Feature-Level Fusion of Surface Electromyography and Mechanomyography Signals for MVC-Normalized Shoulder Abduction...

Area of Science:

• Biomechanics and Movement Science
• Rehabilitation Engineering
• Biomedical Signal Processing

Background:

• Accurate upper-limb force recognition is crucial for wearable monitoring and rehabilitation.
• Combining surface electromyography (sEMG) and mechanomyography (MMG) for shoulder force classification is not fully understood.

Purpose of the Study:

• To investigate the effectiveness of combining sEMG and MMG signals for classifying shoulder abduction force levels.
• To evaluate various machine learning models for this classification task.

Main Methods:

• Ten healthy adults performed shoulder abduction at four maximum voluntary contraction (MVC)-normalized force levels (10%, 30%, 60%, 90%).
• sEMG and MMG signals were collected from the middle deltoid and processed using 500 ms windows with a 150 ms stride.
• Eight classifiers, including histogram-based gradient boosting decision tree (HGBDT), were evaluated using group-aware five-fold cross-validation.

Main Results:

• The HGBDT classifier achieved the highest accuracy (0.904±0.023) and macro-F1 score (0.911±0.018).
• Signal fusion (sEMG + MMG) improved the macro-F1 score from 0.821±0.030 (sEMG-only) and 0.771±0.015 (MMG-only) to 0.911±0.018.
• The best model demonstrated strong performance in classifying MVC-normalized shoulder force levels.

Conclusions:

• The complementary nature of sEMG and MMG signals enhances shoulder force-level classification in healthy adults.
• Further subject-independent and patient-level validation is necessary for clinical rehabilitation applications.