Understanding the Impact of Closed-Loop  Neurofeedback on Neurophysiological Features and  BCI Performance

Shay Englander

Key information

Authors:

Shay Englander (Shay Englander)

Supervisors:

Patrícia Margarida Piedade Figueiredo (Patrícia Figueiredo); Athanasios Vourvopoulos (Athanasios Vourvopoulos)

Published in

November 7, 2024

Abstract

Brain-Computer Interfaces (BCIs) enable direct communication between the brain and external devices, holding great promise for rehabilitation and assistive technologies. However, a persistent challenge is the discrepancy in the BCI performance between training (offline) and closed-loop feedback control, due to the non-stationary nature of EEG signals and their susceptibility to noise and artifacts. This study aims to investigate the effects of closed-loop BCI systems and different virtual feedback modalities on the neurophysiological features, for example, Event-Related Desynchronization (ERD), and the performance changes of the BCI during binary motor imagery (MI) tasks. Fifteen healthy participants performed MI tasks using a 32-electrode electroencephalography (EEG) setup in a virtual reality (VR) environment designed to induce embodiment. The study evaluated classification accuracy using a machine learning model across training, and two virtual feedback conditions, revealing no significant differences between the phases or feedback modalities, suggesting that the protocol effectively maintained consistent performance in closed-loop feedback settings. Feature selection analysis highlighted that many of the most discriminative features were outside the traditional ERD-related regions, indicating the value of exploring non-traditional EEG features for MI task discrimination. Additionally, ERD levels were consistently induced across all conditions, with no significant differences observed between virtual feedback modalities, demonstrating the protocol’s robustness in eliciting expected neural markers for MI tasks. Overall, this study contributes to a better understanding of the changes in neurophysiological properties and classification accuracies under closed-loop BCI conditions and provides insights for future studies in the field of neurorehabilitation.