An ICA-EBM-Based sEMG Classifier for Recognizing Lower Limb Movements in Individuals With and Without Knee Pathology

Area of Science:

• Biomedical Engineering
• Rehabilitation Science
• Signal Processing

Background:

• Surface electromyography (sEMG) is valuable for studying knee pathology during lower limb movements.
• Intersubject variability in sEMG signals poses a significant challenge for accurate analysis.
• Existing methods struggle to reliably differentiate between healthy and pathological knee conditions due to signal variations.

Purpose of the Study:

• To develop and validate a robust classification scheme for analyzing sEMG signals in lower limb movements.
• To overcome the challenge of intersubject variability in sEMG data for knee pathology investigation.
• To create an accurate pattern recognition system capable of distinguishing healthy subjects from those with knee pathology.

Main Methods:

• A three-step classification scheme was implemented: independent component analysis for source decomposition, time-domain feature extraction, and Fisher score-based feature selection.
• Independent Component Analysis (ICA) via entropy bound minimization was used to decompose multichannel sEMG signals.
• Linear Discriminant Analysis (LDA) was applied to dimension-reduced features selected using the Fisher score and a scree-plot technique.

Main Results:

• The developed classification scheme achieved high average classification accuracies: 96.1% for healthy subjects and 86.2% for individuals with knee pathology.
• The method demonstrated effectiveness in discriminating between different lower limb movements (walking, sitting, standing) in both healthy and pathological groups.
• Feature selection using Fisher score and scree-plot significantly improved the discrimination power of the sEMG-based analysis.

Conclusions:

• The proposed three-step classification scheme effectively addresses intersubject variability in sEMG signals for lower limb movements.
• The study presents a promising sEMG-based pattern recognition system for distinguishing knee pathology with high accuracy.
• Further improvements could lead to valuable clinical applications for diagnosing and monitoring knee conditions.