Article Text

Download PDFPDF
E-108 Aladin study: automated large artery occlusion detection in stroke imaging study – a multicenter analysis
  1. C Barreira1,
  2. M Bouslama1,
  3. J Lim2,
  4. A Al-Bayati1,
  5. Y Saleem1,
  6. T Devlin3,
  7. D Haussen1,
  8. M Froehler2,
  9. J Grossberg4,
  10. B Baxter3,
  11. M Frankel1,
  12. R Nogueira1
  1. 1Marcus Stroke Center – Neurology, Grady Memorial Hospital – Emory University, Atlanta, GA
  2. 2Neurology and Neurosurgery, Vanderbilt University Medical Center, Nashville, TN
  3. 3University of Tennessee Health Science Center, Chattanooga, TN
  4. 4Marcus Stroke Center – Neurosurgery, Grady Memorial Hospital – Emory University, Atlanta, GA

Abstract

Introduction Large Vessel Occlusion (LVO) for Acute Ischemic Stroke (AIS) remains a public health issue as there is significant morbidity and mortality when left untreated. Timely recognition, therefore, is of utmost significance, as there are validated therapeutic options involving cerebrovascular reperfusion. Best patient care relies on tailored and expeditious clinical identification, using multimodal neuroimaging and facilitating referrals to comprehensive centers. Artificial Intelligence (AI)-guided technologies applied to medical fields are being used increasingly and may improve LVO recognition. We present an AI-based algorithm for LVO detection.

Methods A multi-center retrospective analysis of CTAs, randomly picked from retrospective cohort of AIS patients, with and without anterior circulation LVOs, admitted at three tertiary stroke centers, from 2014–2017, was performed. CTAs were analyzed and graded by experienced stroke neurologists for presence and site of occlusion to establish the ground truth. These same studies were analyzed by Viz-AI-Algorithm v3.04 – a Convolutional Neural Network programmed to detect MCA-M1 and/or ICA-T occlusions. Our primary analysis included ICA-T and/or MCA-M1 LVOs versus non-LVOs/more distal occlusions. Our secondary analysis also included proximal MCA-M2 vs non-LVOs.

Results Interim analysis of 875 CTAs, of three tertiary stroke centers, show: 49.5% males, bNIHSS 15[IQR 10–20], bASPECTS 10[8–10]. Of all, 45.9% were considered as having LVO, and homogenous baseline characteristics among groups was found. Primary analysis shows an accuracy of 86%, sensitivity of 90.1%, specificity of 82.5, PPV of 81.8 and NPV of 90.6; AUC 86.3% (95% CI 0.83–0.90, p≤0.001) and ICC 84.1% (95% CI 0.81–0.86; p≤0.001). Maximal running time of the algorithm was under five minutes.

Conclusion Viz-AI-Algorithm has an impressive performance for detection of proximal intracranial LVO recognition. Endeavors to optimize detection of the MCA-M2 and all intracranial ICA occlusions have been enforced, as well as posterior circulation. So far, this is the first AI-algorithm for detecting intracranial anterior circulation LVOs.

Disclosures C. Barreira: None. M. Bouslama: None. J. Lim: None. A. Al-Bayati: None. Y. Saleem: None. T. Devlin: None. D. Haussen: None. M. Froehler: None. J. Grossberg: None. B. Baxter: None. M. Frankel: None. R. Nogueira: None.

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.