Large Language Models–Supported Thrombectomy Decision-Making in Acute Ischemic Stroke Based on Radiology Reports: Feasibility Qualitative Study

• DOI: 10.2196/48328
• Publication Date: 2025-02-13T15:46:51Z
• Authors: Jonathan Kottlors; Robert Hahnfeldt; Lukas Görtz; Andra-Iza Iuga; Philipp Fervers; Johannes Bremm; David Zopfs; Kai R Laukamp; Oezguer A Onur; Simon Lennartz; Michael Schönfeld; David Maintz; Christoph Kabbasch; Thorsten Persigehl; Marc Schlamann
• Keywords: Clinical Decision Making; Stroke

Background The latest advancement of artificial intelligence (AI) is generative pretrained transformer large language models (LLMs). They have been trained on massive amounts text, enabling humanlike and semantical responses to text-based inputs requests. Foreshadowing numerous possible applications in various fields, the potential such tools for medical data integration clinical decision-making not yet clear. Objective In this study, we investigate LLMs report-based example acute ischemic stroke (AIS), where image-based information may indicate an immediate need mechanical thrombectomy (MT). purpose was elucidate feasibility integrating radiology report other context therapy using LLMs. Methods A hundred patients with AIS were retrospectively included, which 50% (50/100) indicated MT, whereas not. LLM provided computed tomography report, neurological symptoms onset, patients’ age. performance AI model compared expert consensus regarding binary determination MT indication, sensitivity, specificity, accuracy calculated. Results had overall 88%, a specificity 96% sensitivity 80%. area under curve decision 0.92. Conclusions achieved promising determining eligibility based reports information. Our results underscore radiological integration. This investigation should serve as stimulus further LLMs, be used augmented supporting system human decision-making.