Back

The future of digital health with federated learning

Nicola RiekeJonny HancoxWenqi LiFausto MilletarìHolger R. RothShadi AlbarqouniSpyridon BakasMathieu N. GaltierBennett A. LandmanKlaus Maier-Hein

article2020npj Digital Medicine3,284 citations

Explains how federated learning can unlock siloed medical data without centralizing it and lays out the privacy, governance, and technical challenges that digital-health deployments must solve.

Listen

Modern healthcare systems collect vast amounts of clinical data that could drive powerful artificial intelligence and machine learning tools, yet most of this information remains isolated in institutional silos. Storing and pooling patient data in centralized repositories creates severe regulatory, ethical, and privacy risks, as simple anonymization often fails to prevent patient re-identification from complex imaging or genetic records. Furthermore, healthcare organizations often restrict access to their curated datasets due to high development costs and proprietary value. Without access to diverse, real-world data across multiple demographics and equipment standards, clinical artificial intelligence models risk developing severe demographic biases and poor generalizability to unseen patient populations.

This article evaluates the viability, potential impact, and technical hurdles of federated learning as a collaborative, privacy-preserving paradigm for digital healthcare. The analysis synthesizes evidence from recent multi-institutional implementations across medical imaging, electronic health records, and pharmaceutical discovery to assess how decentralized training performs compared to traditional centralized data pooling.

The review finds that federated learning allows institutions to train shared consensus models locally behind their own firewalls, transferring only model parameters rather than sensitive patient data. Evidence shows that models trained through federated methods consistently achieve diagnostic accuracy comparable to models trained on centrally pooled data lakes, while significantly outperforming models restricted to isolated, single-institution datasets. The approach successfully handles diverse applications, including brain tumor segmentation, breast cancer treatment response prediction, and multi-company pharmaceutical drug discovery. Additionally, moving algorithms to the data rather than transferring storage-intensive files eliminates redundant data duplication and scales efficiently as datasets grow.

These findings indicate that federated learning can accelerate the transition to high-precision, value-based healthcare while mitigating compliance risks associated with strict data privacy laws. By broadening demographic and rare-disease representation without compromising institutional data governance, healthcare networks can deploy more equitable diagnostic tools. However, organizations must account for trade-offs, as advanced privacy safeguards like differential privacy and encrypted model sharing can introduce computational overhead or minor reductions in predictive performance.

To adopt this framework effectively, participating institutions should establish standardized data protocols, synoptic reporting formats, and clear governance agreements regarding contribution metrics and model ownership before initiating training. Organizations must also invest in adequate on-premise compute infrastructure and robust network security to protect against indirect privacy leakage, such as model inversion attacks.

The authors express strong confidence in the clinical and operational value of federated learning, while noting key limitations around data heterogeneity, lack of direct data visibility for debugging failure cases, and non-standardized network environments. Leaders should treat federated learning not as an immediate complete replacement for all data sharing, but as a mature, highly promising framework requiring continuous technical refinement and cross-institutional alignment.

Centre for the Governance of AI (GovAI)German Cancer Research CenterHeidelberg UniversityImperial College LondonIntelKing's College LondonNational Institutes of HealthNVIDIAOpenMinedOwkinTechnical University of MunichUniversity of OxfordUniversity of PennsylvaniaVanderbilt Universitycorrelation analysisfederated learningintegrated dataDataLearningFederated
arXiv: 2003.08119
  • Paper: Federated Machine Learning, Qiang Yang et al. (2019). This foundational survey defines federated learning architectures and classification types that directly enable the privacy-preserving machine learning paradigm applied in the source paper.
  • Paper: Federated Learning: Challenges, Methods, and Future Directions, Tian Li et al. (2019). This paper establishes the core technical challenges of federated learning, providing the foundational optimization and privacy context assumed by the source paper.
Cover for The future of digital health with federated learning

Abstract

Data-driven machine learning (ML) has emerged as a promising approach for building accurate and robust statistical models from medical data, which is collected in huge volumes by modern healthcare systems. Existing medical data is not fully exploited by ML primarily because it sits in data silos and privacy concerns restrict access to this data. However, without access to sufficient data, ML will be prevented from reaching its full potential and, ultimately, from making the transition from research to clinical practice. This paper considers key factors contributing to this issue, explores how federated learning (FL) may provide a solution for the future of digital health and highlights the challenges and considerations that need to be addressed.

