Privacy-Preserving Artificial Intelligence Techniques in Biomedicine

Area of Science:

• Biomedicine
• Computer Science
• Genomics

Background:

• Artificial intelligence (AI) demonstrates significant potential in biomedical applications, including next-generation sequencing data analysis and clinical decision support systems.
• The use of sensitive biomedical data for AI training raises substantial privacy concerns for individual participants.

Purpose of the Study:

• To provide a comprehensive overview of privacy-preserving AI techniques applicable to the biomedical field.
• To categorize and analyze state-of-the-art approaches, highlighting their respective advantages, disadvantages, and areas for future research.

Main Methods:

• A structured review of recent advancements in privacy-preserving AI for biomedicine.
• Classification of existing methods within a unified taxonomy.
• Discussion of the strengths, limitations, and open challenges of these techniques.

Main Results:

• Identified privacy risks associated with training AI on sensitive data, such as genomic information.
• Highlighted how data access restrictions due to privacy concerns hinder collaborative research and scientific progress.

Conclusions:

• Federated machine learning, combined with other privacy-enhancing techniques, represents a promising direction for scalable, privacy-preserving AI in distributed biomedical applications.
• Further research is needed to address the challenges, including computational and network overhead, associated with hybrid privacy-preserving AI approaches.