Sensor-Based Classification of Post-Stroke Motor Impairment Using Fugl-Meyer Lower Extremity Scores

Area of Science:

• Biomedical Engineering
• Rehabilitation Science
• Neurology

Background:

• Post-stroke motor impairment significantly affects mobility and quality of life.
• Objective and automated assessment tools are needed to track recovery and guide rehabilitation.
• Current methods often rely on subjective clinical scales, necessitating more quantitative approaches.

Purpose of the Study:

• To evaluate sensor-based biomarkers from walking for automated estimation of post-stroke motor impairment.
• To compare different feature sets, including spatiotemporal and surface electromyographic (sEMG) data.
• To assess the impact of demographic variables and data augmentation on classification performance.

Main Methods:

• Utilized sensor-based walking data from the ARRA dataset and Hospital of Braga.
• Included 32 post-stroke individuals with varying FMA-LE scores.
• Employed a decision tree classifier with stratified six-fold cross-validation, testing various feature combinations and data augmentation techniques.

Main Results:

• The best model combined sEMG features, age, paretic side, and body mass, with noise-based data augmentation.
• Achieved a validation Matthews Correlation Coefficient (MCC) of 0.85 ± 0.16 and a test MCC of 0.70.
• sEMG features outperformed spatiotemporal features, and reduced muscle subsets yielded comparable results.

Conclusions:

• Sensor-based sEMG features acquired during walking are feasible for classifying post-stroke motor impairment.
• Feature reduction and demographic data integration can optimize model design.
• Data augmentation shows potential for enhancing model generalization, warranting further validation in larger cohorts.