RADIOLOGISTS’ trust in artificial intelligence (AI) depends on accuracy, efficiency, and the practicalities of clinical use, according to a qualitative study exploring clinicians’ perspectives.
The research found that willingness to adopt AI was shaped by a broad range of considerations, including usability, transparency, and confidence in developers. Although most participants expressed optimism about AI, their expectations frequently concerned its future potential rather than the capabilities of currently available systems.
Researchers conducted semi-structured interviews with 18 radiologists recruited from hospitals in the Netherlands and Switzerland. All participants had experience using AI in clinical practice or research, including applications supporting lung nodule assessment, mammography, and prostate MRI.
The interviews explored five domains: user, system, developer, ethical, and patient factors. Researchers used thematic analysis to identify recurring considerations while retaining the context behind participants’ views.
Overall, 11 participants, approximately 61%, described a positive attitude towards AI, while five reported mixed views. These mixed perspectives commonly combined scepticism about existing tools with optimism about future developments.
Accuracy and reliability were highlighted by 17 of the 18 participants. Radiologists valued consistent performance across different cases, scanners, and imaging protocols, alongside the ability to minimise missed findings and unnecessary alerts.
All participants identified execution time and potential time savings as important considerations. The findings suggest that even an accurate system could encounter resistance if it introduces delays or additional work.
Usability was emphasised by 13 participants, approximately 72%, including intuitive interfaces and integration with picture archiving and communication systems. Half highlighted transparency, although the researchers noted that access to technical explanations did not necessarily mean clinicians could readily understand them.
Developer reputation influenced trust amongst 16 participants, approximately 89%. Clinician involvement in system design was mentioned by 10, while 11 highlighted peer-reviewed evidence of performance.
Human oversight remained a prominent consideration, identified by 15 participants, approximately 83%. Radiologists emphasised their role in interpreting findings within the wider clinical picture and reviewing AI outputs before these inform patient care.
Patient privacy and institutional control over data were also raised. The researchers noted a potential tension between maintaining careful oversight and managing workloads that could encourage less critical review.
The small sample, recruitment from two European hospitals, and participants’ previous AI experience limit the transferability of the findings. Reported percentages describe themes mentioned during interviews and do not establish their prevalence across the wider radiology workforce.
Nevertheless, the study highlights the importance of considering clinicians’ perspectives throughout AI development and implementation. Further research could clarify how trust changes with experience and across different healthcare settings.
Reference
Chupetlovska K et al. From resistance to reliance: A human-centered analysis of the spectrum of radiologists’ trust in AI. Eur J Radiol Open. 2026;17:100780.
Featured image: DragonImages on AdobeStock
Each article is made available under the terms of the Creative Commons Attribution-Non Commercial 4.0 License.
