-------->

DeepDefacer is a MRI anonymization tool written in Python, on top of Tensorflow and Keras, that was developed by Anish Khazane in partnership with the Poldrack Lab at Stanford University. It can be used to quickly deface 3D MRI images of any resolution and size on commercial CPUs and GPUs. Its goal is to provide the community with an easy to use, efficient, and fast tool for defacing medical images that require anonymization for compliance with federal privacy laws (e.g HIPAA).

If you use DeepDefacer in your work, please refer to this citation for the current version:

@article{khazane2019state,
  title={DeepDefacer: Automatic Removal of Facial Features from MR Scans Via U-Net Image Segmentation},
  author={Anish Khazane, Julien Hoachuck, Dr. Chris Gorgowelski, Dr. Russell Poldrack},
  journal={in proceedings, arXiv preprint},
  year={2019}
}

If you use any of the architecture code from the ARFF-CNN, please also use the citation above to comply with its authors' instructions on referencing.

• Any Python version between 2.7 and 3.6.
• If you are using the GPU version of this library, please ensure that your GPU drivers are correctly installed and up to date. Please reference GPU Support for Tensorflow-GPU for further details on GPU setup.
• Input MRI images must have 3D structure and be saved in either .nii or .nii.gz format.

Deepdefacer can be easily installed on any operating system via Pypi. There are two versions of this package; CPU or GPU support. Please enter one of the following commands into your terminal window to begin installation, depending on your system specifications and desired python version.

CPU Support

pip install deepdefacer[tf_cpu] / pip3 install deepdefacer[tf_cpu]

GPU Support

pip install deepdefacer[tf_gpu] / pip3 install deepdefacer[tf_gpu]

Note: If you are using a ZSH-type shell, you may need to wrap the package name in quotations in order to successfully initiate the Pip installation. (e.g pip install "deepdefacer[...]").

Once installed, please enter deepdefacer --help into your terminal window to see a list of available options for this program. Defacing a 3D MRI image is extremely simple, and can be done with the following command:

deepdefacer --input_file <input filename>

The program will output a defaced image in the same directory as the input file, unless --defaced_output_path is specified.

You can also save the predicted mask for any MRI image by specifying the desired output filename and directory with the --mask_output_path option.

If you are interested in learning more about the model architecture that powers DeepDefacer, please view the files within ARFF-CNN, or re-visit this repository in the next few weeks to view our published manuscript on this work. We welcome any contributions to this codebase (whether it be a more robust model, package options, etc). If you are particularly interested in re-training DeepDefacer, please review the following steps:

1. Clone ARFF-CNN
2. Store data (MRI Images, volumetric or 2D in nature) within a directory named data in the same file location as ARFF-CNN.
3. Execute python train.py and the model will begin training for a single epoch. The model weights and graph will be stored in a hdf5 file in the same directory as ARFF-CNN.

Please note that ARFF-CNN is not fully integrated with a robust "argparser-like" command-line utility tool. This repository is currently being updated to support more command-line arguments for easier usage.