Investigating the Impact of Training Protocols on Myoelectric Pattern Recognition Control in Upper-Limb Amputees.

Myoelectric control schemes, pivotal in the control of prosthetic limbs, are often developed and evaluated in ideal laboratory conditions. However, these controlled environments may not fully represent the diverse challenges users face in real-world scenarios. The present study aims to tackle some of the existing research limitations by exploring the influence of various model training protocols on myoelectric pattern recognition within a semi-autonomous control system, which has been shown to reduce user cognitive load and enhance overall system performance. Specifically, we focus on the effects of limb movement and weight-bearing activities. We investigate the effect of four distinct training protocols in pattern recognition control for upper-limb prostheses, including training without a prosthetic hand, training with a prosthetic hand and static gestures, training with a prosthetic hand and dynamic movements guided by a graphical user interface (GUI), and training with a prosthetic hand having dynamic transfers and unguided. By examining these conditions, we aim to provide an understanding of how different training protocols and different labeling methods influence myoelectric pattern recognition control. Our results, based on experiments conducted with 14 non-disabled and one amputee participant, suggest that introducing the weight of the prosthetic hand and dynamic movements of the arm to the training data improves the accuracy and robustness of the control scheme. Real-time control experiments with a group of five non-disabled and one amputee participant using a multi-DOF prosthetic hand also verify our findings.