Clinical Report: Automated Identification of Hyperdense Artery Sign on Non-Contrast CT

Overview

This study presents a fully automated deep-learning model for detecting the hyperdense artery sign (HAS) on non-contrast CT, demonstrating high reliability in identifying large vessel occlusion (LVO). The model achieved significant sensitivity and specificity across two validation cohorts, supporting its role as an adjunctive pre-CTA alert to enhance workflow readiness.

Background

Large vessel occlusion (LVO) is a critical condition in acute ischemic stroke, accounting for approximately 30% of cases. Timely detection and intervention are essential for improving patient outcomes, yet traditional imaging methods like CTA can introduce delays. The hyperdense artery sign (HAS) on non-contrast CT offers a potential immediate marker for LVO, which can be enhanced through automated detection methods.

Data Highlights

Key Findings

• The automated model achieved a sensitivity of 76.2% and specificity of 87.0% in a high-acuity cohort.
• In a broader real-world population, the model maintained a sensitivity of 74.3% and specificity of 82.7%.
• The positive predictive value (PPV) was 92.0% in the high-acuity cohort, indicating high reliability of alerts.
• Model assistance improved human reader performance in detecting HAS, increasing the JAFROC Figure of Merit from 0.71 to 0.77.
• The model serves as an adjunctive pre-CTA alert, facilitating earlier workflow readiness in stroke management.

Clinical Implications

The findings suggest that integrating automated HAS detection into clinical workflows can expedite the identification of LVO, potentially improving patient outcomes. Clinicians should consider utilizing this technology as a supplementary tool to enhance decision-making and workflow efficiency in acute stroke care.

Conclusion

The automated detection of the hyperdense artery sign on non-contrast CT shows promise as a reliable adjunctive tool for early identification of large vessel occlusion. This approach may significantly enhance workflow readiness in time-sensitive clinical settings.

References

1. European Radiology, 2023 -- CT Angiography with a Focused Approach for Targeted Imaging of Arteries Involved in Stroke: An Evaluation of Technical Viability
2. Evaluation of Atherosclerosis Load Using AI-Enhanced Coronary Computed Tomography Angiography: A Comparative Study with Intravascular Ultrasound in the INVICTUS Registry
3. Automated Assessment of Aortic Contrast-Enhanced CT Angiograms for Tailored Dose Optimization in Patients
4. npj Digital Medicine -- Volumetric Assessment of Infrarenal Abdominal Aortic Aneurysms Using Deep Learning Techniques on CTA Images
5. 2026 Guideline for the Early Management of Patients With Acute Ischemic Stroke: A Guideline From the American Heart Association/American Stroke Association
6. DAWN Trial design: Patients with acute ischemic stroke
7. Sensitivity and Specificity of the Hyperdense Artery Sign, Stroke
8. 2026 Guideline for the Early Management of Patients With Acute Ischemic Stroke: A Guideline From the American Heart Association/American Stroke Association
9. DAWN Trial design: Patients with acute ischemic st
10. Sensitivity and Specificity of the Hyperdense... : Stroke