --num_classes 2 \
--generator checkpoints/coivd/acgan_convnext_tiny/best_netG.pth
--classifier checkpoints/covid/acgan_convnext_tiny/best_netC.pth
--project_name acgan_convnext_tiny --model_name convnext_tiny --gpu_ids 0

VACGAN

python train.py --dataroot icassp2024/augmented_covid --dataset_name covid --num_classes 2 \
--project_name vacgan_convnext_tiny --model_name convnext_tiny --gpu_ids 0,1

python test.py --dataroot icassp2024/augmented_covid --dataset_name covid --num_classes 2 \
--generator checkpoints/coivd/vacgan_convnext_tiny/best_netG.pth
--classifier checkpoints/covid/vacgan_convnext_tiny/best_netC.pth
--project_name vacgan_convnext_tiny --model_name convnext_tiny --gpu_ids 0

ParaGAN

python train.py --dataroot icassp2024/augmented_covid --dataset_name covid --num_classes 1 \
--lambda_fake 0.2 --lambda_vertical 0.1 --classifier ./pretrained_covid/convnext_tiny_best_netC.pth \
--project_name cyclegan_convnext_tiny --model_name convnext_tiny --gpu_ids 0,1

python test.py --dataroot icassp2024/augmented_covid --dataset_name covid --num_classes 1 \
--generator_A2B checkpoints/covid/cyclegan_convnext_tiny/best_netG_A2B.pth \
--generator_B2A checkpoints/covid/cyclegan_convnext_tiny/best_netG_A2B.pth \
--classifier checkpoints/covid/cyclegan_convnext_tiny/best_netC.pth \
--project_name cyclegan_convnext_tiny --model_name convnext_tiny --gpu_ids 0

if you want to generate the tsne figure and the heatmaps, use the followed two lines in test.py:
  # test_gd(module, data_gd_loader, data_c_loader, memory_allocation, opt)
  # test_heatmap(module, data_gd_loader, data_c_loader, memory_allocation, opt)

yXiangXiong/ParaGAN

Important:

Prerequisites

Dataset Discription

Pretrain the binary classification networks