Classification of human movements with and without spinal orthosis based on surface electromyogram signals

Abstract

Spinal orthoses were designed to correct poor posture; however, they may restrict trunk movements at all times, making daily activities difficult. Detecting trunk movements can provide instructions for adjusting the stiffness of the spinal orthosis. This study evaluated the feasibility of identifying movements based on surface electromyography (sEMG) signals. Ten participants were tested for different movements with two different modalities: motion without the spinal orthosis (Normal) and with the spinal orthosis (Spinal orthosis). The sEMG signals were collected from eight muscles using surface electrodes during four movements [flexion-extension, lateral bending, axial rotation, and stand to sit to stand]. Four time domain features were extracted, with a total of 32 feature vectors. The principal component analysis (PCA) method was adopted to feature selection, and it was found that eight feature dimensions can make cumulative explained variance exceed 95%. The results showed that machine learning algorithms could not only identify Normal and Spinal orthosis movement modalities, but also distinguish four daily movements. Moreover, the classification performance of Random Forest (RF), k-Nearest Neighbor (k-NN), and Support Vector Machine (SVM) algorithms were also compared. The results showed that all three machine algorithms have high classification accuracy. The machine learning methods can accurately identify movement patterns by considering sEMG signals, which may provide instructions for adjusting the stiffness of the spinal orthosis. In the future, the spinal orthosis with adjustable stiffness controlled by sEMG signals could help correct poor posture, and permit the wearer to achieve free movement when needed.

Keywords

• Surface electromyogram
• Machine learning
• Spinal orthosis
• Trunk movements
• Classification