Systematic assessment of the medical utility of radiology and diagnostic Artificial Intelligence in fracture detection (SAMURAI-fracture): a protocol for a multicentre cluster-randomised controlled trial.
Authors
Affiliations (21)
Affiliations (21)
- Oxford Clinical Artificial Intelligence Research (OxCAIR), Oxford, UK [email protected].
- Emergency Medicine Research Oxford (EMROx), Oxford University Hospitals NHS Foundation Trust, Oxford, Oxfordshire, UK.
- Oxford Clinical Artificial Intelligence Research (OxCAIR), Oxford, UK.
- Oxford University Clinical Undergraduate School (OUCAGs), University of Oxford, Oxford, England, UK.
- Radiobotics ApS, Copenhagen, Denmark.
- Emergency Department, Oxford University Hospitals NHS Foundation Trust, Oxford, UK.
- Nuffield Department of Orthopaedics, Rheumatology & Musculoskeletal Sciences, University of Oxford, Oxford, UK.
- Kadoorie Institute for Trauma, Emergency and Critical Care, University of Oxford, Oxford, Oxfordshire, UK.
- University College London Hospitals NHS Foundation Trust, London, England, UK.
- University College London, London, England, UK.
- Health Innovation Oxford and Thames Valley (HIOTV), Oxford, UK.
- PPI Representative, Oxford, UK.
- Royal Berkshire NHS Foundation Trust, Reading, England, UK.
- Buckinghamshire Healthcare NHS Trust, Amersham, England, UK.
- Oxford Health NHS Foundation Trust, Oxford, England, UK.
- Clinical Radiology, Great Ormond Street Hospital for Children, London, UK.
- UCL Great Ormond Street Institute of Child Health, London, UK.
- NIHR Great Ormond Street Hospital Biomedical Research Centre, London, UK.
- Digital Health Validation Lab, University of Glasgow, Glasgow, Scotland, UK.
- Emergency Department, NHS Greater Glasgow and Clyde, Glasgow, Scotland, UK.
- Oxford Biostatistics Ltd, Oxford, UK.
Abstract
Fracture misdiagnosis is a common diagnostic error in emergency departments (EDs) and minor injury units (MIUs), leading to poor patient outcomes, unnecessary treatments and significant healthcare costs. Artificial intelligence (AI)-assisted fracture detection tools are now available for use in radiology workflows; however, the impact of these technologies on patient outcomes, experiences and overall care pathways in the real-world clinical setting is limited. We will conduct a prospective cluster randomised cross-over trial over a 6-month period, assessing the impact of an AI-assisted fracture detection tool in EDs and MIUs across 4 healthcare Trusts in the UK. Patients aged over 2 years old undergoing a plain film radiography for a suspected fracture as part of routine clinical care will be eligible for study enrolment. The trial will deploy Radiobotics' RBfracture, a CE-approved AI-assisted medical device software for fracture detection at each site for 6 months. Randomisation will be at a cluster-level; a site will be randomised to begin with 'AI on' or 'AI off' for a month, followed by alternating active status each month for the remaining 5 months. The primary outcome will evaluate the incidence of 'inappropriate healthcare contacts' among patients receiving imaging for suspected fractures. This composite metric encompasses inappropriate referrals to fracture clinics, repeated hospital attendances and subsequent follow-up communications regarding missed fractures. The rates of these measures will be compared in the 'AI on' versus 'AI off' stages. Secondary outcomes will include patient-reported outcomes, clinician surveys, a predefined health economic evaluation assessing cost-effectiveness and budget impact and the diagnostic performance of the algorithm. The study has received ethical approval from the South Central-Oxford Research Ethics Committee (Reference: 25/SC/0252, approval date: 23 October 2025), and subsequently from the Health Research Authority (IRAS 3 57 391-A). The results of this study will be presented at relevant scientific conferences and peer-reviewed publications will be disseminated on wider public forums such as traditional and social media. This is an ongoing, actively recruiting trial (ISRCTN23087950). Recruitment began in January 2026 and will run for 6 months at each site, with per-patient follow-up of 30 days and subsequent data analysis. This manuscript describes protocol version 1.1. The full protocol and statistical analysis plan are available from the corresponding author on request and will be deposited on the study's public repository (https://github.com/Jason-L-Oke/SAMURAI) prior to publication of results. ISRCTN23087950.