Beyond the algorithm: embedding ethics for trustworthy AI in radiology and oncology

Area of Science:

• Medical Imaging and Oncology
• Artificial Intelligence Ethics
• Sociotechnical Systems

Background:

• AI in radiology and oncology offers diagnostic and efficiency gains but poses ethical and societal challenges.
• Current governance relies on high-level principles like fairness, which lack contextual grounding.
• This study examines ethical and societal AI aspects within the EuCanImage project.

Purpose of the Study:

• To explore the emergence of ethical concerns in real-world AI applications in healthcare.
• To understand how institutional, clinical, and sociotechnical dynamics shape AI ethics.
• To propose a pathway for developing trustworthy AI in cancer care.

Main Methods:

• Multi-method empirical study including literature reviews, interviews, and workshops.
• Involved diverse stakeholders: AI developers, clinicians, and others.
• Explored real-world ethical issues and their contextual influences.

Main Results:

• Ongoing interdisciplinary involvement is crucial for addressing explainability, accountability, bias, and social impact in radiological AI.
• Four dimensions of trustworthy AI (explainability, trust, accountability, fairness) are difficult to operationalize without procedural guidance.
• Ethical issues are not solely technical or abstract; trustworthiness is relational and co-constructed.

Conclusions:

• A structured, multi-stakeholder AI development pathway is proposed.
• This pathway shifts from principle-driven ethics to interdisciplinary approaches grounded in clinical realities.
• Strengthened stakeholder engagement is key for trustworthy AI in cancer care.