> ## Content Index
> Fetch the complete content index at: https://www.coios.me/llms.txt
> Use this file to discover other available public pages before exploring further.

# Video-and-audio AI system detected stroke deficits more sensitively than EMS raters
- URL: https://www.coios.me/video-and-audio-ai-system-detected-stroke-deficits-more/
- Published: 2026-10-25T11:49:59.000Z
- Updated: 2026-10-26T09:01:14.000Z
- Description: Against a consensus of seven neurologists, an AI system reading video and audio of 295 suspected stroke patients detected neurological deficits with 78.5% sensitivity, versus 59.4% for emergency medical services (EMS) personnel, with similar specificity.
- Author: Daniel Ryan
- Tags: Data, models and systems, AI in medicine, Health systems and access, France, #preprint, #new, #item, #source-2026-10, #score-62

Researchers tested APOLLAI, an AI system that detects stroke-related neurological deficits from video and audio recordings. They compared it with emergency medical services (EMS) personnel in 295 patients with suspected stroke at Nimes University Hospital. Seven stroke neurologists scored five National Institutes of Health Stroke Scale items: gaze, facial palsy, arm weakness, aphasia and dysarthria. Their combined scores formed the reference standard, and about 15 EMS raters assessed each patient. Overall, APOLLAI had 78.5% sensitivity versus 59.4% for EMS, specificity of 87.0% versus 85.9%, and AUC 0.882\. Neurologist agreement was only moderate (mean kappa 0.541), and all assessments used recordings at one centre.

*Why it is interesting: Nearly one-third of strokes go unrecognised before hospital arrival. On the same recordings, the AI system flagged more deficits than EMS raters while keeping specificity.*

Source

[medRxiv, 22 October 2026](https://doi.org/10.64898/2026.10.16.26365784?ref=coios.me)

Paper

APOLLAI, a multimodal AI system, versus EMS personnel for prehospital detection of stroke-related deficits

Authors

Javault C, Vargas Gonzalez E, Pibre L, Yriarte F, Thouvenot E

DOI

10.64898/2026.10.16.26365784

Type

Preprint

Design

Prospective single-centre diagnostic accuracy study, 295 suspected-stroke patients, Nîmes University Hospital, Aug 2023–Jul 2025; model vs EMS raters against Dawid-Skene consensus of seven neurologists

Verdict

New finding

Driver

[AI in medicine](https://www.coios.me/d-ai-medicine/)

Driver

[Health systems and access](https://www.coios.me/d-health-systems/)

## From the authors

A prospective field study will start at the end of 2026 in the US to acquire real-world prehospital data under live EMS conditions, which will enable assessment of APOLLAI performance across diverse operational contexts and represent a critical step toward demonstrating clinical utility beyond the controlled conditions of the present study.

Cédric Javault, AI-Stroke