Collaborative Intelligence in Oncology: A Review of Federated Learning Models for Cancer Detection

Cancer remains a major global health burden, and early detection is critical for reducing mortality. Conventional machine learning often depends on centralized data, which raises privacy and data-sharing concerns. Federated learning (FL) offers a solution by enabling collaborative model training across institutions without sharing raw data, safeguarding privacy while improving generalizability. The contribution of this review is to systematically analyze FL applications in cancer detection, highlighting their strengths, limitations, and future directions for clinical translation. The review was conducted through a structured search of Scopus, focusing on peer-reviewed, openaccess articles in English. A total of 42 studies were included, spanning publications from 2020-2024. Eligible studies were screened based on relevance to oncology and FL, with data extraction covering cancer type, data source, FL methods, models, data types, performance metrics, key findings, and research gaps. Results show that FL has been applied to a wide range of cancer types including breast, lung, prostate, colorectal, brain, cervical, melanoma, and multi-cancer datasets. Data sources involve both multi-center collaborations and public datasets, while methods include horizontal, vertical, hybrid, and variants such as FedAvg, FedProx, and transfer learning. Models range from CNNs, ResNets, and UNet derivatives to transformers and ensembles. Reported metrics indicate high performance comparable to centralized learning. Key findings highlight privacy preservation and robust generalization, but research gaps remain in dataset size, heterogeneity, validation, computational cost, and interpretability. In conclusion, FL shows strong promise for collaborative cancer detection, yet future studies must address scalability, data diversity, and transparency to support realworld clinical adoption.