Grounding Radiological AI Ethics: Practice-Centered Ethical Guidelines from a Multi-Stakeholder Perspective
TL;DR
This paper presents an empirically grounded study in radiology in Germany, conducted in collaboration with multiple stakeholders, such as ethicists, AI developers, clinicians, and HCI researchers via an ethical, legal, and social implications (ELSI) workshop and in-depth interviews, and proposes ten practice-centered ethical guidelines for Human-Centered AI (HCAI) in radiology.
Abstract
The use of artificial intelligence (AI) in healthcare promises improved diagnosis, treatment, and care, but raises ethical challenges around accountability, privacy, transparency, data governance, trust, and explainability. Though the body of research on ethical considerations to mitigate these concerns has been growing considerably, such considerations are often insufficiently grounded in real-world practice. This paper presents an empirically grounded study in radiology in Germany, conducted in collaboration with multiple stakeholders, such as ethicists, AI developers, clinicians, and HCI researchers via an ethical, legal, and social implications (ELSI) workshop and in-depth interviews. Our thematic analysis identifies key ethical concerns for AI-supported radiology, including workflow-dependent integration, multi-layered explainability for different stakeholders, data governance and privacy-performance trade-offs, ambiguous responsibility allocation, and risks of automation bias and deskilling. Building on these insights, we propose ten practice-centered ethical guidelines for Human-Centered AI (HCAI) in radiology that translate high-level ethical principles into concrete actions for HCAI design and implementation. This contribution sets out to advance the literature by bridging conceptual ethical knowledge and the practical design of HCAI systems in radiology.
