You are here

Article Text

Article menu

Download PDFPDF

SNIS 21st annual meeting electronic poster abstracts
E-217 Detecting medium vessel occlusions and collateral assessment with multimodality AI approach
Free
  1. M Sowlat1,
  2. N Zaba1,
  3. A Zamarud2,
  4. G Albers3,
  5. B Campbell4,
  6. J Heit5,
  7. A Spiotta1
  1. 1Neurosurgery, Medical University of South Carolina, Charleston, SC
  2. 2Neuroimaging, Stanford hospital, Palo Alto, CA
  3. 3Neurology, Stanford hospital, Palo Alto, CA
  4. 4Neurology, Royal Melbourne hospital, Parkville, Australia
  5. 5Neurosurgery, Stanford hospital, Palo Alto, CA

Abstract

Introduction Recent advances and the use of Artificial intelligence (AI) is changing the standards of care in stroke patients. AI software has a high sensitivity and specificity for identifying Large vessel occlusions (LVO), but the accuracy for detecting Medium Vessel Occlusions (MeVOs) is limited. We investigated the performance of the combination of two different AI modules for the identification of MeVOs.

Methods 150 patients with MeVOs were identified on CTA, using CTP to help localize the lesions, from retrospective databases at MUSC, Stanford, and Royal Melbourne. Additionally, 55 patients with Large Vessel Occlusions (LVOs) on CTA were included. A neuroradiologist confirmed the occluded vessel on CTA for each patient. LVOs were defined as occlusions of the intracranial ICA or horizontal segment of the M1. MeVOs were defined at M2, M3, ACA (A2/A3), or PCA (P2/P3) occlusions. All cases were processed through Rapid LVO, the Rapid Vessel Density module (which assesses hemispheric differences in blood vessel density), and Rapid CTP using automated hypoperfusion intensity ratio (HIR) on CT perfusion (CTP), which estimates collateral flow, differed in LVOs vs. MeVOs.

Results Rapid LVO alone identified occlusions in 41/150 (27.3%) of MeVO patients. Rapid Vessel Density maps identified an additional 38/150 (25.3%) patients with either a 20–25% vessel density reduction or 26–40% reduction. Together, these two modules identified 79/150 (52.6%) of MeVO patients. 54/55 (98.1%) of the LVOs were detected by the Rapid LVO module. The mean HIR was lower in MeVO vs. LVO patients (0.25 vs. 0.45, P-value < 0.001). Patients with M1 or ICA occlusions (LVOs) had average HIRs of 0.46 and 0.47 respectively. Patients with M2, M3, ACA, or PCA occlusions had average HIRs of 0.30, 0.14, 0.22, and 0.20 respectively.

Conclusion MeVos detection was almost doubled with the use of the RapidAI Vessel Density in conjunction with the LVO module.

Abstract E-217 Figure 1
Abstract E-217 Figure 1
View this table:
Abstract E-217 Table 1

Disclosures M. Sowlat: None. N. Zaba: 5; C; Senior CAM, ISV. A. Zamarud: None. G. Albers: 2; C; iSchemaView, Genentech. B. Campbell: None. J. Heit: 2; C; MicroVention, RAPID AI, iSchemaView, Genentech. 4; C; iSchemaView. 6; C; Balt. A. Spiotta: 1; C; Penumbra, RapidAI, Microvention, Medtronic, Stryker. 2; C; Penumbra, RapidAI, Terumo.

https://doi.org/10.1136/jnis-2024-SNIS.322

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.