Federated learning in computational pathology: a literature review.

Area of Science:

• Artificial intelligence in healthcare
• Computational pathology and medical imaging
• Decentralized machine learning techniques

Background:

• Centralized AI models risk patient privacy due to data aggregation.
• Federated learning (FL) offers a privacy-preserving alternative for collaborative model training.
• FL is particularly relevant for sensitive healthcare data and computational pathology.

Purpose of the Study:

• To systematically review the state-of-the-art applications of FL in healthcare.
• To focus on FL's role in computational pathology and medical imaging.
• To identify motivations, strategies, challenges, and solutions in FL for healthcare.

Main Methods:

• Systematic literature review of FL in healthcare, emphasizing computational pathology.
• Analysis of studies using diverse medical imaging modalities (histopathology, MRI, CT, PET).
• Categorization of studies by FL architecture, motivations, implementation, and challenges.

Main Results:

• Growing adoption of FL in healthcare, with increasing use in computational pathology.
• FL models achieve comparable or superior accuracy to centralized models while preserving privacy.
• Federated training is feasible for computational pathology tasks like prediction and segmentation.

Conclusions:

• FL shows significant potential for secure, privacy-preserving healthcare collaboration, especially in computational pathology.
• Key challenges include data heterogeneity, system interoperability, and model interpretability.
• Future research needs to focus on standardization, robustness, and ethical considerations for clinical integration.