Federated Learning for Histopathology Image Classification: A Systematic Review.

Area of Science:

• Digital Pathology
• Machine Learning
• Medical Imaging

Background:

• Machine learning (ML) and deep learning (DL) improve histopathology image classification but face data privacy issues.
• Federated learning (FL) allows collaborative model training without data sharing, addressing privacy concerns.
• FL is crucial for developing robust DL models in sensitive medical domains.

Purpose of the Study:

• To systematically review FL applications in histopathological image classification.
• To identify common methodologies, datasets, and performance metrics used in FL for digital pathology.
• To highlight challenges and future research directions for FL in this field.

Main Methods:

• Systematic review following PRISMA guidelines.
• Analysis of 24 studies published between 2020-2025 from major scientific databases.
• Inclusion criteria focused on FL-based DL models for histopathology with reported performance.

Main Results:

• 10 datasets were used, with accuracies ranging from 69.37% to 99.72%.
• FedAvg was the most common aggregation algorithm; VGG, ResNet, DenseNet, and EfficientNet were frequent architectures.
• Key challenges include communication overhead, computational demands, and inconsistent reporting.

Conclusions:

• Federated learning shows significant potential for privacy-preserving digital pathology.
• Standardized evaluation, efficient aggregation, personalization, and interpretability are crucial for clinical adoption.
• Multi-institutional validation is needed to realize FL's full benefits in histopathology.