Abstract 1122‐000033: Automated Detection of Intracerebral Hemorrhage Using Artificial Intelligence: Pilot Deployment of Viz ICH

Abstract

Introduction: Stroke is a major cause of morbidity and mortality. Hemorrhagic strokes are often more severe and associated with higher mortality when compared to ischemic stroke and account for approximately 13% of all strokes. Initial care for patients with intracerebral hemorrhage (ICH) is in part guided by neuroimaging findings. Non‐contrast computed tomography (NCCT) is often the first imaging obtained in the work up of the acute stroke patient given its diagnostic accuracy for hemorrhage, ubiquity, low cost, and short scan time. Immediate evaluation of imaging by stroke experts, such as neurologists, neurosurgeons, and radiologists, is essential. Artificial intelligence tools can help to expedite image assessment and careteam coordination thereby accelerating time to treatment. In this study, we report on the use of Viz ICH, an AI‐enhanced ICH detection platform, to identify ICH on initial CT and coordinate emergent care in an urban health system with an ICH Center. Methods: All consecutive stroke codes presenting with ICH from May 2019 to August 2019 were eligible for analysis. Non‐contrast CT (NCCT) was conducted for each patient and submitted to the Viz ICH in a prospective fashion. An automated volumetric analysis of these NCCTs was conducted by Viz ICH and assessment was conducted for potential ICH. If suspected ICH was detected, Viz ICH sent an automated prompt to the stroke care team for review. CT impressions provided by radiologists served as the clinical reference standard test and Viz ICH output served as the index test. Diagnostic accuracy tests were then performed. Results: A total of 682 patients were analyzed for ICH, out of which 28 patients were positive for intracerebral hemorrhage (ICH) (4%) and 654 were negative for hemorrhage (96%) based on radiology impressions. Viz ICH was able to correctly identify hemorrhages in 25/28 patients and non‐hemorrhages in 650/654 patients. Overall, the software had high diagnostic accuracy with 89.3% sensitivity, 99.4% specificity, and an overall accuracy of 99.0%. The software also had a positive predictive value of 86.2%, a negative predictive value of 99.5%, a positive likelihood ratio of 145.98, and a negative likelihood ratio of 0.108. Conclusions: Viz ICH is an AI‐enhanced platform that may help in the diagnosis and detection of ICH, with a sensitivity of 89.3 and a specificity of 99.4% in this preliminary study. Though future validation studies with a larger cohort of patients positive for each type of hemorrhage must be conducted for true diagnostic accuracy data, Viz ICH has the potential to be an adjunct tool to streamline ICH triage, reduce treatment delays, and improve outcomes of patients presenting with ICH.

Keywords

• Imaging
• Intracerebral Hemorrhage