HF-Fed: Hierarchical Based Customized Federated Learning Framework for X-Ray Imaging

MICCAI 2024
Tajamul Ashraf1, Tisha Madame1,
1Indian Institute of Technology Delhi,
Paper Code arXiv Data
Deep Breath

Abstract

In clinical applications, X-Ray technology plays a crucial role in noninvasive examinations like mammography, providing essential anatomical information about patients. However, the inherent radiation 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 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 enhancing 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

HF-Fed Architecture

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 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.

Network of Networks (NoN)

In this research, we propose HF-Fed, a hypernetwork-based federated learning framework designed to tackle the non-iid challenge in X-Ray image reconstruction. Similar to FedAvg and FedProx, HF-Fed involves local updates for both the hypernetwork and imaging network, with only the imaging network’s updates aggregated on the server. Unlike FedBN, which normalizes data globally, HF-Fed adapts by performing local normalization due to varied X-Ray data distributions from different machines, challenging FedBN’s assumptions [21]. To address this, we introduce a hypernetwork to modulate feature maps within the imaging network, enabling hierarchical-driven self-normalization

Quantitative Results

Table 1 demonstrates that FL-based methods significantly improve performance with larger datasets, mitigating the non-iid problem. HF-Fed consistently delivers high performance across different dataset sizes, enhancing imaging quality effectively. All methods benefit from larger training samples, with HF-Fed remaining competitive. Our hypernetwork uses X-Ray geometry parameters to modulate feature maps, balancing stability and imaging performance, proving effective in achieving consistent and competitive result

Qualitative Results and Ablation


We conduct ablation studies to highlight the effectiveness of various components in HF-Fed. "†" and "‡" denote imaging networks without and with the hypernetwork, respectively, showing the significant role of the hypernetwork in improving imaging performance by addressing domain gap issues. Further, we evaluate our learning strategy by aggregating only the hypernetwork in rounds labeled "HF-Fed ♢," addressing the challenge of heterogeneous scanning parameters. Additional experiments explore the modulation scope of the hypernetwork, where "HF-Fed ⋆" and "HF-Fed ◦" scenarios focus on encoder and decoder modulation, respectively. Figure 2 shows the boxplots of PSNR based on w/o FL, FedAvg, FedProx, FedBN. Ditto, pFedHN, and HF-Fed for the postprocessing task. Results suggest similar performances across all modulation scenarios, indicating the effectiveness of modulating all layers for consistency and generality

BibTeX

@article{ashraf2024dmastermaskannealedtransformer,
        title={D-MASTER: Mask Annealed Transformer for Unsupervised Domain Adaptation in Breast Cancer Detection from Mammograms}, 
        author={Tajamul Ashraf and Krithika Rangarajan and Mohit Gambhir and Richa Gabha and Chetan Arora},
        year={2024},
        eprint={2407.06585},
        archivePrefix={arXiv},
        primaryClass={cs.CV},
        url={https://arxiv.org/abs/2407.06585}, 
  }