You are here

Article Text

Article menu

Download PDFPDF

SNIS 21st annual meeting oral abstracts
O-044 Large vessel occlusion detection in acute ischemic stroke on non-contrast CT imaging using AI-driven software
Free
  1. A Rai1,
  2. A Al Halak1,
  3. M Abdalkader2,
  4. T Nguyen2,
  5. D Kallmes3,
  6. W Brinjikji3,
  7. A Perry4,
  8. Z Woodhead5,
  9. G Harston4,
  10. D Carone6
  1. 1Neuroradiology, Rockefeller Neuroscience Institute, Morgantown, WV
  2. 2Neurology, Boston Medical Center, Boston, MA
  3. 3Radiology, Mayo Clinic Rochester, Rochester, MN
  4. 4Brainomix, Oxford, UK
  5. 5Brainomix, Chicago, UK
  6. 6Brainomx, Oxford, UK

Abstract

Introduction This study evaluated the predictive capability of a FDA-cleared deep learning algorithm in identifying anterior circulation large vessel occlusions (LVO) on non-contrast CT (NCCT) in patients with suspected acute ischemic stroke (AIS).

Materials and Methods The software was validated in a multisite dataset (N=612), obtained from three clinical sites in the USA. The performance of Brainomix 360 Triage Stroke using non-contrast CT (NCCT) imaging was evaluated and compared against the software’s CT angiography (CTA) based LVO predictive algorithm (Brainomix 360 Triage LVO). The ground truth for these was the baseline CTA reviewed by experienced neuroradiologists aided by clinical information. In a subgroup analysis, the ability to identify LVO patients based on clinical scores (NIHSS) in conjunction with NCCT was explored. In a final subgroup analysis, the performance of Triage Stroke was compared to the unaided performance of radiologists (4 neuroradiologists and 6 general) using the same NCCT.

Results Triage Stroke detected LVO on NCCT with sensitivity of 67% and specificity of 93%. By comparison, CTA-based Triage LVO had a sensitivity of 89% and specificity of 93% for LVO detection. Examining the subset of cases with NIHSS information, it was observed that the combination of clinical information with Triage Stroke yielded a highly specific operating point (99%) and similar sensitivity (65%). Finally, Triage Stroke outperformed radiologists at the same task (radiologists: sensitivity 48%, specificity 89%, p=0.001).

Conclusion Triage Stroke demonstrated strong predictive capabilities for detecting anterior circulation LVO. While less sensitive than using CTA imaging, the device outperformed radiologists. Coupled with NIHSS, it may help centers without routine access to angiography identify thrombectomy candidates, especially in resource constrained environments worldwide.

View this table:
Abstract O-044 Table 1

Predictive values of NCCT/CTA LVO detection

Abstract O-044 Figure 1
Abstract O-044 Figure 1

Disclosures A. Rai: 2; C; Stryker Neurovascular, Cerenovus. A. Al Halak: None. M. Abdalkader: 2; C; Brainomix. T. Nguyen: 2; C; Brainomix. D. Kallmes: 1; C; Brainomix. W. Brinjikji: 1; C; Brainomix. A. Perry: 5; C; Brainomix. Z. Woodhead: 5; C; Brainomix. G. Harston: 5; C; Brainomix. D. Carone: 5; C; Brainomix.

https://doi.org/10.1136/jnis-2024-SNIS.44

Statistics from Altmetric.com

    Request Permissions

    If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.