A regulator-approved artificial intelligence decision support tool was reported to influence radiology workflow when deployed in routine clinical practice, but its real-world impact differed from early expectations, according to a new study published in the Journal of Medical Internet Research.
Researchers at Queensland University of Technology in Australia examined the implementation of an artificial intelligence system designed to flag potential findings on computed tomography imaging studies at a large tertiary referral hospital in Brisbane. Rather than evaluating diagnostic accuracy under controlled conditions, the investigators conducted qualitative interviews across preimplementation, implementation, and postimplementation phases to explore adoption patterns, clinician experiences, and organizational factors shaping use.
The artificial intelligence system was integrated into clinical operations so that computed tomography images were transferred to the platform, which flagged or highlighted potential findings for radiologists to review. Clinicians reported that the tool had the potential to help draw attention to certain cases, particularly during periods of high workload. However, engagement with the system varied across users and clinical contexts.
Adoption was not uniform. Some radiologists consistently interacted with the tool’s outputs, while others used it less frequently. Participants described “information overload,” inconsistent performance, and uncertainty about responsibility and medicolegal liability as barriers to sustained engagement. Interoperability challenges and workflow disruptions — including additional steps required during reporting — also influenced how and when the tool was used.
The authors report that regulatory approval and prior technical validation did not prevent implementation challenges. Interview data highlighted uncertainty about accountability, the need for clearer communication regarding system limitations, and ongoing adjustments to workflow integration.
The authors describe implementation as an ongoing process rather than a one-time installation. Interview findings highlighted the need for continued evaluation, clearer governance structures, and sustained engagement from radiologists during deployment.
The authors declared having no competing interests.
