Accepted for/Published in: Journal of Medical Internet Research
Date Submitted: Jun 19, 2026
Date Accepted: Sep 9, 2026
Artificial Intelligence Decision Support for Emergency Triage of Large Vessel Occlusion Using Noncontrast Computed Tomography: A Systematic Review and Bayesian Diagnostic Test Accuracy Network Meta-Analysis
ABSTRACT
Background:
Emergency triage of anterior circulation large vessel occlusion (LVO) is constrained by delays in vascular imaging, specialist interpretation, and transfer decision-making. Noncontrast computed tomography (NCCT) is often obtained first in suspected stroke, but visual recognition of LVO on NCCT is difficult outside specialist settings. NCCT-based artificial intelligence (AI) may provide an early human-in-the-loop escalation signal.
Objective:
To compare validated NCCT-based AI with human reader paradigms for anterior circulation LVO and assess whether current evidence supports prospective evaluation of AI-assisted escalation pathways.
Methods:
We searched PubMed, Embase, Web of Science Core Collection, and Cochrane Library from inception to May 24, 2026. Eligible studies evaluated NCCT-based LVO detection in validation datasets independent of model development, included within-study head-to-head comparisons, and provided reconstructible 2-by-2 data. Four nodes were compared: Expert Readers, Non-Expert Readers, Unimodal Imaging AI, and Clinically-Informed Multimodal AI. A Bayesian bivariate hierarchical diagnostic test accuracy network meta-analysis estimated sensitivity, specificity, diagnostic odds ratio (DOR), and absolute differences. Risk of bias was assessed with PROBAST+AI, and certainty was rated with GRADE.
Results:
Ten studies comprising 11 validation datasets, 28 study-node arms, and 3632 patients were included. Unimodal Imaging AI had sensitivity of 0.78 (95% credible interval [CrI], 0.68-0.86), specificity of 0.88 (95% CrI, 0.81-0.94), and DOR of 30.89 (95% CrI, 14.19-60.27). Expert Readers and Non-Expert Readers had lower sensitivity estimates of 0.62 (95% CrI, 0.48-0.75) and 0.60 (95% CrI, 0.45-0.75), respectively, with similar specificity estimates of 0.86. In league-table comparisons, Unimodal Imaging AI showed higher sensitivity than Expert Readers by 0.16 (95% CrI, 0.03-0.29) and Non-Expert Readers by 0.18 (95% CrI, 0.03-0.32), with no clear specificity separation. Clinically-Informed Multimodal AI had sensitivity of 0.81 (95% CrI, 0.64-0.92) and specificity of 0.92 (95% CrI, 0.80-0.98), but this sparse node was connected to human readers only through indirect evidence.
Conclusions:
In retrospective validation cohorts, NCCT-based AI showed higher sensitivity than human reader paradigms while maintaining similar specificity. These findings support prospective evaluation of NCCT-based AI as a bounded human-in-the-loop prompt for expert review, computed tomography angiography prioritization, telestroke consultation, or transfer discussion. The evidence remains accuracy-based and does not establish workflow effectiveness, reperfusion acceleration, or functional outcome benefit. Clinical Trial: PROSPERO CRD420261362111; https://www.crd.york.ac.uk/PROSPERO/view/CRD420261362111.
Citation
Request queued. Please wait while the file is being generated. It may take some time.
Copyright
© The authors. All rights reserved. This is a privileged document currently under peer-review/community review (or an accepted/rejected manuscript). Authors have provided JMIR Publications with an exclusive license to publish this preprint on it's website for review and ahead-of-print citation purposes only. While the final peer-reviewed paper may be licensed under a cc-by license on publication, at this stage authors and publisher expressively prohibit redistribution of this draft paper other than for review purposes.