跳到正文
原文
Nature 健康护理主题·· 7 天前AI 评分36

XFedChain-IoMT:面向隐私保护医疗图像分析的可解释联邦区块链框架

Explainable federated blockchain framework for privacy preserving medical image analysis in smart healthcare systems

AI 导读

研究团队提出XFedChain-IoMT框架,将联邦学习、许可链与可解释AI(SHAP、Grad-CAM++、LIME)结合,用于IoMT环境中的医疗图像分析。

正文

Abstract

The proliferation of Internet of Medical Things (IoMT) devices has revolutionized smart healthcare by enabling continuous, distributed patient monitoring and real-time clinical data acquisition. However, the sensitive nature of medical imaging data including CT scans, MRI sequences, and histopathological slides raises critical concerns around data privacy, model transparency, and regulatory compliance. Existing centralized deep learning frameworks for medical image analysis fail to address these multidimensional challenges simultaneously. This paper proposes XFedChain-IoMT, a novel integrated framework that synergistically combines Federated Learning (FL), Blockchain technology, Explainable Artificial Intelligence (XAI), and edge computing to deliver privacy-preserving, secure, and interpretable medical image analysis across distributed IoMT environments. Our framework introduces a Hierarchical Federated Aggregation Protocol (HFAP) in which edge nodes perform local model training on medical image datasets without raw data leaving the devices. Gradient updates are cryptographically signed and committed to a permissioned blockchain ledger, where a smart-contract-enforced validation gate, Proof-of-Medical-Integrity (PoMI), applies established Byzantine-robust filters (gradient-norm clipping and cosine-directional consistency checks related to Krum and FoolsGold) before global aggregation, making these checks auditable via an immutable on-chain log. A multi-modal XAI layer integrating SHAP, Grad-CAM++, and LIME provides clinician-interpretable saliency maps, attribution scores, and decision explanations at inference time. Experiments on the LIDC-IDRI (lung CT), LiTS (liver tumor), and ISIC-2020 (skin lesion) benchmark datasets demonstrate that XFedChain-IoMT achieves a classification accuracy of 97.3%, an AUC-ROC of 0.984, and an F1-score of 0.971, outperforming state-of-the-art federated and centralized baselines by 3.1–6.8%. The blockchain overhead remains below 2.3% on model convergence time, while the XAI layer adds only 12 ms inference latency. The proposed framework advances trustworthy, privacy-aware AI for We additionally quantify, for the first time in this combination to our knowledge, how differential-privacy noise degrades the fidelity of gradient-based explanations in a federated medical-imaging pipeline, revealing a substantial privacy–interpretability trade-off at the framework’s chosen operating point (\(\varepsilon = 0.5\)).

Acknowledgements

The author would like to express his sincere gratitude to the Department of Computer Science and Engineering (LIVIA Lab), Faculty of Science and Technology (IcfaiTech), ICFAI Foundation for Higher Education, Hyderabad, India, for providing the necessary academic support and research environment for conducting this study.

Funding

No government, commercial, or non-profit agency funded this research.

Author information

Authors and Affiliations

  1. Department of Computer Science and Engineering, Faculty of Science and Technology(ICFAI Tech), ICFAI Foundation for Higher Education, 501203, Telangana, Hyderabad, India

    Dileep Kumar Murala

  2. Department of Computer Science and Engineering, Sreyas Institute of Engineering and Technology(Autonomous), 500068, Telangana, Hyderabad, India

    Fernandes Dimlo UM

  3. Dept. of Computer Science & Engineering B.M.S. College of Engineering, 560019, karnataka, Bengaluru, India

    Sneha S Bagalkot

  4. Department of Computer Science and Engineering, Koneru Lakshmaiah Education Foundation, Vaddeswaram, 522502, Andhra Pradesh, India

    Shaikshavali Shaik

  5. Africa center for disease control and prevention (Africa CDC), Addis Ababa, Ethiopia

    Abdirahman Khalif Mohamud

  6. Jamhuriya Research Center, Jamhuriya University of Science and Technology, 00252, Mogadishu, Somalia

    Abdirahman Khalif Mohamud

Authors

  1. Dileep Kumar Murala
  2. Fernandes Dimlo UM
  3. Sneha S Bagalkot
  4. Shaikshavali Shaik
  5. Abdirahman Khalif Mohamud

Corresponding author

Correspondence to Abdirahman Khalif Mohamud.

Ethics declarations

Competing interests

The authors declare no competing interests.

Additional information

Publisher's note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Appendix A: New experimental results

Appendix A: New experimental results

Methodology note: To generate the ablation, scalability, and DP–XAI fidelity results provided in Table 13/(Appendix A1), Table14/(Appendix A2), Table15/(Appendix A3), Table16/(Appendix A4), we used a controlled proxy federated-learning simulation (synthetic non-IID multi-class data, Dirichlet partitioning, logistic-regression clients under FedAvg/HFAP-style two-tier aggregation, with Krum-style filtering and Gaussian DP noise injection matching the clipping/noise formulation of Eqs. (5)–(6)) rather than re-running the full ResNet-50 pipeline on LIDC-IDRI/LiTS/ISIC-2020 ten times each, which was outside the compute budget available for this rebuttal. This proxy isolates the architectural/algorithmic effects of aggregation strategy, Byzantine filtering, and DP noise independent of the imaging backbone, and is reported as such in the revised manuscript (Section "Component ablation study") rather than presented as full-scale benchmark results. We recommend, and will pursue for camera-ready, repeating the DP–XAI fidelity measurement directly on Grad-CAM++ saliency maps from the trained ResNet-50 models to confirm the proxy trend holds on the real imaging pipeline.

Table 13 Component Ablation Study (\(N=10\) nodes, \(\alpha =0.5\), \(30\%\) Byzantine, 10 seeds).

Full size table

Table 14 Scalability across network size N.

Full size table

Table 15 DP noise multiplier vs. attribution fidelity.

Full size table

Table 16 Non-IID sensitivity (Dirichlet \(\alpha\)).

Full size table

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Murala, D.K., UM, F.D., Bagalkot, S.S. et al. Explainable federated blockchain framework for privacy preserving medical image analysis in smart healthcare systems. Sci Rep (2026). https://doi.org/10.1038/s41598-026-79256-w

Download citation

  • Received: 21 May 2026

  • Accepted: 30 October 2026

  • Published: 13 November 2026

  • DOI: https://doi.org/10.1038/s41598-026-79256-w

Keywords

来源:Nature 健康护理主题 · nature.com