Related Experiment Video
Updated: Jan 9, 2026
A Method for Quantifying Upper Limb Performance in Daily Life Using Accelerometers
Published on: April 21, 2017
Artificial intelligence to predict modified rankin score after acute stroke symptoms using wrist-worn triaxial
Benjamin R Kummer1, Alexander Gerlach2, Shaun Kohli3
1Department of Neurology, Icahn School of Medicine at Mount Sinai, New York, NY, USA; Clinical Neuro-Informatics Center, Icahn School of Medicine at Mount Sinai, New York, NY, USA; Windreich Department of Artificial Intelligence and Human Health, Icahn School of Medicine at Mount Sinai, New York, NY, USA; Hasso Plattner Institute for Digital Health at Mount Sinai, New York, NY, USA.
Artificial intelligence (AI) using wearable sensors can predict stroke recovery outcomes. This technology offers an objective method to track functional changes after stroke, improving patient monitoring.
Area of Science:
- Biomedical Engineering
- Neurology
- Data Science
Background:
- The modified Rankin Scale (mRS) is standard for post-stroke functional assessment but has limitations including subjectivity and cultural barriers.
- Wearable accelerometry and AI present a potential for objective, continuous monitoring of functional status after stroke.
Purpose of the Study:
- To evaluate the efficacy of AI models in predicting modified Rankin Scale (mRS) scores and functional changes post-stroke using wearable sensor data.
- To explore the potential of AI-driven accelerometry for objective characterization of post-stroke functional status.
Main Methods:
- Analysis of data from the REACH Stroke-Sleep study, including triaxial wrist-worn accelerometry.
- Training and evaluation of logistic regression (LR), random forest (RF), and long short-term memory (LSTM) models to predict 1- and 6-month mRS scores and mRS changes.
- Model performance assessed using area under the receiver-operating curve (AUROC) with 5-fold cross-validation.
Main Results:
- Random Forest models demonstrated superior performance in predicting 1-month mRS scores (AUROC 0.65) and 6-month functional changes (AUROC 0.62) compared to LSTM and LR models.
- AI models achieved prediction accuracy above chance for short-term mRS scores and functional changes post-stroke.
- RF models showed better predictive power for 1-month mRS scores and 1-to-6 month mRS changes.
Conclusions:
- AI applied to wearable accelerometry data provides a proof-of-concept for capturing meaningful post-stroke functional variations.
- This approach can potentially inform future real-time monitoring frameworks for stroke patients.
- Integrating multimodal wearable and clinical data may further enhance the prediction accuracy of post-stroke functional outcomes.
