Related Experiment Video
Updated: Jul 7, 2026
09:10
Determining the Functional Status of the Corticospinal Tract Within One Week of Stroke
Published on: February 22, 2020
Advanced Multimodal AI for Predicting Long-Term Functional Outcomes After Ischemic Stroke Using Only Admission Data
Fiona McBride1, Haoxu Huang2, Anjali Kiran Kapoor3
1New York University Medical Center: NYU Langone Health.
Research Square
|June 29, 2026
Summary
A new AI model using admission data accurately predicts stroke outcomes, outperforming existing tools for better early clinical decisions. This advanced prognostic tool improves patient risk stratification after acute ischemic stroke.
Area of Science:
- Artificial Intelligence in Medicine
- Neurology
- Medical Prognostics
Background:
- Current stroke prognostication relies on limited variables, despite richer admission data.
- Established risk scores like THRIVE and SPAN-100 have limitations in predicting functional outcomes.
Purpose of the Study:
- To develop and validate a multimodal AI model for predicting functional outcomes (modified Rankin Scale - mRS) after acute ischemic stroke.
- To compare the AI model's performance against existing prognostic tools (THRIVE, SPAN-100).
Main Methods:
- Retrospective study of ischemic stroke/TIA patients (n=6,915).
- Trained modality-specific AI models on non-contrast head CT, clinical notes, and structured data.
- Ensembled models to predict binary and ordinal mRS at discharge and 90 days, validated on an external cohort.
Main Results:
- The multimodal AI ensemble significantly outperformed THRIVE and SPAN-100 in predicting discharge and 90-day binary mRS (e.g., AUROC 0.859 vs. THRIVE/SPAN-100 at discharge).
- AI model demonstrated superior performance in ordinal mRS prediction with better QWK and lower MAE.
- The AI model reclassified one-third of patients and improved outcome prediction accuracy in 74% of discordant cases.
Conclusions:
- A well-calibrated multimodal AI model accurately predicts post-stroke functional outcomes using only admission data.
- This AI model surpasses existing prognostic tools, offering enhanced support for early clinical decision-making in stroke care.