Topics

correlation analysisfederated learningintegrated dataprivacy concerns

Keywords

DataLearningFederatedHealthFederated LearningDigitalTrainingModelDigital HealthMay

Table of Contents

  • 1 Introduction
  • 2 Data-driven Medicine Requires Federated Efforts
  • 2.1 The Reliance on Data
  • 2.2 The Promise of Federated Efforts
  • 2.3 Impact on Stakeholders
  • 3 Technical Considerations
  • 3.1 Federated Learning Definition
  • 3.2 Relation to Similar Strategies
  • 3.3 Challenges and Considerations
  • 3.3.1 Privacy and Security
  • 3.3.2 Data heterogeneity
  • 3.3.3 Traceability and accountability
  • 3.3.4 System architecture
  • 3.3.5 Initiatives and consortia
  • 4 Conclusion
  • 5 Acknowledgement
  • References

Knowls

  1. Knowl 1 — Mathematical Formulation of Federated Learning for Distributed Healthcare Data

    definition

    Federated learning (FL) is a distributed learning paradigm wherein multiple decentralized participating parties collaboratively optimize a global statistical model without sharing or centralizing their private datasets.

    Let KK denote the number of participating institutions. Each party k∈{1,…,K}k \in \{1, \dots, K\} holds a private local dataset XkX_k that remains behind its institutional firewall. The global objective function L(X;ϕ)\mathcal{L}(X; \phi), parameterized by model weights ϕ\phi, is formulated as a weighted combination of local loss functions Lk\mathcal{L}_k:

    min⁡ϕL(X;ϕ)whereL(X;ϕ)=∑k=1KwkLk(Xk;ϕ)\min_{\phi} \mathcal{L}(X; \phi) \quad \text{where} \quad \mathcal{L}(X; \phi) = \sum_{k=1}^K w_k \mathcal{L}_k(X_k; \phi)

    Here, wk>0w_k > 0 represents the weighting coefficient assigned to client kk (typically proportional to the client's dataset size, satisfying ∑k=1Kwk=1\sum_{k=1}^K w_k = 1).

    In standard implementations, each participant conducts multiple local optimization steps on Lk(Xk;ϕ)\mathcal{L}_k(X_k; \phi) before synchronizing parameter updates. As the number of local optimization steps per round increases, the convergence guarantees towards the exact minimizer of the global objective function decrease, introducing a trade-off between communication efficiency and theoretical convergence.

  2. Knowl 2 — Centralized Hub-and-Spoke Federated Averaging Algorithm

    algorithm

    In a centralized hub-and-spoke federated learning architecture, a central aggregation server coordinates iterative model distribution and weight averaging across KK parallel participating nodes without inspecting local patient datasets.

    procedure FederatedAveraging
        Input: Total communication rounds TT, number of participating clients KK
        Initialize global model weights W(0)W^{(0)}
        for t=1t = 1 to TT do
            for each client k∈{1,…,K}k \in \{1, \dots, K\} in parallel do
                Send global model W(t−1)W^{(t-1)} to client kk
                Execute local training at client kk to minimize local loss Lk(Xk;W(t−1))\mathcal{L}_k(X_k; W^{(t-1)})
                Compute local model update ΔWk(t−1)\Delta W_k^{(t-1)} and record local training iterations/sample count NkN_k
                Send (ΔWk(t−1),Nk)(\Delta W_k^{(t-1)}, N_k) back to central server
            end for
            Aggregate updates at the server:
            W(t)←W(t−1)+1∑k=1KNk∑k=1K(Nk⋅ΔWk(t−1))W^{(t)} \leftarrow W^{(t-1)} + \frac{1}{\sum_{k=1}^K N_k} \sum_{k=1}^K (N_k \cdot \Delta W_k^{(t-1)})
        end for
        return W(T)W^{(T)}
    end procedure

    The central aggregation node collects weight updates ΔWk(t−1)\Delta W_k^{(t-1)} weighted by the local sample contribution NkN_k, updating the consensus parameter vector W(t)W^{(t)} before the next training iteration.

  3. Knowl 3 — Taxonomy of Network Topologies and Compute Plans in Medical FL

    model/method

    Federated learning systems are classified across two orthogonal architectural dimensions: network topology and execution compute plans.

