Dependable and Trustworthy Artificial Intelligence in Clinical Practice - Does It Exist?
Description
Artificial intelligence (AI) is moving rapidly into medical imaging workflows, yet questions remain about whether these tools can truly be considered dependable and trustworthy in clinical practice. This session will address three critical dimensions: technical reliability and lifecycle monitoring; interpretability and the calibration of trust between human experts and AI systems; and economic and workflow impact, including both the potential gains and the pitfalls in measuring return on investment. By bringing together diverse perspectives, the session will explore not only where AI is delivering value but also where caution is warranted, providing attendees with practical insights into the promises and limits of AI in patient care. The session will review methods for evaluating AI systems as part of clinical commissioning. Proper commissioning and site-specific validation are key aspects of building trust in real-world clinically deployed systems. Appropriate volumes and types of test cases, metrics, best practices for documentation, and practical aspects of implementation will be discussed with emphasis on empowering attendees to gauge the appropriate dependability of AI systems in their own environments.