Related Experiment Video
Updated: Jun 7, 2025
09:10
Determining the Functional Status of the Corticospinal Tract Within One Week of Stroke
Published on: February 22, 2020
8.5K
Machine learning for early dynamic prediction of functional outcome after stroke
Julian Klug1, Guillaume Leclerc2, Elisabeth Dirren1
1Stroke Research Group, Department of Clinical Neurosciences, University Hospital and Faculty of Medicine, Geneva, Switzerland.
Communications Medicine
|November 13, 2024
Summary
This study introduces a novel machine learning model for hourly stroke outcome prediction, integrating diverse patient data within 72 hours. The model accurately forecasts mortality and morbidity, outperforming static methods.
Area of Science:
- Neurology
- Artificial Intelligence
- Medical Informatics
Background:
- Accurate prediction of stroke outcomes is crucial for effective treatment planning and resource management.
- Dynamic changes in patient condition during the initial days post-stroke complicate outcome prediction.
- Existing prediction methods often lack the granularity to adapt to rapid physiological changes.
Purpose of the Study:
- To develop and validate a machine learning model for hourly prediction of mortality and morbidity after ischemic stroke.
- To integrate continuous clinical, physiological, imaging, and biological data within 72 hours of hospital admission.
- To identify key predictors influencing stroke outcomes using explainable AI methods.
Main Methods:
- A transformer-based machine learning model was developed using 2,492 ischemic stroke admissions (2,131,752 data points).
- The model incorporated continuous data streams (clinical, physiological, imaging, biological) recorded within 72 hours of admission.
- Shapley additive explanations (SHAP) were employed for feature importance analysis; external validation was performed using the MIMIC-III database.
Main Results:
- The transformer model achieved an Area Under the Receiver Operating Characteristic Curve (AUC) of 0.830 on admission for 3-month mortality prediction, improving to 0.893 by 72 hours.
- The model demonstrated superior performance compared to all static models when validated on an independent cohort.
- Key predictors included continuous clinical assessment, patient demographics, time to acute treatment, and markers of inflammation and organ dysfunction.
Conclusions:
- Machine learning models integrating longitudinal clinical, physiological, imaging, and biological data show significant potential for improving stroke outcome prediction.
- The developed transformer model offers hourly updated predictions and explanations, enhancing its clinical applicability.
- This approach provides a dynamic and interpretable tool for managing stroke patients.
