Automated Identification of Hyperdense Artery Sign on Non-Contrast CT for Swift Detection of Large Vessel Occlusion: A Multicenter Validation Analysis - Summary - MDSpire
Advertisement
Automated Identification of Hyperdense Artery Sign on Non-Contrast CT for Swift Detection of Large Vessel Occlusion: A Multicenter Validation Analysis
To develop and validate a fully automated deep-learning model for detecting the hyperdense artery sign (HAS) on non-contrast CT (NCCT) as an adjunctive pre-CTA alert for swift detection of large vessel occlusion (LVO).
Key Findings:
In Part 1A, sensitivity was 76.2%, specificity 87.0%, accuracy 79.9%, and PPV 92.0%.
In Part 1B, sensitivity was 74.3%, specificity 82.7%, PPV 44.1%, NPV 94.6%, and accuracy 81.4%.
The model significantly improved HAS detection performance in the reader study, increasing JAFROC Figure of Merit from 0.71 to 0.77 (p < 0.01).
Interpretation:
The model demonstrated high reliability of positive alerts in a high-acuity setting and preserved sensitivity/specificity in a broader population, supporting its role as an adjunctive tool for early workflow readiness.
Limitations:
The model's PPV was lower in the broader cohort, indicating potential challenges in lower-prevalence settings.
The study was retrospective and may not fully capture real-world variability.
Conclusion:
The automated HAS detection model enables rapid identification of potential LVO on NCCT, supporting earlier workflow readiness in stroke care without replacing CTA.
