A study on interfacing with the peripheral nervous system for the development of sensory prostheses

Abstract

The work presented in this thesis focuses on current technological limitations and knowledge gaps faced when interfacing with the peripheral nervous system (PNS). Specifically, this thesis was performed in the context of the SenseBack project which aimed to investigate new methodologies to interface with peripheral nerves and provide sensory information to upper-limb prosthetic users. The main challenges addressed in this work include examining the compromise between complexity, acuity, and invasiveness of peripheral nerve interfaces; and investigating how to use signal processing and stimulation protocols to improve the information that can be transmitted or received by these interfaces. The original contributions of this thesis are presented in the form of three acute in-vivo studies. The first in-vivo study investigated the influence of the implant position and the number of recording channels on the discrimination of sensory neural signals. Two multi-channel nerve cuffs were implanted concurrently on the sciatic nerve of rats to record the afferent signals in response to mechanical stimulation of the hindlimb. The results showed that the discrimination task was easier when the cuff was implanted distally, taking advantage of the fasciculation of the nerve. The results also showed that the implant location played a more significant role than the electrode channel count in discriminating the sensory signals. Thus, this study emphasises the importance of considering the underlying nerve anatomy before designing and implanting a nerve cuff electrode. Following these findings, more complex signal analysis techniques were explored to further improve the classification accuracy of the sensory signals collected in the first in-vivo study. A novel feature extraction framework, that incorporates spatio-temporal focus and dynamic time warping, achieved high performance accuracies while keeping a low computational cost. This framework outperformed the remaining frameworks tested in the study and has improved the discrimination accuracy of the sensory signals. Thus, this study contributes by extending the tools available to extract features from sensory population activity electroneurographic (ENG) signals