Introduction/Purpose Noncontrast CT (NCCT) is the first line screening modality to evaluate for intracerebral hemorrhage (ICH) and early ischemic changes in the setting of acute ischemic stroke (AIS). Large vessel occlusions (LVOs) are a major cause of AIS, but challenging to detect on NCCT alone. The purpose of this study is to evaluate a fully-automated AI platform called RAPID NCCT Stroke (iSchemaView, Menlo Park, CA) for ICH and LVO detection as compared to expert neuroimaging readers.
Materials and Methods In this IRB approved retrospective, multicenter study, 244 patients with suspected acute stroke were included. Stand-alone performance of the RAPID NCCT Stroke platform was assessed (based on the consensus of 3 neuroradiologists who determined reference standard from CT/CTA) and sensitivity and specificity were determined. The platform’s performance was then compared to expert interpretation by neuroimaging readers comprised of eight general radiologists (GR) and three neuroradiologists (NR) in detecting ICH and hyperdense vessel sign (HVS) as a marker of LVO. The primary hypothesis was that the sensitivity of RAPID NCCT Stroke for detection of LVO would be superiority to GR readers and non-inferiority to NR readers. Receiver operating characteristics (ROC) curve was used to evaluate the performance of each reader based on a rating system from 1 to 5 to express confidence in LVO detection. P < 0.05 was considered significant.
Results A total of 244 cases were included in this study. Of the 244, 115 were LVOs and 26 were ICHs. One hundred three cases did not have LVO nor ICH. Stand-alone performance of the platform demonstrated sensitivities and specificities of 96.2% and 99.5% respectively for ICH as well as 63.5% and 95.1% for LVO detection. In assessing non-inferiority of the platform compared to all 11 readers, the platform achieved a significantly higher sensitivity (63.5% versus 43.6%, p < 0.0001), meeting criteria for both non-inferiority and superiority. When compared to the eight GR readers only, the platform also showed superiority with significantly higher sensitivity (63.5% versus 40.9%, p = 0.001).
Conclusion The RAPID NCCT Stroke platform demonstrated superior performance to radiologists for detecting LVO from a NCCT. This could lead to earlier identification of LVO and faster treatment times. Prospective studies are needed for further validation.
Disclosures V. Yedavalli: 2; C; RAPID (iSchemaview, Menlo Park, CA). J. Heit: 2; C; RAPID (iSchemaview, Menlo Park, CA). S. Dehkharghani: None. H. Haerian: None. J. McMenamy: None. J. Honce: None. V. Timpone: None. C. Harnain: None. A. Kesselman: None. A. Filly: None. A. Beardsley: None. B. Sakamoto: None. C. Song: None. J. Montuori: None. B. Navot: None. F. Mena: None. D. Giurguitiu: None. F. Kitamura: None. F. Lima: None. H. Coelho: None. F. Mont’Alverne: None. G. Albers: 4; C; RAPID (IschemaView, Menlo Park, CA).
Statistics from Altmetric.com
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.