Educational content on VJNeurology is intended for healthcare professionals only. By visiting this website and accessing this information you confirm that you are a healthcare professional.

Share this video  

WSC 2026 | Stroke Agent: an AI decision support tool for stroke diagnosis and treatment

Jing Zhao, MD, PhD, Minhang Hospital, Fudan University, Shanghai, China, discusses the current integration of artificial intelligence (AI) in stroke diagnosis and treatment decision-making, highlighting the limitations of existing AI tools. Dr Zhao introduces the Stroke Agent, a decision support tool designed to address these limitations, which has shown promising results in mimicking the real emergency room timeline and providing evidence-based suggestions. This interview took place at the 18th World Stroke Congress (WSC) in Seoul, Korea.

These works are owned by Magdalen Medical Publishing (MMP) and are protected by copyright laws and treaties around the world. All rights are reserved.

Transcript

Most current stroke AI tools only read one scan image, but real stroke care has very difficult needs. Let me first talk about the limits of today’s AI and then introduce our stroke agent. We see three big mismatches between existing AI and real-world stroke care. First, a timing mismatch. Old AI treats each scan as one single task, but stroke decisions happen step by step. Second, data mismatch...

Most current stroke AI tools only read one scan image, but real stroke care has very difficult needs. Let me first talk about the limits of today’s AI and then introduce our stroke agent. We see three big mismatches between existing AI and real-world stroke care. First, a timing mismatch. Old AI treats each scan as one single task, but stroke decisions happen step by step. Second, data mismatch. Many systems need extra format conversion and data restructuring before they work. This is preceding with precious time. Third, audit mismatch. Many AI tools are black boxes. If a high-risk decision goes wrong, we cannot find when a mistake happened. That is exactly why we built Stroke Agent. It is a decision support tool for the whole acute ischemic stroke pathway. It copies the real emergency room timeline with three stages. Stage 1, emergency triage. Stage 2, the initial thrombolysis decision. Stage 3, final reperfusion plan. It reads original data directly: CT scans, ECG reports, and free doctor notes. No extra conversion. Every suggestion comes with its own evidence: two outputs and guidelines clauses and trial results. So we can trace the audit at every step. Our results are encouraging. We tested 100 respective cases under the same step-by-step information limits as doctors. Final decision coordination reached 86% and the complete pathway coordination was 76%. Three general language models reached 73% and 47%, and six neurologists reached 69% and 53%. Our biggest advantage appeared exactly at the thrombolysis decision step. For safety, our rate of output with serious potential harm was only 2%, and far below the other models. We also saved about 23% of processing time. But I must be honest, this only shows our system matches expert diagnoses. It doesn’t prove that patient outcomes improve. We still need prospective clinical validation; also, our data set is small and we haven’t tested real doctors working together with this system. The Stroke Agent is designed as an updatable, auditable, second opinion for hospitals without stroke specialists on site. It doesn’t replace doctors. It’s goal is to help more patients get reperfusion treatment within the golden time window.

This transcript is AI-generated. While we strive for accuracy, please verify this copy with the video.

Read more...

Choose your cookie settings

VJNeurology uses necessary cookies to make the site work. With your permission we'd also like to use analytics, preference and marketing cookies to understand how the site is used and improve it.

Choose an option to continue. You can change it at any time on our Cookie Policy page, linked at the bottom of every page.