Abstract
Low-dose CT denoising is crucial for reducing radiation exposure while maintaining diagnostic quality. However, developing robust denoising models often requires large datasets that are distributed across multiple institutions, raising privacy concerns. In this work, we propose FedFDD, a federated learning approach with frequency domain decomposition for low-dose CT denoising.
Our method introduces a frequency domain decomposition framework that enables effective federated learning while preserving data privacy. The framework consists of three key components: (1) a frequency domain decomposition module that separates low and high-frequency components, (2) a federated learning protocol that enables collaborative training without sharing raw data, and (3) a denoising module that reconstructs high-quality CT images.
We evaluate our approach on multiple low-dose CT datasets from different institutions. Experimental results demonstrate that our frequency domain approach significantly improves denoising performance in federated settings compared to traditional methods. The method shows robust performance across different noise levels and imaging protocols while maintaining strict privacy standards.
The proposed framework represents a significant advancement in privacy-preserving medical image processing, providing effective denoising that could improve diagnostic quality while reducing radiation exposure.
BibTeX
@inproceedings{
chen2024fedfdd,
title={Fed{FDD}: Federated Learning with Frequency Domain Decomposition for Low-Dose {CT} Denoising},
author={Xuhang Chen and Zeju Li and Zikun Xu and Kaijie Xu and Cheng Ouyang and Chen Qin},
booktitle={Medical Imaging with Deep Learning (MIDL 2024)},
year={2024},
url={https://openreview.net/forum?id=Zg0mfl10o2},
}