HF-Fed: Hierarchical based customized Federated Learning Framework for X-Ray Imaging

This is the official implementation of the MICCAI 2024 workshop paper HF-Fed: Hierarchical based customized Federated Learning Framework for X-Ray Imaging by Tajamul Ashraf and Tisha Madame.

Abstract

In clinical applications, X-ray technology plays a crucial role in noninvasive examinations like mammography, providing essen�tial anatomical information about patients. However, the inherent radi�ation risk associated with X-Ray procedures raises significant concerns. X-Ray reconstruction is crucial in medical imaging for creating detailed visual representations of internal structures, and facilitating diagnosis and treatment without invasive procedures. Recent advancements in deep learning (DL) have shown promise in X-ray reconstruction. Nevertheless, conventional DL methods often necessitate the centralized aggregation of substantial large datasets for training, following specific scanning protocols. This requirement results in notable domain shifts and privacy issues. To address these challenges, we introduce the Hierarchical Framework-based Federated Learning method (HF-Fed) for customized X-Ray Imaging. HF-Fed addresses the challenges in X-ray imaging optimization by decomposing the problem into two components: local data adaptation and holistic X-ray imaging. It employs a hospital-specific hierarchical framework and a shared common imaging network called the Network of Networks (NoN) for these tasks. The emphasis of the NoN is on acquiring stable features from a variety of data distributions. A hierarchical hypernetwork extracts domain-specific hyperparameters, conditioning the NoN for customized X-ray reconstruction. Experimental results demonstrate HF-Fed’s competitive performance, offering a promising solution for en�hancing X-Ray imaging without the need for data sharing. This study significantly contributes to the evolving body of literature on the potential advantages of federated learning in the healthcare sector. It offers valuable insights for policymakers and healthcare providers holistically.

Dataset

The dataset includes the following files and directories:

RSNA-Breast-Cancer-Detection-Dataset/
├── test_images/
│   ├── image_001.dcm
│   ├── image_002.dcm
│   └── ...
├── train_images/
│   ├── image_001.dcm
│   ├── image_002.dcm
│   └── ...
├── sample_submission.csv
├── test.csv
└── train.csv

Getting Started

1. Clone the Repository:

   git clone https://github.com/Tajamul21/HF-Fed.git
   cd HF-Fed

2. Download the Dataset:

   • Visit the Kaggle competition page
   • Download the dataset files
   • Extract the dataset files into the appropriate directories

Requirements

Our codes were implemented by PyTorch 1.10 and 11.3 CUDA version. If you wanna try our method, please first install the necessary packages as follows:

pip install requirements.txt

Our implementation is based on CTLib in simulating data and training IR-based methods. If you are interested in data simulation and IR-based networks, we recommend installing it. Furthermore, HF-Fed can be easily integrated into transformer-based methods with minor modifications.

Acknowledgments

Special thanks to Prof. Aditeshwar Seth for his support and guidance!

Contact

If you have any questions or suggestions about our work, please get in touch with me. [Email].

Citation

If you find this work useful in your research, please cite:

@misc{li2023diffusion,
      title={Your Diffusion Model is Secretly a Zero-Shot Classifier}, 
      author={Alexander C. Li and Mihir Prabhudesai and Shivam Duggal and Ellis Brown and Deepak Pathak},
      year={2023},
      eprint={2303.16203},
      archivePrefix={arXiv},
      primaryClass={cs.LG}
}