Implementation of AI in radiology: the perspective of referring physicians

AI offers considerable potential to improve diagnostic accuracy and efficiency in radiology. However, its successful implementation depends largely on the trust and acceptance of referring physicians. This study examines physicians' attitudes toward AI in radiology, identifying key facilitators and barriers to its clinical integration.

A total of 169 licensed physicians in Germany, including surgeons, internists, and general practitioners who frequently refer patients to radiology, were surveyed. Participants were recruited via a systematic review of hospital and practice websites. A structured online questionnaire assessed perceptions of AI, focusing on trust-related factors, preferred applications, and adoption barriers. Statistical analysis was conducted using R and Python.

Overall, 60% of respondents evaluated AI positively for enhancing diagnostic accuracy (mean score 3.7 ± 1.2). The most influential trust factor was model transparency (56.3%), followed by legal clarity on liability (25.0%) and strong data protection (11.7%). Transparency was rated significantly higher than other factors (p < 0.001). Preferred AI applications included lesion detection, research data analysis, and workflow management. Barriers to adoption included the "black box" nature of AI, unclear accountability, and data privacy concerns. Subgroup analysis revealed no significant variation in trust factors between specialties (p = 0.21).

Physicians see AI as a promising tool in radiology but emphasize the need for greater transparency, clear legal responsibility, and secure data handling. Addressing these concerns through explainable AI models, legal frameworks, and robust data protection measures is essential for fostering trust and facilitating successful AI integration in clinical practice.

Understanding physicians' concerns about AI transparency, liability, and data privacy is essential. Addressing these barriers is critical to ensuring responsible implementation, building trust, and enabling effective integration of AI into clinical radiology workflows.

AI acceptance in radiology faces transparency and liability concerns. Lesion detection and data analysis were rated most beneficial by physicians. Clear regulation and explainability are key for clinical AI trust.