TY - JOUR
T1 - The effect of arm position on classification of hand gestures with intramuscular EMG
AU - Jochumsen, Mads
AU - Waris, Asim
AU - Kamavuako, Ernest Nlandu
PY - 2018
Y1 - 2018
N2 - The arm position affects discrimination between upper limb motion classes when using surface EMG (sEMG). In this study, the effect of arm position on motion class discrimination was investigated using intramuscular EMG (iEMG). Eight able-bodied subjects performed five motion classes (hand grasp, hand open, rest, wrist extension, wrist flexion) in four different arm positions (0, 45, 90, 135°). Three classification scenarios were evaluated using Hudgins’ time domain features and a Bayes classifier; within position classification (WPC), across position classification (APC), and between position classification (BPC). The same analysis was performed using sEMG and with combined surface and iEMG. For WPC, similar classification accuracies were obtained using the different types of EMG (93–98%). The mean absolute value and waveform length were associated with the highest classification accuracies compared to zero crossing and slope sign changes for WPC. For APC, classification accuracies dropped to 85–95%, and for BPC, classification accuracies dropped to 69–83% with hand opening being the least discriminable motion class. The degree of decreased performance was computed as: 1) APC/WPC: 0.94 ± 0.03 (sEMG) and 0.92 ± 0.05 (iEMG), and 2) BPC/WPC: 0.81 ± 0.06 (sEMG) and 0.78 ± 0.12 (iEMG), indicating that arm position affects iEMG in a similar degree as sEMG, which is a practicality issue for the clinical application of pattern recognition based control schemes.
AB - The arm position affects discrimination between upper limb motion classes when using surface EMG (sEMG). In this study, the effect of arm position on motion class discrimination was investigated using intramuscular EMG (iEMG). Eight able-bodied subjects performed five motion classes (hand grasp, hand open, rest, wrist extension, wrist flexion) in four different arm positions (0, 45, 90, 135°). Three classification scenarios were evaluated using Hudgins’ time domain features and a Bayes classifier; within position classification (WPC), across position classification (APC), and between position classification (BPC). The same analysis was performed using sEMG and with combined surface and iEMG. For WPC, similar classification accuracies were obtained using the different types of EMG (93–98%). The mean absolute value and waveform length were associated with the highest classification accuracies compared to zero crossing and slope sign changes for WPC. For APC, classification accuracies dropped to 85–95%, and for BPC, classification accuracies dropped to 69–83% with hand opening being the least discriminable motion class. The degree of decreased performance was computed as: 1) APC/WPC: 0.94 ± 0.03 (sEMG) and 0.92 ± 0.05 (iEMG), and 2) BPC/WPC: 0.81 ± 0.06 (sEMG) and 0.78 ± 0.12 (iEMG), indicating that arm position affects iEMG in a similar degree as sEMG, which is a practicality issue for the clinical application of pattern recognition based control schemes.
KW - Arm position
KW - Intramuscular EMG
KW - Prosthetics
KW - Surface EMG
KW - Upper extremity
UR - http://www.scopus.com/inward/record.url?scp=85042731496&partnerID=8YFLogxK
U2 - 10.1016/j.bspc.2018.02.013
DO - 10.1016/j.bspc.2018.02.013
M3 - Journal article
SN - 1746-8094
VL - 43
SP - 1
EP - 8
JO - Biomedical Signal Processing and Control
JF - Biomedical Signal Processing and Control
ER -