A review on federated learning in computational pathology.

Area of Science:

• Computational Pathology
• Machine Learning
• Data Privacy

Background:

• Generalizable computational pathology (CPATH) algorithms require large, multi-institutional datasets.
• Healthcare data privacy regulations impede the creation of such datasets.
• Federated Learning (FL) offers a solution by enabling collaborative model training while preserving data privacy.

Purpose of the Study:

• To review key developments in Federated Learning (FL) for computational pathology (CPATH) applications.
• To evaluate the current status and performance of FL in CPATH.
• To identify challenges and propose steps for broader adoption of FL in CPATH.

Main Methods:

• Systematic review of 15 studies on FL for CPATH.
• Evaluation of different FL approaches, including model aggregation, domain alignment, and privacy preservation methods.
• Comparison of federated and centralized training performance.

Main Results:

• Proof-of-concept studies demonstrate FL models achieve performance equivalent to centralized models.
• Model alignment methods show significant performance improvements ( $$ ).
• Privacy preservation methods maintain performance with only slight degradation ( $$ lower).

Conclusions:

• FL is a viable approach for training CPATH algorithms, maintaining data privacy.
• Advancements in model and domain alignment are crucial for maximizing FL performance.
• Standardized frameworks and guidelines are needed for real-world FL adoption in CPATH.