    1. Network Topologies (Communication Architecture):
    • Centralized (Hub & Spoke): A central aggregation server coordinates communication, collects model updates, performs aggregation, and redistributes consensus models to local training nodes.
    • Decentralized (Peer-to-Peer): Participating nodes directly connect to one or more peer nodes without an intermediary server; aggregation is computed independently on individual nodes in parallel.
    • Hierarchical: Nodes are organized into distinct sub-federations (e.g., regional or organizational clusters) that aggregate locally before passing updates to higher-level coordination layers.
    • Hybrid Hierarchical: A composition of peer-to-peer subnetworks and aggregation-server hubs operating across institutional tiers.
    1. Compute Plans (Model Trajectories):
    • Aggregation Server Plan: The global model parameters are broadcast to nodes, updated locally in parallel, and returned to a centralized aggregation step per round.
    • Peer-to-Peer Plan: Model parameters propagate directly across connected neighbor sites, where each node aggregates incoming weights from its neighbors.
    • Sequential Training / Cyclic Transfer Learning: A single model traverses participating institutions sequentially, updating its weights on each local site's private data before migrating to the next participant.

  4. Knowl 4 — Data Heterogeneity and Local Optimality Discrepancies in Healthcare FL

    limitation

    Healthcare data across institutions violates the independent and identically distributed (IID) assumption due to variations in medical imaging acquisition protocols, scanner manufacturers, clinical workflows, and regional patient demographics.

    Standard federated optimization algorithms (such as FedAvg) degrade in performance or diverge under non-IID conditions. Furthermore, when data distributions are inhomogeneous across sites, the globally optimal parameter configuration ϕ∗=arg⁡min⁡ϕ∑k=1KwkLk(Xk;ϕ)\phi^* = \arg\min_\phi \sum_{k=1}^K w_k \mathcal{L}_k(X_k; \phi) may not correspond to the locally optimal configuration ϕk∗=arg⁡min⁡ϕLk(Xk;ϕ)\phi_k^* = \arg\min_\phi \mathcal{L}_k(X_k; \phi) for a specific participant kk. Consequently, global model convergence can result in suboptimal inference accuracy on an individual clinic's specific population unless explicitly handled by optimization adaptations (such as proximal regularization via FedProx, multi-task learning, part-data sharing, or domain adaptation techniques).

  5. Knowl 5 — Information Leakage Vectors and Privacy Trade-Offs in Federated Learning

    limitation

    Although federated learning prevents direct transmission of patient records, sharing parameter updates ΔW\Delta W or gradients ∇ϕLk\nabla_\phi \mathcal{L}_k exposes medical data to indirect information leakage risks:

    • Model Inversion and Gradient Leakage: Adversaries observing model updates over successive rounds can mathematically reconstruct underlying training inputs or raw medical images via gradient-matching optimization and inversion attacks.
    • Membership and Property Inference: Neural networks are susceptible to memorization, enabling white-box or black-box attacks that determine whether a specific patient's record was present in a local training cohort.
    • Adversarial and Model Manipulation Attacks: Malicious participants can execute gradient-ascent attacks to deliberately induce memorization in other nodes or poison the global consensus model.

    Mitigating these attacks requires deploying differential privacy (DP), secure multi-party computation (SMPC), or gradient truncation. However, adding differential privacy noise introduces an inherent trade-off between privacy protection and model utility, which can degrade diagnostic accuracy.

  6. Knowl 6 — Trust Frameworks in Healthcare Federated Consortia

    model/method

    Federated networks in healthcare operate under two distinct trust regimes that determine security requirements:

    1. Trusted Consortia: Collaborations among a known, limited number of vetted healthcare institutions bound by enforceable legal agreements and governance contracts. Because participant motivations are assumed non-adversarial, lightweight security mechanisms suffice, reducing compute and communication overhead.
    2. Non-Trusted Consortia: Large-scale or open networks where participant identity and behavior cannot be legally or procedurally guaranteed. Such environments require defensive architectures, including secure party authentication, execution integrity verification, encrypted model aggregation, differential privacy noise injection, and protections against adversarial data poisoning or reconstruction attacks.

