Federated Learning: collaboration without compromise
The chapter on Federated Learning shows how this technology enables machine learning models to be trained on distributed data, without such data ever being shared or centralized. In healthcare, this means that hospitals, laboratories, and research centers can collaborate in developing predictive and diagnostic models while keeping sensitive patient data within their own infrastructures.
FL is structured into two main modes: cross-device (on personal devices) and cross-silo (between trusted institutions such as hospitals and research centers). It can also be horizontal (similar data on different users) or vertical (different data on the same user). Both configurations present technical challenges but also offer significant advantages in terms of privacy, security, scalability, and regulatory compliance, particularly relevant in regulated domains such as GDPR.
Concrete applications in healthcare and outcomes
The report illustrates numerous use cases of FL in healthcare, demonstrating how it is already a technology capable of concretely improving clinical practice and research:
• In oncology, federated models have improved brain tumor segmentation in international multicenter studies.
• In radiology, the EXAM project employed FL to predict oxygen needs in COVID-19 patients, achieving an accuracy above 92%.
• In intensive care, FL enables the development of risk scores for sepsis or organ failure without sharing sensitive data.
• In genomics and rare diseases, FL allows data to be aggregated from multiple centers to identify genetic patterns otherwise invisible, accelerating diagnosis and the development of targeted therapies.
A vision for the future of healthcare
Intesa Sanpaolo Innovation Center identifies Federated Learning as a turning point in several domains, especially in healthcare, within a context characterized by an aging population and growing demand for personalized treatments. The ability to collaborate on a large scale without compromising data privacy opens new avenues for AI-driven innovation, healthcare system sustainability, and equity in access to care even when data availability is limited, as in the case of rare diseases.
