Swarm Learning for decentralized and confidential clinical machine learning.

Area of Science:

• Artificial Intelligence in Medicine
• Bioinformatics
• Data Privacy

Background:

• Precision medicine aims for rapid, reliable detection of complex diseases.
• Machine learning on blood transcriptomes can identify conditions like leukemia.
• Privacy legislation poses challenges for integrating diverse medical data.

Purpose of the Study:

• Introduce Swarm Learning, a decentralized machine learning approach.
• Facilitate global medical data integration while upholding privacy laws.
• Demonstrate Swarm Learning's feasibility for developing disease classifiers.

Main Methods:

• Swarm Learning combines edge computing, blockchain, and peer-to-peer networking.
• A decentralized approach that maintains confidentiality without a central coordinator.
• Applied to four heterogeneous diseases: COVID-19, tuberculosis, leukemia, and lung pathologies.

Main Results:

• Utilized over 16,400 blood transcriptomes and 95,000 chest X-ray images from 127 studies.
• Swarm Learning classifiers demonstrated superior performance compared to site-specific models.
• The approach inherently complies with local data confidentiality regulations.

Conclusions:

• Swarm Learning offers a privacy-preserving solution for distributed medical data analysis.
• This decentralized method accelerates the advancement of precision medicine.
• Enables robust disease classification across diverse datasets and patient populations.