AI Model Passport: Data and system traceability framework for transparent AI in health

Area of Science:

• Medical Informatics
• Artificial Intelligence
• Health Systems Engineering

Background:

• Current AI frameworks in healthcare lack scalability and machine interpretability due to manual documentation.
• Existing methods fail to provide verifiable AI model identities, hindering reproducibility and trust.
• There is a critical need for standardized, verifiable documentation for AI in biomedical applications.

Purpose of the Study:

• To introduce the AI Model Passport, a structured framework for AI model digital identity and verification.
• To address limitations in transparency, accountability, and reproducibility of AI in health systems.
• To establish a new standard for trustworthy and regulation-compliant AI in healthcare.

Main Methods:

• Concept introduction of the AI Model Passport for capturing essential AI model lifecycle metadata.
• Development and implementation of AIPassport, an MLOps tool for automated metadata collection and versioning.
• Demonstration using a medical imaging use case (lesion segmentation) within the ProCAncer-I project.

Main Results:

• The AI Model Passport provides unique identification, verification, and traceability for AI models.
• AIPassport tool automates metadata collection, versioning, and integrates with development environments.
• The framework demonstrated enhanced transparency, reproducibility, and regulatory readiness in a medical imaging use case.

Conclusions:

• The AI Model Passport offers a scalable and verifiable solution for AI documentation in healthcare.
• This framework promotes trust and accountability in AI-driven healthcare solutions.
• The AI Model Passport serves as a foundation for transparent and regulation-compliant AI systems across domains.