  7. Knowl 7 — Structural Distinctions of Cross-Silo Healthcare FL versus Edge-Device FL

    model/method

    Cross-silo federated learning in healthcare differs fundamentally from consumer edge-device federated learning (e.g., mobile phone deployments):

    • Compute Capability: Healthcare nodes consist of dedicated on-premise servers or private cloud instances equipped with high-performance computational hardware (GPUs), allowing the training of large deep learning architectures (millions of parameters) and more local optimization steps per communication round.
    • Network Reliability: Participating hospital institutions operate over high-throughput, reliable institutional networks, allowing richer model parameter exchange compared to low-bandwidth mobile connections.
    • Governance and Administration: Healthcare federations often incorporate an 'honest broker' (a trusted third-party intermediary orchestrating access and aggregation without inspecting raw data) or deploy trustless cryptographic architectures to enforce protocol compliance, node scheduling, and communication redundancy.

  8. Knowl 8 — Traceability, Verification, and Local Inspection Constraints in Clinical FL

    model/method

    Deploying federated learning in safety-critical clinical environments imposes governance and interpretability constraints:

    • Auditability and Execution Integrity: Full provenance tracking of data access histories, hardware/software training environments, hyperparameter configurations, and parameter convergence curves is required for regulatory compliance and clinical validation.
    • Data Shapley and Contribution Valuation: Evaluating the computational resources spent and the dataset quality provided by each participating institution enables fair revenue models and compensation schemes across collaborative consortia.
    • Black-Box Data Inspection Barrier: Because developers cannot directly access or visualize raw out-of-institution training data, identifying the root causes of individual model failure cases or unexpected statistical behaviors is restricted. Mitigating this requires secure intra-node viewing facilities or localized model explainability methods that preserve patient confidentiality.

Coverage note — Detailed descriptions of specific third-party clinical initiatives (e.g., FeTS, MELLODDY, HealthChain) and prior empirical research studies cited in the perspective were omitted to focus exclusively on the paper's core conceptual, architectural, mathematical, and governance contributions for federated digital health.

