Brain-inspired spiking neural networks for decoding and understanding muscle activity and kinematics from electroencephalography signals during hand movements

Kaushalya Kumarasinghe,Nikola Kasabov,Denise Taylor

doi:10.1038/s41598-021-81805-4

Kaushalya Kumarasinghe, Nikola Kasabov + Show 1 more

Open Access

https://doi.org/10.1038/s41598-021-81805-4

Copy DOI

Abstract

Compared to the abilities of the animal brain, many Artificial Intelligence systems have limitations which emphasise the need for a Brain-Inspired Artificial Intelligence paradigm. This paper proposes a novel Brain-Inspired Spiking Neural Network (BI-SNN) model for incremental learning of spike sequences. BI-SNN maps spiking activity from input channels into a high dimensional source-space which enhances the evolution of polychronising spiking neural populations. We applied the BI-SNN to predict muscle activity and kinematics from electroencephalography signals during upper limb functional movements. The BI-SNN extends our previously proposed eSPANNet computational model by integrating it with the ‘NeuCube’ brain-inspired SNN architecture. We show that BI-SNN can successfully predict continuous muscle activity and kinematics of upper-limb. The experimental results confirmed that the BI-SNN resulted in strongly correlated population activity and demonstrated the feasibility for real-time prediction. In contrast to the majority of Brain–Computer Interfaces (BCIs) that constitute a ‘black box’, BI-SNN provide quantitative and visual feedback about the related brain activity. This study is one of the first attempts to examine the feasibility of finding neural correlates of muscle activity and kinematics from electroencephalography using a brain-inspired computational paradigm. The findings suggest that BI-SNN is a better neural decoder for non-invasive BCI.

Highlights

Compared to the abilities of the animal brain, many Artificial Intelligence systems have limitations which emphasise the need for a Brain-Inspired Artificial Intelligence paradigm
The primary goal of this study is to evaluate the feasibility of Brain-Inspired Spiking Neural Networks to construct a novel, interpretable neural decoder which can incrementally learn to predict an upcoming movement from EEG signals
The results suggest that the Brain-Inspired Spiking Neural Network (BI-Spiking Neural Networks (SNN)) is able to predict each individual motor signal more accurately than the pure eSPANNet or Generalised Linear Model (GLM) models

Summary

Introduction

Compared to the abilities of the animal brain, many Artificial Intelligence systems have limitations which emphasise the need for a Brain-Inspired Artificial Intelligence paradigm. We applied the BI-SNN to predict muscle activity and kinematics from electroencephalography signals during upper limb functional movements. This study is one of the first attempts to examine the feasibility of finding neural correlates of muscle activity and kinematics from electroencephalography using a brain-inspired computational paradigm. Conventional neural decoders that utilise the sensorimotor rhythms of electroencephalography (EEG) generate distinct neural commands through Event-related Synchronisation/ Desynchronisation evoked as a result of moving different parts of the body This results in un-naturalistic control when applied to neurorehabilitation due to the cognitive disconnection between the targeted and intended action. Several recent studies report the feasibility of extracting neuro-muscular interactions from EEG during functional upper limb movements such as grasp and lift. Synergies from EEG during upper limb reaching tasks using unified independent component analysis. these studies provide promising empirical results on extracting neural signals from EEG useful to control and manipulate objects through BCIs

Objectives

Methods

Results

Conclusion