Artificial intelligence triages chest radiographs: Optimising diagnostic workflow efficiency in hospital settings.
Authors
Affiliations (2)
Affiliations (2)
- Cardiac Catheterisation Laboratory, Greenslopes Private Hospital, Ramsay Health Care, Brisbane, Australia; Queensland University of Technology, School of Clinical Sciences, Faculty of Health, Brisbane, Australia. Electronic address: [email protected].
- Queensland University of Technology, School of Clinical Sciences, Faculty of Health, Brisbane, Australia; Queensland University of Technology, Centre for Biomedical Technologies, Brisbane, Australia; Medical Imaging Department, Redcliffe Hospital, Brisbane, Australia. Electronic address: [email protected].
Abstract
This narrative review assesses the impact of artificial intelligence (AI)-driven chest radiograph (CXR) triage on the efficiency and accuracy of diagnostic workflows in hospital radiology departments. It includes regulatory-approved CXR triage software introduced since 2021 and evaluates key metrics such as sensitivity and report turnaround time in both real-world and simulated clinical environments. Twelve peer-reviewed studies, including single-centre and multicentre investigations across nine countries, were synthesised. AI-driven CXR triage detected selected critical abnormalities and prioritised urgent cases, with reductions in report turnaround time reported across workflow simulations, retrospective evaluations, prospective studies, and one large randomised controlled trial. However, evidence of improved downstream diagnosis or patient outcomes remains limited. Current evidence supports the potential clinical value of AI-driven CXR triage as adjuncts to radiologists and radiographers in hospital settings, demonstrating meaningful improvements in diagnostic efficiency and workflow performance. Continued large-scale, real-world validation, alongside attention to ethical, educational, and integration factors, is needed to support safe and sustainable adoption in radiographic practice. The findings of this review offer valuable insights to inform clinical workflows in hospitals worldwide, particularly those managing substantial CXR caseloads. AI-driven CXR triage should support the interpretation by reporting practitioners. Safe implementation requires local validation, integration with existing workflows, clear governance, staff education, and ongoing monitoring of diagnostic accuracy, reporting efficiency, and downstream patient outcomes. Because faster reporting does not necessarily accelerate diagnosis or treatment, adoption should be supported by coordinated pathway redesign and continued human oversight.