From resistance to reliance: A human-centered analysis of the spectrum of radiologists' trust in AI

Background: Artificial intelligence (AI) is increasingly applied in radiology, yet research has focused mostly on technical implementation, while human factors, particularly radiologists’ trust, which is critical for adoption, remain underexplored. Methods: We conducted semi-structured interviews with 18 radiologists from two hospitals using a guideline covering five domains: user, system, developer, ethical, and patient factors. Interviews allowed participants to elaborate, revisit points, contradict themselves, and introduce unanticipated topics. Thematic analysis with systematic coding identified key themes while preserving contextual nuance. Results: All participants had prior AI experience. Most (61%, 11/18) were optimistic about AI's potential for specific tasks. Accuracy/reliability (94%, 17/18) and time-saving (100%, 18/18) were consistently highlighted as the most critical factors for trust and adoption. Usability, including intuitive interfaces and seamless PACS integration, was emphasized by most (72%, 13/18), while half (50%, 9/18) noted the importance of transparency. Institutional reputation influenced trust in the large majority (89%, 16/18), with preference for non-commercial or reputable entities. Other important factors included clinician involvement in system design (56%, 10/18) and peer-reviewed evidence of performance (61%, 11/18). Ethical considerations included retaining human oversight (83%, 15/18), protecting patient privacy (44%, 8/18), and ensuring institutional data ownership (33%, 6/18). Conclusions: Successful integration of AI in radiology requires attention to radiologists’ perspectives and human factors throughout development, implementation, and use. The diversity and complexity of trust-related factors highlight the importance of human-centered approaches to AI adoption and the need for further research to guide effective implementation.