UbiStroke: Multimodal Sensing of Stroke-Related Deficits

Nadir Weibel, Vish Ramesh, Andrew Nguyen, Brett C. Meyer, Gert Cauwenberghs, Kunal Agrawal
Last updated on Sep 28, 2026

Overview

Stroke assessment rests on human observation. Diagnosing a stroke quickly enough to treat it requires a specialist, and the deficit scales in clinical use record symptoms categorically, which makes outcome prediction crude: existing scales are generally unable to say whether a patient will do well or very poorly.

UbiStroke develops techniques to identify and quantify stroke-related deficits from multiple sensors at once, producing a per-patient stroke signature from body posture, facial droop, pupil tracking, speech patterns, and other symptoms. Machine learning models trained on this data detect stroke-associated hemiparesis and support the clinician rather than replacing the exam.

The work evaluated UbiStroke alongside the clinician-performed NIHSS exam, and studied how clinicians rely on a computational aid during an acute assessment.

Funding and External Collaborations

UbiStroke is a collaboration with the UCSD Stroke Center, the UCSD Bioengineering Department, the Institute of Neural Computation, and HomniHealth. Research leading to UbiStroke has been funded by the National Science Foundation, the NSF I-Corps, the National Library of Medicine at NIH, IBM Research, and UCSD’s Office of Research Affairs.

HealthCare AI Sensing Earlier
Nadir Weibel
Nadir Weibel
Professor of Computer Science and Engineering
Vish Ramesh
Vish Ramesh
Postdoctoral Researcher at UCSD
Biomedical Informatics
(NIH NLM Fellow, Co-Advised with Gert Cauwenberghs, Bioengineering)

Publications

Assessing Clinicians' Reliance on Computational Aids for Acute Stroke Diagnosis
PervasiveHealth 2020
The rapid rise of computational aids for stroke diagnosis have led to important concerns about clinicians developing an over-dependence on technology. Other studies have assessed reliance on clinical decision support systems in fields like diabetes, but no such study exists for stroke diagnosis.
Cite Project DOI
Assessing Clinicians' Reliance on Computational Aids for Acute Stroke Diagnosis
Developing aids to assist acute stroke diagnosis
CHI EA 2020
The only known therapy for stroke, a major leading cause of death and disability, has to be administered within 3 hours of the onset of symptoms for it to be effective.
Cite Project DOI
Developing aids to assist acute stroke diagnosis
Stroke-associated hemiparesis detection using body joints and support vector machines
PervasiveHealth 2018
Hemiparesis, the weakness of one side of the body, affects the ability of stroke survivors to move and walk.
Cite Project DOI
Stroke-associated hemiparesis detection using body joints and support vector machines
Exploring stroke-associated hemiparesis assessment with support vector machines
PervasiveHealth 2017
Hemiparesis, the weakness of one side of the body, affects the ability of stroke survivors to move and walk. With prevalence in 80% of survivors, hemiparesis is an important measure for stroke severity.
Cite Project DOI