References

  1. 1.LeCun, Y., Bengio, Y. & Hinton, G. Deep learning. Nature 521, 436 (2015).
  2. 2.Wang, F., Casalino, L. P. & Khullar, D. Deep learning in medicine—promise, progress, and challenges. JAMA Intern. Med. 179, 293–294 (2019).
  3. 3.Chartrand, G. et al. Deep learning: a primer for radiologists. Radiographics 37, 2113–2131 (2017).
  4. 4.De Fauw, J. et al. Clinically applicable deep learning for diagnosis and referral in retinal disease. Nat. Med. 24, 1342 (2018).
  5. 5.Sun, C., Shrivastava, A., Singh, S. & Gupta, A. Revisiting unreasonable effectiveness of data in deep learning era. In Proceedings of the IEEE international conference on computer vision, 843–852 (IEEE, 2017).
  6. 6.Van Panhuis, W. G. et al. A systematic review of barriers to data sharing in public health. BMC Public Health 14, 1144 (2014).
  7. 7.Rocher, L., Hendrickx, J. M. & De Montjoye, Y.-A. Estimating the success of re-identifications in incomplete datasets using generative models. Nat. Commun. 10,1–9 (2019).
  8. 8.Schwarz, C. G. et al. Identification of anonymous mri research participants with face-recognition software. N. Engl. J. Med. 381, 1684–1686 (2019).
  9. 9.McMahan, B., Moore, E., Ramage, D., Hampson, S. & y Arcas, B. A. Communication-efficient learning of deep networks from decentralized data. In Artificial Intelligence and Statistics, 1273–1282. https://scholar.google.de/scholar?hl=de&as_sdt=0%2C5&q=Communicationefficient+learning+of+deep+networks+from+decentralized+data&btnG= (2017).
  10. 10.Li, T., Sahu, A. K., Talwalkar, A. & Smith, V. Federated learning: Challenges, methods, and future directions. IEEE Signal Processing Magazine 37, 50–60 (IEEE, 2020).
  11. 11.Yang, Q., Liu, Y., Chen, T. & Tong, Y. Federated machine learning: concept and applications. ACM Trans. Intell. Syst. Technol. (TIST) 10, 12 (2019).
  12. 12.Kairouz, P. et al. Advances and open problems in federated learning. arXiv preprint arXiv:1912.04977 (2019).
  13. 13.Lee, J. et al. Privacy-preserving patient similarity learning in a federated environment: development and analysis. JMIR Med. Inform. 6, e20 (2018).
  14. 14.Brisimi, T. S. et al. Federated learning of predictive models from federated electronic health records. Int. J. Med. Inform. 112, 59–67 (2018).
  15. 15.Roy, A. G., Siddiqui, S., Pölsterl, S., Navab, N. & Wachinger, C. Braintorrent: a peer-to-peer environment for decentralized federated learning. arXiv preprint arXiv:1905.06731 (2019).
  16. 16.Li, W. et al. Privacy-preserving federated brain tumour segmentation. In International Workshop on Machine Learning in Medical Imaging, 133–141 (Springer, 2019).
  17. 17.Sheller, M. J., Reina, G. A., Edwards, B., Martin, J. & Bakas, S. Multi-institutional deep learning modeling without sharing patient data: a feasibility study on brain tumor segmentation. In International MICCAI Brainlesion Workshop, 92–104 (Springer, 2018).
  18. 18.Li, X. et al. Multi-site fmri analysis using privacy-preserving federated learning and domain adaptation: abide results. arXiv preprint arXiv:2001.05647 (2020).
  19. 19.Huang, L. et al. Patient clustering improves efficiency of federated machine learning to predict mortality and hospital stay time using distributed electronic medical records. J. Biomed. Inform. 99, 103291 (2019).
  20. 20.Xu, J. & Wang, F. Federated learning for healthcare informatics. arXiv preprint arXiv:1911.06270 (2019).
  21. 21.Roy, A. & Banerjee, A. Ibm’s merge healthcare acquisition. https://www.reuters.com/article/us-merge-healthcare-m-a-ibm/ibm-to-buy-merge-healthcare-in-1-billion-deal-idUSKCN0QB1ML20150806 (2015) (Accessed 10 February 2020).
  22. 22.Nhs scotland’s national safe haven. https://www.gov.scot/publications/charter-safe-havens-scotland-handling-unconsented-data-national-health-service-patient-records-support-research-statistics/pages/4/ (2015) (Accessed 10 February 2020).
  23. 23.Cuggia, M. & Combes, S. The french health data hub and the german medical informatics initiatives: Two national projects to promote data sharing in healthcare. Yearbook Med. Informat. 28, 195–202 (2019).
  24. 24.Health Data Research UK. https://www.hdruk.ac.uk/ (Health Data Research UK, 2020) (Accessed 10 Feb 2020).
  25. 25.Sporns, O., Tononi, G. & Kötter, R. The human connectome: a structural description of the human brain. PLoS Comput. Biol. 1, e42, https://doi.org/10.1371/journal.pcbi.0010042 (2005).
  26. 26.Sudlow, C. et al. Uk biobank: an open access resource for identifying the causes of a wide range of complex diseases of middle and old age. PLoS Med. 12, e1001779. https://doi.org/10.1371/journal.pmed.1001779 (2015).
  27. 27.Clark, K. et al. The cancer imaging archive (tcia): maintaining and operating a public information repository. J. Digit. Imaging. 26, 1045–1057 (2013).
  28. 28.Wang, X. et al. Chestx-ray8: Hospital-scale chest X-ray database and benchmarks on weakly-supervised classification and localization of common thorax diseases. In Proceedings of the IEEE conference on computer vision and pattern recognition, 2097–2106 (IEEE, 2017).
  29. 29.Yan, K., Wang, X., Lu, L. & Summers, R. M. Deeplesion: automated mining of large-scale lesion annotations and universal lesion detection with deep learning. J Med. Imaging. 5, 036501 (2018).
  30. 30.Tomczak, K., Czerwińska, P. & Wiznerowicz, M. The cancer genome atlas (tcga): an immeasurable source of knowledge. Contemp. Oncol. 19, A68 (2015).
  31. 31.Jack Jr., C. R. et al. The alzheimer’s disease neuroimaging initiative (adni): Mri methods. J. Magn. Reson. Imaging 27, 685–691 (2008).
  32. 32.Grand Challenge-a Platform for End-to-end Development of Machine Learning Solutions in Biomedical Imaging. https://grand-challenge.org/ (2020) (Accessed 24 July 2020).
  33. 33.Litjens, G. et al. 1399 h&e-stained sentinel lymph node sections of breast cancer patients: the camelyon dataset. GigaScience 7, giy065 (2018).
