A Method for Locomotion Mode Identification Using Muscle Synergies

• Published in: IEEE transactions on neural systems and rehabilitation engineering Vol. 25; no. 6; pp. 608 - 617
• Main Authors: Afzal, Taimoor, Iqbal, Kamran, White, Gannon, Wright, Andrew B.
• Format: Journal Article
• Language: English
• Published: United States IEEE 01.06.2017; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Amputation; Artificial neural networks; Ascent; Classification; Computer Simulation; Descent; Discriminant Analysis; Electromyography; Electromyography - methods; Feature extraction; Female; Gait - physiology; Humans; Intent recognition; Legged locomotion; Limbs; Locomotion; Locomotion - physiology; Lower Extremity - physiology; Male; Matrices (mathematics); Mechanical sensors; Models, Biological; Muscle Contraction - physiology; muscle synergies; Muscle, Skeletal - physiology; Muscles; Neural networks; Pattern Recognition, Automated - methods; Prostheses; Prosthetics; Reproducibility of Results; Sensitivity and Specificity; Steady state; Support Vector Machine; Support vector machines; Toe; Walking; Young Adult
• ISSN: 1534-4320; 1558-0210; 1558-0210
• DOI: 10.1109/TNSRE.2016.2585962

Active lower limb transfemoral prostheses have enabled amputees to perform different locomotion modes such as walking, stair ascent, stair descent, ramp ascent and ramp descent. To achieve seamless mode transitions, these devices either rely on neural information from the amputee's residual limbs or sensors attached to the prosthesis to identify the intended locomotion modes or both. We present an approach for classification of locomotion modes based on the framework of muscle synergies underlying electromyography signals. Neural information at the critical instances (e.g., heel contact and toe-off) was decoded for this purpose. Non-negative matrix factorization was used to extract the muscles synergies from the muscle feature matrix. The estimation of the neural command was done using non-negative least squares. The muscle synergy approach was compared with linear discriminant analysis (LDA), support vector machine (SVM), and neural network (NN) and was tested on seven able-bodied subjects. There was no significant difference (p > 0.05) in transitional and steady state classification errors during stance phase. The muscle synergy approach performed significantly better (p <; 0.01) than NN and LDA during swing phase while results were similar to SVM. These results suggest that the muscle synergy approach can be used to discriminate between locomotion modes involving transitions.