Classification of naturally evoked compound action potentials in peripheral nerve spatiotemporal recordings

Abstract

Peripheral neural signals have the potential to provide the necessary motor, sensory or autonomic information for robust control in many neuroprosthetic and neuromodulation applications. However, developing methods to recover information encoded in these signals is a significant challenge. We introduce the idea of using spatiotemporal signatures extracted from multi-contact nerve cuff electrode recordings to classify naturally evoked compound action potentials (CAP). 9 Long-Evan rats were implanted with a 56-channel nerve cuff on the sciatic nerve. Afferent activity was selectively evoked in the different fascicles of the sciatic nerve (tibial, peroneal, sural) using mechano-sensory stimuli. Spatiotemporal signatures of recorded CAPs were used to train three different classifiers. Performance was measured based on the classification accuracy, F 1 -score, and the ability to reconstruct original firing rates of neural pathways. The mean classification accuracies, for a 3-class problem, for the best performing classifier was 0.686 ± 0.126 and corresponding mean F 1 -score was 0.605 ± 0.212. The mean Pearson correlation coefficients between the original firing rates and estimated firing rates found for the best classifier was 0.728 ± 0.276. The proposed method demonstrates the possibility of classifying individual naturally evoked CAPs in peripheral neural signals recorded from extraneural electrodes, allowing for more precise control signals in neuroprosthetic applications.

Tutorial: a guide to techniques for analysing recordings from the peripheral nervous system

“…The concept of a neural spatiotemporal signature was first introduced in [22], [61], [83]. This technique involves extracting neural recordings from a group of fibres associated with a particular function (e.g., a motor command to a single muscle) using a 2D array of electrode contacts (longitudinally and circumferentially).…”

“…In acute surgical experiments, the electrodes (recording, stimulation, or both) are implanted in anesthetised animals and the target nerve is explored over the course of a few hours whilst the animal remains unconscious [17], [25], [72], [83], [91], [92]. The animal is then terminated at the end of the experiment.…”

Advancing Peripheral Nerve Interfaces in a Large Animal Model

“…The spatiotemporal approach was further validated in acute experiments using naturally evoked CAPs and four classification techniques: (i) matched filter, (ii) random forest, (iii) neural net, and (iv) random forest applied to signals pre-processed using rectified-bin-integration in different lengths of time-window. Similar to simulated data, the acute experiments indicated that the spatiotemporal approach could outperform BSF and velocity-selective recording approaches [79]. However, a sampling frequency of 30 kHz was used for both studies, which may affect the velocity-selective recording performance.…”