Automated Identification of Hyperdense Artery Sign on Non-Contrast CT for Swift Detection of Large Vessel Occlusion: A Multicenter Validation Analysis - Takeaways - MDSpire

Automated Identification of Hyperdense Artery Sign on Non-Contrast CT for Swift Detection of Large Vessel Occlusion: A Multicenter Validation Analysis

  • By

  • Hirofumi Tsuji

  • Akira Ishii

  • Hidehisa Nishi

  • Yu Abekura

  • Takuya Fuchigami

  • Atsushi Tachibana

  • Hirotaka Ito

  • Yoshiki Arakawa

  • April 20, 2026

  • 0 min

Share

  • 1

    A deep-learning model was developed for automated detection of the hyperdense artery sign (HAS) on non-contrast CT to aid in identifying large vessel occlusion.

  • 2

    The model demonstrated high reliability in a multicenter CSC triage cohort, achieving a positive predictive value of 92.0% and sensitivity of 76.2%.

  • 3

    In a broader real-world cohort, the model maintained a sensitivity of 74.3% and a negative predictive value of 94.6%, indicating preserved diagnostic performance.

  • 4

    The observer study showed that AI assistance significantly improved human detection performance of HAS, enhancing the JAFROC Figure of Merit.

  • 5

    The findings support the model's role as an adjunctive pre-CTA alert to facilitate earlier workflow readiness in time-sensitive stroke care.

Original Source(s)

Automated Identification of Hyperdense Artery Sign on Non-Contrast CT for Swift Detection of Large Vessel Occlusion: A Multicenter Validation Analysis

  • by Hirofumi Tsuji, Akira Ishii, Hidehisa Nishi, Yu Abekura, Takuya Fuchigami, Atsushi Tachibana, Hirotaka Ito, Yoshiki Arakawa

  • April 20, 2026

Related Content

Dual impact of flow diverters on ischemic and hemorrhagic outcomes in vertebral artery dissecting aneurysms presenting with ischemia

Deep-Motion-Net: GNN-based volumetric liver shape reconstruction from single-view 2D projections

Case Report: Spontaneous Dissection of the Renal Pelvis and Ureter