Advancing breast, lung and prostate cancer research with federated learning. A systematic review

Anshu Ankolekar, Sebastian Boie, Maryam Abdollahyan, Emanuela Gadaleta, Seyed Alireza Hasheminasab, Guang Yang, Charles Beauville, Nikolaos Dikaios, George A. Kastis, Michael Bussmann, Claude Chelala, Sara Khalid, Hagen Kruger, Philippe Lambin, Giorgos Papanastasiou

Introduction

Modern oncology AI relies on massive, diverse datasets, yet most medical data remains trapped in institutional silos. Privacy regulations like GDPR and the technical overhead of moving large-scale imaging data often prevent the creation of centralized "data lakes." Consequently, many clinical models suffer from overfitting, performing well at their home institution but failing in external deployments. Federated Learning (FL) offers a solution by allowing models to train on decentralized data without moving sensitive records. This systematic review evaluates the impact of FL in breast, lung, and prostate cancer research between 2020 and 2023.

Methodology and Technical Landscape

The authors analyzed 25 peer-reviewed papers, evaluating "FL rigour" by comparing FL performance against centralized machine learning (ML) baselines. The technical landscape was categorized by architecture (Horizontal vs. Vertical FL), model types, and technical tasks.

Distribution of ML model types and technical tasks in the reviewed literature (2021-2023).

As shown in the figure above, classification remains the dominant technical task, while model architectures have evolved from classic ML to more complex structures like UNets and large pre-trained models. The review also noted a lack of transparency in many papers regarding specific aggregation strategies and privacy-preserving techniques, which can hinder reproducibility.

Clinical Applications and FL Scope

The review highlights a shift from simple classification to more complex clinical needs. FL is being leveraged for tumor identification, disease differentiation, and severity assessment. Beyond specific tasks, the primary motivations for adopting FL include improving model generalizability and ensuring data privacy.

Breakdown of clinical applications and the primary scope of federated learning implementations.

Key Insights and Results

The review challenges the notion that FL requires a trade-off in accuracy for privacy. In 60% of the studies (15 out of 25), FL models actually outperformed centralized baselines. This "performance paradox" suggests that the increased data diversity accessible through FL leads to more robust and generalizable models than those trained on single-center silos.

Key findings include:

  • Superiority: 18 out of 25 studies showed FL performance was comparable or superior to centralized ML.
  • Precision Medicine: FL enables multi-modal integration (imaging, EHR, genomics) across institutions, facilitating holistic cancer analysis.
  • Scalability: One standout lung cancer study achieved a staggering 99.69% accuracy by combining FL with blockchain for secure classification.
  • Generalizability: FL models consistently maintained higher performance when tested on "unseen" data from new institutions compared to locally trained models.

Challenges and Future Directions

Despite its promise, the field faces several hurdles:

  • Reproducibility: Only 44% of rigorous studies provided accessible code, hindering the ability to build on existing work.
  • Data Heterogeneity: "Client drift" due to varying imaging protocols or demographics remains a challenge. Advanced aggregation methods like FedProx or FedNova are recommended to handle non-IID (not identically and independently distributed) data.
  • Standardization: There is a critical need for consistent evaluation metrics and benchmark datasets tailored for federated contexts.

Future research is expected to focus on how large, pre-trained foundation models can be fine-tuned via FL to further enhance clinical utility in oncology.

Summary updated

Abstract

Federated Learning (FL) has emerged as a promising solution to address the limitations of centralised machine learning (ML) in oncology, particularly in overcoming privacy concerns and harnessing the power of diverse, multi-center data. This systematic review synthesises current knowledge on the state-of-the-art FL in oncology, focusing on breast, lung, and prostate cancer. Distinct from previous surveys, our comprehensive review critically evaluates the real-world implementation and impact of FL on cancer care, demonstrating its effectiveness in enhancing ML generalisability, performance and data privacy in clinical settings and data. We evaluated state-of-the-art advances in FL, demonstrating its growing adoption amid tightening data privacy regulations. FL outperformed centralised ML in 15 out of the 25 studies reviewed, spanning diverse ML models and clinical applications, and facilitating integration of multi-modal information for precision medicine. Despite the current challenges identified in reproducibility, standardisation and methodology across studies, the demonstrable benefits of FL in harnessing real-world data and addressing clinical needs highlight its significant potential for advancing cancer research. We propose that future research should focus on addressing these limitations and investigating further advanced FL methods, to fully harness data diversity and realise the transformative power of cutting-edge FL in cancer care.
BibTeX
@article{ankolekar2025advancing,
  author = {Anshu Ankolekar and Sebastian Boie and Maryam Abdollahyan and Emanuela Gadaleta and Seyed Alireza Hasheminasab and Guang Yang and Charles Beauville and Nikolaos Dikaios and George A. Kastis and Michael Bussmann and Claude Chelala and Sara Khalid and Hagen Kruger and Philippe Lambin and Giorgos Papanastasiou},
  title = {{Advancing breast, lung and prostate cancer research with federated learning. A systematic review}},
  journal = {Npj Digit. Medicine},
  year = {2025},
  doi = {10.1038/s41746-025-01591-5},
}
Figure