MITS-GAN

This is the implementation of our Computers in Biology and Medicine 2024 work 'MITS-GAN: Safeguarding Medical Imaging from Tampering with Generative Adversarial Networks'. This study introduces MITS-GAN, a novel approach to prevent tampering in medical images, with a specific focus on CT scans. The approach disrupts the output of the attacker's CT-GAN architecture by introducing imperceptible but yet precise perturbations. The original paper can be found here.
Please leave a star ⭐ and cite the following paper if you use this repository for your project.

@article{PASQUALINO2024109248,
      title = {MITS-GAN: Safeguarding medical imaging from tampering with generative adversarial networks},
      journal = {Computers in Biology and Medicine},
      volume = {183},
      pages = {109248},
      year = {2024},
      issn = {0010-4825},
      doi = {https://doi.org/10.1016/j.compbiomed.2024.109248},
      url = {https://www.sciencedirect.com/science/article/pii/S0010482524013337},
      author = {Giovanni Pasqualino and Luca Guarnera and Alessandro Ortis and Sebastiano Battiato}
}

MITS-GAN Architecture

Installation

Choose one of the two installation options:

Google Colab

Quickstart here 👉
Or load and run the MITS-GAN.ipynb on Google Colab following the instructions inside the notebook.

Installation on your PC

Install the required dependencies by running the following command in your terminal:

pip install --upgrade scipy matplotlib pandas tensorflow keras SimpleITK pydicom torch

Dataset

Dataset is available here

Training

Create a new folder Named MITS-GAN and put inside the following files/folders:

MITS-GAN
|
└───data/
│    └───ct_scan_1.raw
│    └───ct_scan_1.mhd
│    │        
│    |       ...
|    |
│    └───ct_scan_n.raw
│    └───ct_scan_n.mhd
|
└───models/
|     └───INJ
|     |    └───G_model_inj.h5
|     |    └───normalization.npy
|     |    └───equalization.pkl
|     |  
│     └───REM
│          └───G_model_rem.h5
|          └───normalization.npy
|          └───equalization.pkl
|
└───procedures/
│     └───attack_pipeline.py
|
└───utils/
|     └───dataloader.py
|     └───dicom_utils.py
|     └───equalizer.py
|     └───utils.py
|
│- config.py
│- discriminator.py
|- generator.py
|- scanDataset.py
|- train.py

Download the dataset and the pretrained model from the CT-GAN repo by contacting the authors and place them respectively inside the data and model folders. Run the following training script:

python3 MITS-GAN/train.py

Testing

To generate Protected images set save_img_result=True inside the training script. Then use the GUI.py script provided by the authors using the protected generated images.

Results

Qualitative results on the reconstruction task compared with images as manipulation targets.

Metric results evaluated between the following pairs on the: real-MITS-GAN, real-TAFIM, real-MITS-GAN tampered and real-TAFIM tampered. Lower values are better for RMSE and LPIPS, higher for PSNR.

Metric	Real (MITS-GAN)	Real (TAFIM)	Tampered (MITS-GAN)	Tampered (TAFIM)
RMSE	169.481	194.943	198.253	233.780
PSNR	27.949	21.702	21.237	21.469
LPIPS	0.170	0.383	0.226	0.391