Systematic Assessment of Medical Utility of Radiology Artificial Intelligence - CT Head (SAMURAI-CT)

Background: Non-contrast CT head (NCCTH) is the most commonly requested cross-sectional imaging test in emergency departments (EDs). Rising demand combined with limited radiology capacity has led to widespread delays in reporting, which may affect patient outcomes and prolong ED waiting times. Artificial intelligence (AI) offers a promising solution by helping clinicians identify critical findings and normal studies more quickly, reducing reliance on immediate radiologist input. Annalise Enterprise CTB is a commercially available AI algorithm with high diagnostic performance, shown to improve radiologist accuracy in simulated studies. However, its impact in real-world clinical settings is yet to be established. Objectives: This project aims to implement and evaluate Annalise CTB in four NHS EDs to support clinician interpretation of NCCTH scans. The goal is to reduce patient waiting times and improve operational efficiency by expediting discharge for patients with normal scans and prioritising urgent cases for faster review. Methods: The study will follow a phased, real-world implementation of AI-assisted CT head interpretation. Work packages include: i) retrospective validation to set detection thresholds; ii) algorithm implementation and ED clinician training; iii) a multicentre, prospective, block-randomised trial to evaluate patient pathway impact; iv) stakeholder engagement; v) patient and public involvement; vi) health economic analysis. Primary outcome: time from CT acquisition to discharge for patients with normal scans. Secondary outcomes: report turnaround time for critical cases, proportion of patients discharged within 4 hours, safety metrics (missed findings, recall rate), patient and clinician satisfaction, and cost-effectiveness through improved flow and reduced occupancy. Expected Results: Retrospective validation has demonstrated strong performance in distinguishing urgent from non-urgent cases. We expect a ?20% reduction in time to discharge when AI is used versus standard care. Clinicians will provide feedback on integration challenges and opportunities for improvement. Analysis will be stratified by demographics to explore variation in system performance. Structured surveys will assess acceptability and value from patient and clinician perspectives. Economic modelling will estimate direct and indirect cost savings to support future scale-up. Dissemination and Impact: Findings will be shared through journals, conferences, and social media targeting clinicians, policymakers, and the public. This project will generate real-world evidence for AI integration into emergency imaging workflows. By enabling ED clinicians to interpret NCCTH scans more effectively, we aim to ensure timely intervention for critical cases and quicker discharge for others. This will enhance care quality, reduce ED pressure, and support radiology prioritisation. Outcomes will guide best practices for AI use in emergency radiology and support wider NHS implementation.