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

Key numbers

• 78.5% versus 59.4%
• 87.0% versus 85.9% for the

Caveats

• Not randomized: its findings are associations.
• Preprint; not yet peer reviewed.

Background and Purpose: Prehospital stroke recognition remains suboptimal, with nearly one-third of strokes unidentified before hospital arrival, and neurological assessment being inherently subjective, with substantial interobserver variability even among certified clinicians. APOLLAI, a multimodal artificial intelligence system, was developed to detect stroke-related neurological deficits from video and audio recordings. We compared the diagnostic performance of APOLLAI and emergency medical services (EMS) personnel against an expert consensus reference standard. Methods: In this prospective study, video recordings of 295 patients with suspected stroke (mean age, 70.3 {+/-}13.3 years) were obtained at Nimes University Hospital (August 2023-July 2025). Five National Institutes of Health Stroke Scale items (best gaze, facial palsy, motor arm, aphasia, dysarthria) were independently annotated by seven certified stroke neurologists, combined into consensus labels using the Dawid-Skene algorithm, and separately assessed by EMS personnel (mean of 15 raters per patient). APOLLAI, a multimodal machine-learning framework that combines computer vision, audio-based deep learning, and a final support vector machine (SVM) fusion classifier, was evaluated against EMS using sensitivity, specificity, and the area under the receiver operating characteristic curve (AUC). Results: Interrater agreement among the seven neurologists was moderate overall (mean {kappa}=0.541, range, 0.367-0.732), indicating an intrinsic difficulty of neurological assessment. APOLLAI showed higher sensitivity than EMS for all five neurological tasks and for the global assessment (78.5% versus 59.4%), with broadly comparable specificity (87.0% versus 85.9% for the global assessment). AUC values ranged from 0.839 (facial palsy) to 0.966 (best gaze), with a global AUC of 0.882. Conclusions: APOLLAI demonstrated higher sensitivity than EMS personnel for detecting stroke-related neurological deficits, with preserved specificity. These findings support its potential as a standardized, scalable adjunct to prehospital stroke recognition, warranting prospective validation under real-world EMS conditions.

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

Leverage3 / 5

A method or resource many groups across the field will adopt within a year.

10% of this desk’s rubric

Magnitude3 / 5

Large gain: roughly 2x, or a clear new state of the art on a hard, unsaturated problem.

20% of this desk’s rubric

Evidence3 / 5

Solid: multiple benchmarks or cohorts, ablations, fair baselines, released code or data.

32% of this desk’s rubric

Novelty2 / 5

A new combination of known ideas.

8% of this desk’s rubric

Trajectory2 / 5

Some room to improve with obvious engineering.

5% of this desk’s rubric

Stakes3 / 5

Meaningful benefit to many people within a few years.

25% of this desk’s rubric

• Citations0 (reference 20, via semantic-scholar, Oct 23, 2026, 06:17 UTC)

• Reviewed by heuristic-v7 on Oct 23, 2026, 06:17 UTC. Paper type: clinical.
• Categories: neurology
• BRIEF, No.6 in the Medicine edition of October 23, 2026.