Purpose or Learning objective: This study examines whether established population‐screening principles suffice to assess the ethical and societal acceptability of AI-enabled screening programs, and proposes actionable, radiology-oriented norms to support responsible design, evaluation, and deployment of AI systems for screening.
Methods or Background: We conducted a conceptual and normative analysis of classic screening criteria in light of AI’s specific affordances and potential impacts on screening workflows. We review the emergence and evolution of population screening principles, outline AI’s potential in population screening, and consider ESR recommendations alongside the EU legislative landscape. Drawing on the WHO’s 2021 guiding principles for governing AI in healthcare and the notion of “techno-moral change”, we examine where automation, multi-disease detection, and data-driven personalization create ethical blind spots. From this, we develop a set of recommendations in the form of additional criteria.
Results or Findings: Established screening criteria do not fully address the concerns AI-enabled screening may raise. In particular, the enhanced detection capabilities and the impacts of automation justify an extension. We propose additional norms for assessing AI-enabled screening across four domains: I. appropriateness of using AI in a specific context, II. AI system requirements, III. conditions for trustworthy automation and IV. requirements for handling additional findings and personalized approaches. Our framework emphasizes the need for demonstrable benefits of using AI over traditional methods, evidence-based analysis of benefits and risks, acceptable levels of fairness and interpretability of algorithmic predictions, clear accountability mechanisms, safe and transparent use of AI, and respect for patient autonomy.
Conclusion: Together with previous principles, these additional criteria will facilitate a comprehensive evaluation of AI-enabled screening and, hence, its responsible implementation.
Limitations: Normative synthesis without empirical validation; applicability may vary by modality, setting, and jurisdiction; evolving regulation and evidence may require periodic revision.