CoChat

AI-enabled Parkinson's disease screening using smile videos

The DOI matches its Crossref record; 59 of the 59 references it cites were found in a registry or an index.

Abstract

BACKGROUND: Parkinson's disease (PD) diagnosis is challenging owing to insufficient access to clinical care. We present an efficient and accessible artificial intelligence-driven PD screening method leveraging the largest video dataset of facial expressions from 1452 unique participants, including 391 with PD - 300 of whom have been clinically diagnosed and 91 of whom self-reported the condition. METHODS: We recruited individuals across multiple countries - primarily North America - via social media, email outreach, and a PD research registry; patients undergoing in-person PD assessments at a U.S. clinic; clients of a U.S.-based PD wellness center and their caregivers; and individuals in Bangladesh identified as being at high risk for PD. Participants used an online tool to record themselves (either at home or in a clinical setting) mimicking three facial expressions (i.e., smile, disgust, and surprise). Facial landmarks and action unit-based features were extracted to quantify hypomimia. Machine-learning models were trained on these features to distinguish between individuals with and without PD. The model's generalizability was tested on external test datasets (from the U.S. clinic and Bangladesh). RESULTS: An ensemble of models trained on smile videos achieved an accuracy of 87.9 ± 0.1% and an area under the receiver operating characteristic curve (AUROC) of 89.3 ± 0.3% in 10-fold cross-validation, with a 76.8 ± 0.4% sensitivity, 91.4 ± 0.3% specificity, 73.3 ± 0.5% positive predictive value (PPV), and 92.7 ± 0.1% negative predictive value (NPV). On the U.S. clinic test set, it achieved an 80.3 ± 1.6% accuracy and an 83.3 ± 1.4% AUROC, with a 80.0 ± 2.5% sensitivity, and 80.5 ± 2.0% specificity. On the test set from Bangladesh, performance reached an 85.3 ± 1.4% accuracy with an 81.5 ± 1.8% AUROC. The specificity, sensitivity, and NPV remained competitive, while PPV declined to 35.7 ± 4.8%. No detectable bias was observed across sex and ethnic subgroups, except on the test dataset from Bangladesh, for which performance was significantly better for female participants. CONCLUSIONS: Smiling videos can effectively differentiate between individuals with and without PD, offering a potentially easy, accessible, and cost-efficient way to screen for PD, especially when access to clinical diagnosis is limited. (Funded by the National Institute of Neurological Disorders and Stroke of the National Institutes of Health number, P50NS108676 and others.).

Overview

Integrity signals

Registry

Crossref · 10.1056/aioa2400950

Cites

59 of 59 references verified0 citation errors · 0 not checkable

Statistics

Recomputing p-values needs the paper’s full text.

Verification

Claims the paper makes

We couldn’t retrieve this paper’s full text, so its claims and self-check weren’t run. The registry and citation checks still apply.

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Citations

Cites

Research works printed but not linked: 5

  1. Simon P. The choice of ignorance: the debate on ethnic and racial statistics in France. In: Simon P, Piché V, Gagnon AA, eds. Social statistics and ethnic diversity: cross-national perspectives in classifications and identity politics. Cham, Switzerland: Springer, 2015:65-87.
  2. Lugaresi C Tang J Nash H et al. MediaPipe: a framework for perceiving and processing reality. In: Third Workshop on Computer Vision for AR/VR at IEEE Computer Vision and Pattern Recognition (CVPR) vol. 2019. Long Beach CA: IEEE 2019 (https://static1.squarespace.com/static/5c3f69e1cc8fedbc039ea739/t/5e130ff310a69061a71cbd7c/1578307584840/NewTitle_May1_MediaPipe_CVPR_CV4ARVR_Workshop_2019.pdf).
  3. Dietterich TG. Ensemble learning. In: Arbib MA, ed. The handbook of brain theory and neural networks, second edition. Cambridge, MA: The MIT Press, 2002:405-412.
  4. Polikar R. Ensemble learning. In: Zhang C, Ma Y, eds. Ensemble machine learning: methods and applications. New York, NY: Springer, 2012:1-34.
  5. Bower KM. When to use Fisher’s exact test. American Society for Quality Six Sigma Forum Magazine vol. 2 August 1 2003 (https://www.researchgate.net/publication/265026286_When_to_Use_Fisher’s_Exact_Test).

Cited by

See all 5

Metadata

Authors
Adnan, T., Islam, M. S., Lee, S., Wasifur Rahman Chowdhury, E. M., Tithi, S. D., Noshin, K., Islam, M. R., Sarker, I., Rahman, M. S., Schneider, R. B., Adams, J. L., Dorsey, E. R., & Hoque, E.
Affiliations
  • Bangladesh University of Engineering and Technology (BD)
  • University of Rochester (US)
  • Department of Computer Science, University of Rochester, NY
  • University of Rochester Medicine (US)
  • National Institute of Neurosciences & Hospital (BD)
Venue
NEJM AI
Published
2025-06-26
Volume
2
Issue
7
Pages
AIoa2400950-AIoa2400950
ISSN
2836-9386 (ISSN-L)
OpenAlex
W4411672446
Cited by
5
References
59
Citations by year
2026: 4 · 2025: 1
FWCI
1.76
Type
journal article
Language
en
Open access
closed
Locations
  • NEJM AI (publishedVersion)
  • PubMed (publishedVersion)
Topics
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Face recognition and analysis
  • Physical Sciences › Computer Science › Artificial Intelligence › Emotion and Mood Recognition
  • Health Sciences › Health Professions › Health Information Management › Artificial Intelligence in Healthcare
Keywords
Parkinson's disease, screening, facial expression analysis, artificial intelligence, early detection
MeSH terms
Machine Learning, Aged, Artificial Intelligence, Facial Expression, Female, Humans, Male, Mass Screening, Middle Aged, Parkinson Disease, Smiling, Video Recording

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.