  34. 34.Menze, B. H. et al. The multimodal brain tumor image segmentation benchmark (brats). IEEE Trans. Med. Imaging 34, 1993–2024 (2014).
  35. 35.Bakas, S. et al. Identifying the best machine learning algorithms for brain tumor segmentation, progression assessment, and overall survival prediction in the brats challenge. arXiv preprint arXiv:1811.02629 (2018).
  36. 36.Bakas, S. et al. Advancing the cancer genome atlas glioma MRI collections with expert segmentation labels and radiomic features. Sci. Data 4, 170117 (2017).
  37. 37.Simpson, A. L. et al. A large annotated medical image dataset for the development and evaluation of segmentation algorithms. arXiv preprint arXiv:1902.09063 (2019).
  38. 38.Yeh, F.-C. et al. Quantifying differences and similarities in whole-brain white matter architecture using local connectome fingerprints. PLoS Comput. Biol. 12, e1005203 (2016).
  39. 39.Chang, K. et al. Distributed deep learning networks among institutions for medical imaging. J. Am. Med. Inform. Assoc. 25, 945–954 (2018).
  40. 40.Shokri, R., Stronati, M., Song, C. & Shmatikov, V. Membership inference attacks against machine learning models. In 2017 IEEE Symposium on Security and Privacy (SP), 3-18 (IEEE, 2017).
  41. 41.Sablayrolles, A., Douze, M., Ollivier, Y., Schmid, C. & Jégou, H. White-box vs black-box: Bayes optimal strategies for membership inference. In Chaudhuri, K. & Salakhutdinov, R. (eds) Proceedings of the 36th International Conference on Machine Learning, {ICML} 97, 5558–5567. http://proceedings.mlr.press/v97/sablayrolles19a.html (PMLR, 2019).
  42. 42.Zhang, C., Bengio, S., Hardt, M., Recht, B. & Vinyals, O. Understanding deep learning requires rethinking generalization. In 5th International Conference on Learning Representations, {ICLR}. https://openreview.net/forum?id=Sy8gdB9xx, (OpenReview.net, 2017).
  43. 43.Carlini, N., Liu, C., Erlingsson, Ú., Kos, J. & Song, D. The secret sharer: evaluating and testing unintended memorization in neural networks. In Heninger, N. & Traynor, P. (eds) 28th {USENIX} Security Symposium ({USENIX} Security 19, 267–284. https://www.usenix.org/conference/usenixsecurity19/presentation/carlini ({USENIX} Association, Santa Clara, CA, USA, 2019).
  44. 44.Abadi, M. et al. Deep learning with differential privacy. In Proceedings of the 2016 ACM SIGSAC Conference on Computer and Communications Security, 308–318 (ACM, 2016).
  45. 45.Shokri, R. & Shmatikov, V. Privacy-preserving deep learning. In Proceedings of the 22nd ACM SIGSAC conference on computer and communications security, 1310–1321 (ACM, 2015).
  46. 46.Langlotz, C. P. et al. A roadmap for foundational research on artificial intelligence in medical imaging: from the 2018 nih/rsna/acr/the academy workshop. Radiology 291, 781–791 (2019).
  47. 47.Kim, Y., Sun, J., Yu, H. & Jiang, X. Federated Tensor Factorization for Computational Phenotyping. In Proceedings of the 23rd {ACM} {SIGKDD} International Conference on Knowledge Discoveryand Data Mining. 887–895. https://doi.org/10.1145/3097983.3098118 (ACM, Halifax, NS, Canada, 2017).
  48. 48.He, C., Annavaram, M. & Avestimehr, S. Fednas: Federated deep learning via neural architecture search. https://sites.google.com/view/cvpr20-nas/ (2020).
  49. 49.Trustworthy federated data analytics (tfda). https://tfda.hmsp.center/ (2020) (Accessed 28 May 2020).
  50. 50.Joint Imaging Platform (Jip). https://jip.dktk.dkfz.de/jiphomepage/ (2020) (Accessed 28 May 2020).
  51. 51.Medical institutions collaborate to improve mammogram assessment ai. https://blogs.nvidia.com/blog/2020/04/15/federated-learning-mammogram-assessment/ (2020) (Accessed 28 May 2020).
  52. 52.Healthchain consortium. https://www.substra.ai/en/healthchain-project (2020) (Accessed 28 May 2020).
  53. 53.The federated tumor segmentation (fets) initiative. https://www.fets.ai (2020) (Accessed 28 May 2020).
  54. 54.Machine learning ledger orchestration for drug discovery. https://cordis.europa.eu/project/id/831472 (2020). Accessed 28 May 2020.
  55. 55.Konečny`, J., McMahan, H. B., Ramage, D. & Richtárik, P. Federated optimization: Distributed machine learning for on-device intelligence. arXiv preprint arXiv:1610.02527 (2016).
  56. 56.Lalitha, A., Kilinc, O. C., Javidi, T. & Koushanfar, F. Peer-to-peer federated learning on graphs. arXiv preprint arXiv:1901.11173 (2019).
  57. 57.Li, T., Sahu, A. K., Zaheer, M., Sanjabi, M., Talwalkar, A. & Smith, V. Federated optimization in heterogeneous networks. arXiv preprint arXiv:1812.06127 (2018).
  58. 58.Zhao, Y. et al. Federated learning with non-iid data. arxivabs/1806.00582 (2018).
  59. 59.Li, X., Huang, K., Yang, W., Wang, S. & Zhang, Z. On the convergence of fedavg on non-IID data. https://openreview.net/forum?id=HJxNAnVtDS (2020).
  60. 60.Wu, B. et al. P3sgd: patient privacy preserving SGD for regularizing deep CNNs in pathological image classification. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (pp. 2099–2108) (2019).
  61. 61.Zhu, L., Liu, Z. & Han, S. Deep leakage from gradients. In Wallach, H. M. et al. (eds) Advances in Neural Information Processing Systems 32: Annual Conference on Neural Information Processing Systems, 14747–14756. http://papers.nips.cc/paper/9617-deep-leakage-from-gradients (2019).
  62. 62.Wang, Z. et al. Beyond inferring class representatives: user-level privacy leakage from federated learning. In 2019 {IEEE} Conferenceon Computer Communications, {INFOCOM} 2512–2520. https://doi.org/10.1109/INFOCOM.2019.8737416 (IEEE, Paris, France, 2019).
  63. 63.Hitaj, B., Ateniese, G. & Perez-Cruz, F. Deep models under the gan: information leakage from collaborative deep learning. In Proceedings of the 2017 ACM SIGSAC Conference on Computer and Communications Security, CCS’17, 603–618 (Association for Computing Machinery, New York, NY, USA, 2017).
  64. 64.Ghorbani, A. & Zou, J. Data shapley: Equitable valuation of data for machine learning. In International Conference on Machine Learning (pp. 2242-2251) (2019).

