Assessment of Performance, Interpretability, and Explainability in Artificial Intelligence-Based Health Technologies:...

Area of Science:

• Medical device development
• Artificial intelligence in healthcare
• Health technology assessment

Background:

• Development of AI-based medical devices (MDs) requires robust assessment frameworks.
• Existing guidelines for health technology assessment (HTA) need specific criteria for AI-MDs.
• Confidence in AI-based MDs hinges on understanding algorithm behavior.

Purpose of the Study:

• To identify essential concepts for AI-based MD development.
• To analyze assessment methodologies for AI-MDs, focusing on performance, interpretability, and explainability.
• To provide recommendations and decision support for AI-MD development and regulatory assessment.

Main Methods:

• Literature review to identify key HTA criteria for AI-MDs.
• Analysis of existing assessment methodologies for selected criteria.
• Scoping review of HTA agency guidelines.

Main Results:

• Performance, interpretability, and explainability are key criteria for AI-MD confidence.
• Recommendations provided for evaluating AI-MD performance based on model structure and data.
• Methods described to support the evaluation of interpretability and explainability.
• A decision support flowchart for regulatory requirements is proposed.

Conclusions:

• Increasing emphasis on explainability and interpretability in HTA for AI-MD accountability.
• Proposed tools and methods aid in understanding AI algorithm predictions and decision-making processes.
• Clear criteria and assessment methods are vital for the safe and effective development of AI-based MDs.