Abstract WP168: Creating an Artificial Intelligence-Driven Chatbot to Triage Patients for Mechanical Thrombectomy

Abstract

Introduction: Mechanical thrombectomy (MT) is the standard of care in patients presenting with acute ischemic stroke (AIS) due to large vessel occlusion (LVO). Early recanalization is critical to restoring blood flow to ischemic brain tissue. Previous literature has reported disparities in access to MT capable centers in patients with AIS. Early identification of MT eligibility could help stratify access to MT capable centers. Our study created an artificial intelligence (AI)-driven chatbot to help determine eligibility for MT. Methods: Guidelines established by the American Heart Association (AHA) and evidence from randomized clinical trials were used to determine MT eligibility based on presence of LVO, absence of intracranial hemorrhage (ICH), National Institute of Health Stroke Scale (NIHSS) and ASPECTS score. These criteria were utilized to develop a custom chatbot using the Python library for Open AI (San Francisco, California). The chatbot was used to determine eligibility and report reasons for each decision. Responses on MT eligibility were compared to the surgeon’s decision on MT eligibility at our institution between May 2024- August 2024. Results: 34 patients presented with AIS and 52.9% (n= 18) underwent MT. The chatbot’s response agreed with the surgeon’s decision in 61.7% patients (n= 21). All patients who underwent MT were successfully identified by the chatbot. There were 13 cases of disagreement where the chatbot deemed patients to be ineligible for MT, however, the surgeon chose to perform MT after outweighing the benefits and risks of intervention. The chatbot was made available in the form of an open access web application (Figure 1, 2). https://huggingface.co/spaces/thrombectomypredictor/MT_eligibility Conclusion: Our AI-driven chatbot identified all instances of MT eligibility and could be incorporated into AI-based imaging software to streamline the process of referrals to MT-capable centers.

Ähnliche Arbeiten