Deep learning-based BCI for gait decoding from EEG with LSTM recurrent neural network

Abstract

Objective. Mobile Brain/Body Imaging (MoBI) frameworks allowed the research community to find evidence of cortical involvement at walking initiation and during locomotion. However, the decoding of gait patterns from brain signals remains an open challenge. The aim of this work is to propose and validate a deep learning model to decode gait phases from Electroenchephalography (EEG). Approach. A Long-Short Term Memory (LSTM) deep neural network has been trained to deal with time-dependent information within brain signals during locomotion. The EEG signals have been preprocessed by means of Artifacts Subspace Reconstruction (ASR) and Reliable Independent Component Analysis (RELICA) to ensure that classification performance was not affected by movement-related artifacts. Main results. The network was evaluated on the dataset of 11 healthy subjects walking on a treadmill. The proposed decoding approach shows a robust reconstruction (AUC > 90%) of gait patterns (i.e. swing and stance states) of both legs together, or of each leg independently. Significance. Our results support for the first time the use of a memory-based deep learning classifier to decode walking activity from non-invasive brain recordings. We suggest that this classifier, exploited in real time, can be a more effective input for devices restoring locomotion in impaired people.

Experimental Protocol to Assess Neuromuscular Plasticity Induced by an Exoskeleton Training Session

“…Previous research that has focused on cortical activity during walking has been limited by technical constraints, e.g., the EEG was recorded either without recording synchronised muscle activity [ 15 ], or only healthy subjects [ 64 , 65 ], or using a limited number of EEG electrodes instead of a high-density EEG [ 66 ], or separately from EMG, kinematics and clinical scales [ 67 ]. In addition, the limited number of studies synchronizing EEG, EMG and kinematics recordings during overground walking with and without an exoskeleton are mainly conducted in able-bodied subjects and on treadmills [ 68 ]. To specifically overcome these technical and study design limitations and answer to our research questions: A high-density EEG set-up is required to reconstruct the brain cortical activity minimizing leakage effects, targeting the brain areas involved in the tasks (e.g., through power spectra or time-frequency analysis) and their interactions (e.g., effective connectivity and network/graph analysis [ 69 , 70 , 71 ]) The synchronous kinematics is required to detect gait phases to segment the EEG and EMG signals according to foot-strike and foot-off events and quantify gait symmetry and complexity [ 45 , 53 ] The synchronous EMG signals are required to identify antagonistic-agonistic muscular activation during gait and muscles response to exoskeleton.…”

Analysis of Human Gait Using Hybrid EEG-fNIRS-Based BCI System: A Review

“…The ASR algorithm is usually sufficient for removing the physiological noise of large amplitude, such as large amplitude movements, ocular artifacts, and typical muscle burst in EEG signals (Zhang et al, 2017;Contreras-Vidal et al, 2018;Tortora et al, 2020). Bulea et al (2015) used ASR to remove high amplitude artifact from the EEG recorded for speed control during walking.…”

“…Ghonchi et al (2020) used a combination of convolutional (extracting spatial features) and recurrent neural networks (extracting temporal features) to achieve an accuracy of 99.63% with the proposed model. Tortora et al (2020) used LSTM deep neural network to differentiate between swing and stance states for both individuals and combine leg movements. Similarly, spatiospectral representation learning (DNN topology) is used to differentiate between four walking conditions using EEG signals (Goh et al, 2018).…”

A review of Hidden Markov models and Recurrent Neural Networks for event detection and localization in biomedical signals

“…Tortora et al [160], 2020 BCI for gait decoding from EEG EEG data were preprocessed to remove motion artifacts through high pass filtration and independent component analysis. Different frequency bands were then extracted and a separate classifier is trained based on each frequency band.…”