Evaluating the influence of subject-related variables on EMG-based hand gesture classification

In this study we evaluated the effect of subjectrelated variables, i.e. hand dominance, gender and experience in using, on the performances of an EMG-based system for virtual upper limb and prosthesis control. The proposed system consists in a low density EMG sensors arrangement, a purpose-built signal-conditioning electronic circuitry and a software able to classify the gestures and to replicate them via avatars. The classification algorithm was optimized in terms of feature extraction and dimensionality reduction. In its optimal configuration, the system allows to accurately discriminate five different hand gestures (accuracy = 88.85 ± 7.19%). Statistical analysis demonstrated no significant difference in classification accuracy related to hand-dominance (handedness) and to gender. In addition, maximum accuracy in dominant hand is achieved since first use of the system, whilst accuracy in classifying gestures of the non-dominant hand significantly increases with experience. These results indicate that this system can be potentially used by every trans-radial upper-limb amputee for virtual/real limb control.