DeepStroke+: An Advanced Mobile AI Diagnostic Tool for Fast and Precise Detection of Acute Strokes in Mobile Stroke Units, Emergency Rooms, and Telestroke Triage

Project Summary/Abstract Nearly one million Americans suffer from a stroke every year, often resulting in serious long-term disability or death. Early detection and treatment of stroke is critical to improving patient outcomes. Several tools have previously been developed that attempt to screen and help identify patients having a stroke, but there are still many stroke cases that are missed, even after the patient arrives to the emergency room. This study aims to improve the identification and triage of stroke patients in emergency departments using artificial intelligence (AI) on multimedia patient data. Our hypothesis is that real-time, standardized, and reproducible stroke assessment tools using AI can improve stroke triage and decrease missed diagnoses in patients with minor-to-moderate neurological symptoms. The study plans to develop a DeepStroke+ augmented-intelligence framework to triage any kind of strokes (ischemic stroke, hemorrhagic stroke, transient ischemic attack, etc) versus stroke mimics commonly seen in emergency settings. We developed a DeepStroke+ framework for stroke triage using facial videos of English-speaking patients who are describing the “Cookie Theft” picture from the Boston Diagnostic Aphasia Examination. Based on this technology, we will develop a stroke triage app and validate it in different triage scenarios from mobile stroke units to emergency room triage in local and telestroke scenarios.