NeuroScreen: A Novel Tool to Differentiate Stroke vs. Migraine with Aura Using Visual Characterization and Machine Learn

Overview

Stroke and migraine with aura, which affect over 300 million people globally, present with similar visual symptoms requiring differentiation through costly neuroimaging that may not be available in underserved, rural, or low-resource communities. No existing tool combines multimodal visual assessment with machine learning for rapid web-based screening, and NeuroScreen fills this gap. We hypothesized that ML classification based on a novel, low-cost combination of four visual assessments could generate quantitative probability scores to classify stroke, migraine with aura, and controls. After providing informed consent, healthy controls, stroke survivors, and migraine with aura patients completed four modular visual assessments (static perimetry, contrast sensitivity, motion coherence, and flicker fusion) on the NeuroScreen web application (~6 minutes). A total of 55 participants (9 stroke, 15 migraine with aura, 31 controls) generated 70 features each (11 risk factors and 59 module characteristics) across six incremental ensemble models combining Random Forest, Logistic Regression, and Gradient Boosting. The most accurate version (v1.5, trained on 31 samples) correctly predicted 78.3% of held-out participants (p < 0.001). AUC values reached up to 0.964 for stroke (v1.3), 0.786 for migraine with aura (v1.5), and 0.980 for controls (v1.5) across model versions, indicating strong discriminative ability. Stroke, the highest-stakes diagnosis, achieved up to 100% predictive sensitivity (v1.3). These results support the hypothesis that web-based screening can detect distinct visual processing patterns and complement neuroimaging to support triage decision-making. Further data collection and validation (currently being pursued in a clinical setting) could improve classification accuracy and drive NeuroScreen towards real-world utility.

Competition history

• CSEF 2026 Medicine & Physiology (Track 2) (Senior Division) · Entry S-21-14

Related projects

Closest projects by meaning, across every fair and year in the corpus.

Browse more like this

• All Biomedical and Health Sciences projects
• CSEF Biomedical and Health Sciences projects
• Every CSEF 2026 project