Related Experiment Video
Updated: Oct 16, 2025
08:40
Isokinetic Robotic Device to Improve Test-Retest and Inter-Rater Reliability for Stretch Reflex Measurements in Stroke Patients with Spasticity
Published on: June 12, 2019
7.6K
Elucidating factors influencing machine learning algorithm prediction in spasticity assessment: a prospective
Natiara Mohamad Hashim1, Jingye Yee2, Nurul Atiqah Othman3
1Department of Rehabilitation Medicine, Faculty of Medicine, Universiti Teknologi MARA, Sungai Buloh, Malaysia.
Computer Methods in Biomechanics and Biomedical Engineering
|October 20, 2021
Summary
Machine learning models (MLM) show promise in rehabilitation but struggle with spasticity prediction. Data biases significantly reduce MLM accuracy on ambiguous datasets, highlighting the need for careful training considerations.
Area of Science:
- Biomedical Engineering
- Rehabilitation Technology
- Artificial Intelligence in Medicine
Background:
- Machine learning models (MLM) are increasingly used in rehabilitation for prediction and AI training.
- High-quality data is crucial for developing effective MLM algorithms.
- Factors influencing MLM performance in predicting spasticity severity are not well understood.
Purpose of the Study:
- To train and validate an MLM algorithm for spasticity assessment.
- To evaluate the MLM's predictive performance on ambiguous spasticity datasets.
- To identify factors affecting MLM accuracy in spasticity prediction.
Main Methods:
- Recruited 47 participants with central nervous system pathology.
- Collected four biomechanical properties of spasticity using wearable sensors.
- Trained and validated MLMs (SVM, Decision Tree, Random Forest) on clean data and tested on ambiguous datasets.
Main Results:
- MLM accuracy on validation data: SVM (96%), Decision Tree (52%), Random Forest (72%).
- MLM accuracy on ambiguous datasets significantly dropped: SVM (20%), Decision Tree (23%), Random Forest (23%).
- Data biases and variances in disease background, pathophysiology, and anatomy impact MLM performance.
Conclusions:
- MLM performance in spasticity prediction is highly sensitive to data quality and characteristics.
- Biases and variances in training data can lead to substantial reductions in predictive accuracy.
- Future MLM development for spasticity assessment requires careful consideration of data heterogeneity.
