Radiologists' Trust in AI-Based Systems.
TL;DR
Investigating radiologists' perceptions of AI in clinical practice and examines how these perceptions relate to trust in AI-assisted diagnostic decision-making finds barriers to trust emerge when education, accountability frameworks, and technological maturity do not sufficiently support clinicians' ability to critically engage with AI-assisted radiology tools in healthcare.
Abstract
Artificial intelligence (AI) is increasingly introduced into radiological practice to support image interpretation, workflow optimization, and diagnostic decision-making. However, successful clinical adoption depends not only on technical performance but also on radiologists' trust in these systems. This study investigates radiologists' perceptions of AI in clinical practice and examines how these perceptions relate to trust in AI-assisted diagnostic decision-making. We conducted a cross-sectional online survey among practicing radiologists in Sweden. The survey comprised eight thematic sections addressing AI knowledge and training, attitudes toward AI, responsibility attribution, information and transparency, evaluation and risk perception, professional development, and trust in AI system functionality. Survey responses were analyzed using descriptive statistics, and open-ended responses were analyzed inductively. Fifty-seven radiologists across a range of experience levels and clinical settings participated in the study. While most reported generally positive attitudes towards AI, routine clinical use and formal training were limited. Trust was primarily linked to diagnostic accuracy, empirical validation, transparency, responsibility, and continued human oversight, while concerns focused on data representativeness, system robustness in complex cases, over-reliance on AI, and insufficient transparency. Participants largely assigned responsibility for AI-assisted decisions to AI developers, radiologists, and healthcare organizations rather than to the AI systems themselves. Trust in clinical AI develops through the alignment of technological reliability, professional preparedness, and transparent governance structures. Barriers to trust emerge when education, accountability frameworks, and technological maturity do not sufficiently support clinicians' ability to critically engage with AI-assisted radiology tools in healthcare. Strengthening trust therefore requires clearer definitions of the roles, responsibilities, and competencies expected of radiologists working with AI in clinical practice.