Recent methodological advances in federated learning for healthcare

Area of Science:

• Medical Informatics
• Machine Learning
• Data Science

Background:

• Healthcare data is often siloed due to ethical and privacy concerns, preventing traditional centralized machine learning.
• Federated learning (FL) offers a solution by enabling model training across distributed datasets without data aggregation.
• Healthcare data presents unique challenges including class imbalance, missing values, and non-standardized variables, complicating FL implementation.

Purpose of the Study:

• To systematically review and analyze novel federated learning methodologies developed for healthcare data challenges.
• To identify systemic issues and limitations in current federated learning approaches for medical applications.
• To provide recommendations for enhancing future federated learning methodology development in healthcare.

Main Methods:

• Systematic literature review of Scopus-indexed papers published between January 2015 and February 2023.
• Focused on papers describing new federated learning methodologies addressing healthcare data complexities.
• Reviewed 89 papers that met the inclusion criteria.

Main Results:

• Identified significant systemic issues across many reviewed federated learning methodologies.
• Found that common healthcare data challenges (silos, imbalance, missing data, distribution shifts) are not adequately addressed by existing FL methods.
• Highlighted the methodological complexity introduced by federated learning, including distributed optimization and model aggregation.

Conclusions:

• Current federated learning methodologies for healthcare data suffer from significant systemic flaws.
• There is a critical need for improved and robust federated learning methods tailored to the unique challenges of medical data.
• Recommendations are provided to guide future research and development in this domain.