Beyond regulatory compliance: evaluating radiology artificial intelligence applications in deployment.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Clinical Governance

Background:

• Regulatory approval for AI applications does not fully address practical concerns like reliability, accountability, safety, and governance.
• Lack of professional confidence and perceived need for enhanced validation hinder AI adoption in routine clinical practice.
• Existing validation and compliance measures may be insufficient for novel and relatively untested AI technologies.

Purpose of the Study:

• To propose an approach for validating artificial intelligence (AI) applications beyond standard regulatory compliance.
• To enhance trust and prevent harm associated with AI implementation in healthcare settings.
• To provide a framework for rigorous evaluation of AI tools in both laboratory and clinical practice.

Main Methods:

• Independent benchmarking of AI applications in a laboratory setting.
• Conducting clinical audits of AI applications in real-world practice.
• Implementing a validation strategy that exceeds baseline regulatory requirements.

Main Results:

• The proposed approach aims to increase professional confidence in AI technologies.
• Enhanced validation methods are expected to improve the safety and reliability of AI tools.
• The strategy focuses on preventing harm through comprehensive evaluation.

Conclusions:

• Standard regulatory compliance is insufficient for ensuring the safe and effective implementation of AI in healthcare.
• Independent benchmarking and clinical audits are crucial for building trust in AI applications.
• A proactive approach to validation is necessary to overcome barriers to AI adoption and ensure patient safety.