Simultaneous sEMG Classification of Hand/Wrist Gestures and Forces

Area of Science:

• Biomedical Engineering
• Rehabilitation Engineering
• Neuroprosthetics

Background:

• Surface electromyography (sEMG) is a promising non-invasive method for amputees to control prosthetic hands.
• Existing methods often struggle to control multiple degrees of freedom (DoFs) naturally, limiting prosthetic functionality.
• Simultaneous decoding of both hand gestures and applied forces using sEMG pattern recognition (PR) remains an underexplored area.

Purpose of the Study:

• To develop and evaluate a hierarchical classification approach for decoding motor intentions in prosthetic hand control.
• To simultaneously assess desired hand/wrist gestures and force levels for grasping tasks.
• To compare the performance of Non-Linear Logistic Regression (NLR) with Linear Discriminant Analysis (LDA) for this application.

Main Methods:

• A hierarchical classification architecture managed by a Finite State Machine was implemented.
• Three Non-Linear Logistic Regression (NLR) classifiers were used to decode hand/wrist gestures and force levels.
• The system was evaluated on 31 healthy subjects, with results validated against Linear Discriminant Analysis (LDA).

Main Results:

• The hand/wrist gesture classifier achieved a mean accuracy of 98.78% for seven distinct gestures.
• The force classifier demonstrated high accuracy, with 98.80% for spherical grasp force and 96.09% for tip grasp force.
• Statistical analysis (Wilcoxon Signed-Rank test) showed no significant difference in F1-score performance between NLR and LDA.

Conclusions:

• The proposed hierarchical classification system effectively decodes multiple hand/wrist gestures and associated force levels using sEMG.
• Non-Linear Logistic Regression (NLR) is a viable and effective alternative to benchmark LDA for EMG pattern recognition.
• This approach offers a pathway towards more intuitive and functional control of multifunctional prosthetic hands.