JMIR Publications

ABSTRACT

Background:

Medical imaging remains at the forefront of advancements in adopting digital health technologies in clinical practice. Regulatory approved artificial intelligence (AI) clinical decision support systems are commercially available and being embedded into routine practices for radiologists internationally. These decision support solutions show promising clinical validity compared to standard practice conditions; however, their implementation in practice is poorly understood.

Objective:

The study presents findings from an end-to-end qualitative implementation-evaluation of an AI clinical decision support tool within a large hospital medical-imaging department.

Methods:

This prospective implementation-evaluation study was conducted in a large public tertiary referral hospital in Brisbane, Australia. One-to-one participant interviews were conducted across three implementation phases: pre-implementation, peri-implementation and post-implementation. Participants comprised radiology consultants and registrars in addition to radiographers. Eligibility criteria included involvement in chest computed tomography (CT) studies during the study timeframe. Interviews were informed by the Non-Adoption, Abandonment, Scale-up, Spread, and Sustainability (NASSS) framework.

Results:

In total, 43 one on one interviews were conducted across the 2-year study period. This consisted of 6 radiographers (14%), 21 registrar radiologists (49%) and 16 consultant radiologists (37%). Nine (21%) participants were interviewed across multiple timepoints. Perceptions of, and level of engagement with, the AI decision support solution were mixed during real-world image interpretation and reporting in practice. Responses highlighted the importance of addressing the sociotechnical conditions of implementation through early stakeholder involvement, consistent communication, agile training models, workflow-compatible design, and mechanisms for feedback and iterative refinement. Moreover, findings highlight long-term sustainability of such technologies depends on institutional commitment through intentional resourcing and implementation planning.

Conclusions:

The success, or otherwise, of AI clinical decision support solutions in real-world practice is dependent on many ongoing non-technical factors. Co-design principles that place clinical users at the centre of system development and integration remain as important as ever as we progress into the era of AI-based clinical decision support.