Assessing workload in using electromyography (EMG)-based prostheses

Area of Science:

• Biomedical Engineering
• Human-Computer Interaction
• Rehabilitation Engineering

Background:

• Prosthetic device use imposes a significant cognitive load on users.
• Accurate assessment of cognitive workload is crucial for optimizing prosthetic design and usability.

Purpose of the Study:

• To investigate and compare classification models for assessing cognitive workload in electromyography (EMG)-based prosthetic devices.
• To identify key input features that best predict cognitive workload.

Main Methods:

• Utilized electromyography (EMG) signals, eye-tracking measures, task performance, and cognitive performance model (CPM) outcomes as input features.
• Applied feature selection algorithms, hyperparameter tuning (grid search), and k-fold cross-validation for model optimization.
• Evaluated model performance using classification accuracy, AUC, precision, recall, and F1 scores.

Main Results:

• Task performance measures, pupillometry data, and CPM outcomes were identified as highly informative features.
• Naïve Bayes (NB) and Random Forest (RF) algorithms demonstrated the most promising performance in classifying cognitive workload.
• The developed models achieved high accuracy with low computational cost.

Conclusions:

• Machine learning models, particularly NB and RF, can effectively classify cognitive workload in EMG-based prosthetics.
• These models can assist manufacturers and clinicians in predicting cognitive workload during early design phases.
• The findings support the use of these algorithms for assessing prosthetic device usability.