Myoelectric walking mode classification for transtibial amputees

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Human-Computer Interaction

Background:

• Myoelectric control offers potential for prosthetic limbs to adapt to user intent and diverse walking conditions.
• Developing accurate walking mode classifiers is crucial for intuitive prosthetic control in transtibial amputees.

Purpose of the Study:

• To develop and evaluate a myoelectric walking mode classifier for transtibial amputees.
• To assess the performance of linear discriminant analysis (LDA) and support vector machine (SVM) classifiers using myoelectric signals.
• To investigate the stability of the classifier under electrode shift disturbances.

Main Methods:

• Myoelectric signals were recorded from four leg muscles (tibialis anterior, medial gastrocnemius, vastus lateralis, biceps femoris) in nonamputee and transtibial amputee subjects.
• Signals were processed into features (mean absolute value, variance, wavelength, slope sign changes, zero crossings) across gait cycle subwindows.
• LDA and SVM algorithms were employed to classify seven walking modes: level ground (3 speeds), ramp ascent/descent, and stair ascent/descent.

Main Results:

• The classifier achieved high accuracy: 97.9% for amputees and 94.7% for nonamputees across all walking modes.
• Stair ascent/descent exhibited the highest classification accuracy (99.8% for amputees, 100.0% for nonamputees).
• Electrode shift of the medial gastrocnemius significantly impacted classification accuracy; SVM did not outperform LDA for this dataset.

Conclusions:

• The developed myoelectric classifier effectively distinguishes between various walking modes in transtibial amputees.
• The classifier's robustness is influenced by electrode placement, particularly for the medial gastrocnemius.
• This algorithm represents a significant step towards adaptive and intuitive prosthetic leg control.