Machine-learning myoelectric prosthesis control: towards the advancement of assessing functional use and control skill

The technological state-of-the-art in upper limb prostheses has demonstrated impressive and rapid advancements. However, the introduction of these technologies has opened up new questions in two domains of the assessment paradigms – namely, whether the current clinical tests of prosthetic upper limb function are still adequate to fully assess these prostheses and how the user skill in controlling these devices can be quantified. The aim of this thesis was to investigate and to advance the assessment paradigms in machine-learning based myoelectric upper limb prostheses, with a focus on the assessment of functional prosthesis use and on the assessment of the user skill in controlling the prosthesis. For this purpose, this thesis investigates how state-of-the-art hand prostheses are used and which potential advantages (and disadvantages) come with the introduction of multi-function prosthetic hands and machine-learning based control interfaces. Subsequently, this thesis outlines how these technologies should be evaluated and how this could be realized within the constraints of clinical test routines for functional prosthesis use. Furthermore, this thesis investigates how characteristics of the users’ muscle activation patterns relate to the users’ control performance of machine-learning based myoelectric prostheses. Subsequently, this thesis investigates potential reasons why a common set of such muscle activation pattern characteristics paradoxically falls short of showing a strong association with the control performance. Finally, this thesis discusses the wider implications of the findings with regard to the assessment paradigm and the conceptual assumptions in machine-learning myoelectric prosthesis control.