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

Abstract

Federated learning (FL) is advancing cancer research by enabling privacy-preserving collaborative training of machine learning (ML) models on diverse, multi-centre data. This systematic review synthesises current knowledge on state-of-the-art FL in oncology, focusing on breast, lung, and prostate cancer. Unlike previous surveys, we critically evaluate FL's real-world implementation and impact, demonstrating its effectiveness in enhancing ML generalisability and performance in clinical settings. Our analysis reveals that FL outperformed centralised ML in 15 out of 25 studies, spanning diverse models and clinical applications, including multi-modal integration for precision medicine. Despite challenges identified in reproducibility and standardisation, FL demonstrates substantial potential for advancing cancer research. We propose future research focus on addressing these limitations and investigating advanced FL methods to fully harness data diversity and realise the transformative power of cutting-edge FL in cancer care.

Fig. 1. Vertical bar graphs showing research trends for all categories of machine learning (ML) models and technical tasks, identified.

Publication record over time for all ML types (a) and technical tasks (b) identified. In a, “Other” included recurrent neural network, capsule neural network and region-based CNN.

Fig. 2. Vertical bar graphs showing research trends for all categories of clinical applications and FL scopes, identified.

Publication record over time for all clinical applications addressed (a) and FL scopes (b) identified. In a, “Other” included side effect prediction (1) and tumour recurrence assessment (1). In b, “Other” involved domain adaptation (1) and training time reduction (1). NM: not mentioned; FL: federated learning.

Fig. 3. Sankey diagram depicting relationships (combinations) between the following technical aspects extracted across studies (represented as nodes): data, central ML model, technical task addressed and FL method.

The width of each flow is proportional to the quantity being represented: thicker width corresponds to a higher combination prevalence across the reviewed papers, and vice versa. EHR: electronic health records; CT: computed tomography; MRI: magnetic resonance imaging; PET-CT: hybrid positron emission tomography-computed tomography; WSI: whole slide imaging; ML: machine learning; FL: federated learning; NM: not mentioned.

Fig. 4. Sankey diagram illustrating combinations between the following application aspects extracted across studies (represented as nodes): data, clinical application, FL scope and organ area.

EHR: electronic health records; CT: computed tomography; MRI: magnetic resonance imaging; PET-CT: hybrid positron emission tomography-computed tomography; WSI: whole slide imaging; FL: federated learning; NM: not mentioned.

Fig. 5. PRISMA flow of the systematic review process.

The flow presents inclusion and exclusion of papers at each review stage.