Citation

MLA
Rieke, N., et al. “The Future of Digital Health with Federated Learning”. Npj Digital Medicine, vol. 3, no. 1, 2020, https://doi.org/10.1038/s41746-020-00323-1.
APA
Rieke, N., Hancox, J., Li, W., Milletarì, F., Roth, H. R., Albarqouni, S., Bakas, S., Galtier, M. N., Landman, B. A., Maier-Hein, K., Ourselin, S., Sheller, M., Summers, R. M., Trask, A., Xu, D., Baust, M., & Cardoso, M. J. (2020). The future of digital health with federated learning. Npj Digital Medicine, 3(1). https://doi.org/10.1038/s41746-020-00323-1
Chicago
Rieke, N., J. Hancox, W. Li, et al. 2020. “The Future of Digital Health with Federated Learning”. Npj Digital Medicine 3 (1). https://doi.org/10.1038/s41746-020-00323-1.
Harvard
Rieke, N. et al. (2020) “The future of digital health with federated learning”, npj Digital Medicine, 3(1). Available at: https://doi.org/10.1038/s41746-020-00323-1.
Vancouver
1. Rieke N, Hancox J, Li W, et al (2020) The future of digital health with federated learning. npj Digital Medicine. https://doi.org/10.1038/s41746-020-00323-1

BibTeX

@article{Rieke_2020, title={The future of digital health with federated learning}, volume={3}, ISSN={2398-6352}, url={http://dx.doi.org/10.1038/s41746-020-00323-1}, DOI={10.1038/s41746-020-00323-1}, number={1}, journal={npj Digital Medicine}, publisher={Springer Science and Business Media LLC}, author={Rieke, Nicola and Hancox, Jonny and Li, Wenqi and Milletarì, Fausto and Roth, Holger R. and Albarqouni, Shadi and Bakas, Spyridon and Galtier, Mathieu N. and Landman, Bennett A. and Maier-Hein, Klaus and Ourselin, Sébastien and Sheller, Micah and Summers, Ronald M. and Trask, Andrew and Xu, Daguang and Baust, Maximilian and Cardoso, M. Jorge}, year={2020}, month=Sept }
Metadata:Crossref

Access the Paper

This paper is available from its original source. Click below to access the PDF.

Open PDF