Journal: IEEE transactions on neural systems and rehabilitation engineering : a publication of the IEEE Engineering in Medicine and Biology Society

Article Title: Validity and Impact of Methods for Collecting Training Data for Myoelectric Prosthetic Control Algorithms

doi: 10.1109/TNSRE.2024.3400729

Figure Lengend Snippet: Two approaches to collecting training data for supervised machine-learning algorithms for myoelectric prostheses. A) With mimic training, the user watches a prosthesis move through pre-programmed movements and attempts to mimic the movement of the prosthesis with their phantom limb. The resulting training dataset consists of EMG from the residual muscles of the phantom hand and the pre-programmed kinematics of the prosthesis. B) With mirror training, the user performs bilaterally mirrored movements, such that the motion of their intact contralateral hand mirrors that of their phantom limb. The resulting training dataset consists of EMG from the residual muscles of the phantom hand and the mirrored kinematics of the contralateral limb (determined by motion capture).

Article Snippet: In the present study, the prosthesis control algorithms were trained in a biomimetic way, such that the users intended kinematics were consistent with the prosthesis action.

Techniques: Muscles

Journal: IEEE transactions on neural systems and rehabilitation engineering : a publication of the IEEE Engineering in Medicine and Biology Society

Article Title: Validity and Impact of Methods for Collecting Training Data for Myoelectric Prosthetic Control Algorithms

doi: 10.1109/TNSRE.2024.3400729

Figure Lengend Snippet: Performance of amputee participants completing a clothespin relocation task (CRT) with a myoelectric prosthesis trained with mirror or mimic training. A) Participants complete the CRT faster with mirror training. B) Success rate was similar for mirror and mimic training (out of 39 attempts for mimic training and out of 42 attempts for mirror training). C) The cognitive load associated with mirror and mimic training is also similar both during training and online use. D) There were also no significant differences between embodiment scores associated with mirror and mimic training. Data from four unilateral transradial amputees for A and B. Data from three unilateral transradial amputees for C & D. Violin plots show the kernel density estimation. Black horizontal lines denote the mean, white circles denote the median, and vertical gray lines denote the interquartile range. * p < 0.05, Wilcoxon rank-sum test.

Article Snippet: In the present study, the prosthesis control algorithms were trained in a biomimetic way, such that the users intended kinematics were consistent with the prosthesis action.

Techniques: