Elsevier

Clinical Neurophysiology

Volume 126, Issue 1, January 2015, Pages 154-159
Clinical Neurophysiology

A brain–computer interface for single-trial detection of gait initiation from movement related cortical potentials

https://doi.org/10.1016/j.clinph.2014.05.003Get rights and content

Highlights

  • Accurate single trial detection of the intention of step initiation from scalp EEG.

  • Independent component analysis (ICA) preprocessing helps to automatically remove EEG artifacts and enhances detection performance.

  • All participating subjects were BCI/EEG naïve subjects, implying general applicability of the proposed approach.

Abstract

Objective

Applications of brain–computer interfacing (BCI) in neurorehabilitation have received increasing attention. The intention to perform a motor task can be detected from scalp EEG and used to control rehabilitation devices, resulting in a patient-driven rehabilitation paradigm. In this study, we present and validate a BCI system for detection of gait initiation using movement related cortical potentials (MRCP).

Methods

The templates of MRCP were extracted from 9-channel scalp EEG during gait initiation in 9 healthy subjects. Independent component analysis (ICA) was used to remove artifacts, and the Laplacian spatial filter was applied to enhance the signal-to-noise ratio of MRCP. Following these pre-processing steps, a matched filter was used to perform single-trial detection of gait initiation.

Results

ICA preprocessing was shown to significantly improve the detection performance. With ICA preprocessing, across all subjects, the true positive rate (TPR) of the detection was 76.9 ± 8.97%, and the false positive rate was 2.93 ± 1.09 per minute.

Conclusion

The results demonstrate the feasibility of detecting the intention of gait initiation from EEG signals, on a single trial basis.

Significance

The results are important for the development of new gait rehabilitation strategies, either for recovery/replacement of function or for neuromodulation.

Introduction

Neurological conditions, such as stroke, spinal cord injury or Parkinson’s disease, often result in impaired motor control and consequent difficulty of the patient to perform activities of daily living. One of the goals of rehabilitation is to promote the patient’s independency with the aim of restoring the loss of movement ability.

Conventional approaches of rehabilitation promote motor recovery through a “bottom-up” approach, focused on peripheral training, often with robotic trainers. Robotic training has several advantages (a reduction of the effort of physical therapists per patient, the possibility to objectively quantify rehabilitation parameters and training output) (Pennycott et al., 2012) and allows for peripheral activity compatible with unconstrained tasks (Gizzi et al., 2012). However its effectiveness may also be reduced by the autonomous ability of the robot to complete the movement without the need for patient involvement. Active participation of the patient has been demonstrated to be crucial in improving the outcome of rehabilitation (Pennycott et al., 2012, Duff et al., 2013).

As a complementary and promising branch within motor rehabilitation and assistance are brain–computer interfaces (BCI). BCI technologies provide the means for conveying control commands directly from the brain and can be used either for directly controlling rehabilitation devices (function recovery or replacement) or to provide feedback to the patient based on his/her brain activity (neuromodulation). In the latter case, the patient is actively involved in the rehabilitation process. The feedback is provided by the action of rehabilitation devices (e.g., the movement of an orthotics system) triggered by the brain activity (brain switch).

When the brain activation related to motor intention is measured using non-invasive EEG, the information carried in different frequency bands may be extracted, interpreted and used as the command signal to external devices. These strategies include sensory motor rhythms (SMR), on which most past studies on BCI for neuromodulation have focused (Neuper et al., 2006, Kaiser et al., 2011, Ramos-Murguialday et al., 2013). A disadvantage of this approach, however, is the need for numerous training sessions until the user is able to control the signal adequately. Alternatively, movement related cortical potentials (MRCP) have also been proposed for detecting motor intention from EEG. MRCP is a slow cortical potential that occurs naturally as a person commences or imagines the start of a movement (Gangadhar et al., 2009, Niazi et al., 2011, Garipelli et al., 2013, Xu et al., 2014). The advantage of this approach is that no extensive prior training of the user is required. Moreover, MRCPs can also be used to discriminate between different types of tasks as well as the way a task is executed (Do Nascimento et al., 2008, Gu et al., 2009). One potential confounding factor is that the size of the MRCP is relatively small (∼10 μV) and is prone to many movement artifacts that influence the EEG measures.

MRCPs have been studied during gait initiation, with focus on Parkinsonian patients (Vidailhet et al., 1995, Shoushtarian et al., 2011). Moreover, the study by Do Nascimento et al. (2005) on healthy subjects demonstrated that MRCPs contain rich information regarding gait initiation, which made a strong case for utilizing MRCPs for detecting the intention of gait initiation. However, the ability to detect MRCPs depends on the signal quality and the presence of artifacts, such as due to eye movements or to facial muscle contractions that can significantly affect the performance and robustness of a BCI detection system. This study aims at investigating the possibility of detecting the intention of gait initiation from MRCPs after artifacts were removed in a semi-automatic way. We focused on the step initiation in the forward direction, as it is most relevant for the targeted application. The main objective is to develop and test a brain switch based on the intention to initiate locomotion and, in future developments, to integrate this brain switch into non-ambulatory robotic systems for rehabilitation of walking to promote plasticity in stroke patients (Belda-Lois et al., 2011).

Section snippets

Subjects

Nine subjects (M6, F3, 21–38 yrs), denoted by SUB1–SUB9, participated in the experiment. No subject had any known neurological disorders. Except for SUB5, all other subjects had no prior experience with BCI systems before the experiment, and were thus considered as naïve BCI subjects. The experiment protocol was approved by the research ethics committee of the University Medical Center Göttingen.

Experimental protocol

An active EEG electrode system (activCap, Brainproducts GmbH) was used in all the experiments. The

Results

All subjects successfully completed the experimental session. A representative example of the ICA artifact rejection procedure is presented for one of the subjects in Fig. 2. By properly selecting independent components both from the time course and the scalp map, the artifacts within the EEG were effectively removed. The ICA algorithm (FastICA) converged for all but one subject (SUB6), in which case the original raw EEG was used for subsequent processing.

The MRCP templates of all subjects,

Discussion

We demonstrated the possibility of single trial detection of the intention of gait initiation from scalp EEG of healthy individuals during normal gait. Across subjects, the best performance corresponded to TPR 83% and FPR 1.64 per min. The average detection performance was similar to prior studies on isometric dorsiflexion detection when subjects were in seated position (Niazi et al., 2011). The experimental condition (standing and gait) in the current study was more challenging than the

Acknowledgment

This work was supported and by the EU project BETTER (contract # 247935) (to D.F.).

References (33)

  • J.B. Toledo et al.

    High beta activity in the subthalamic nucleus and freezing of gait in Parkinson’s disease

    Neurobiol Dis

    (2014)
  • J.-M. Belda-Lois et al.

    Rehabilitation of gait after stroke: a review towards a top-down approach

    J Neuroeng Rehabil

    (2011)
  • B. Blankertz et al.

    The Berlin brain–computer interface: non-medical uses of BCI technology

    Front Neurosci

    (2010)
  • L. Deecke et al.

    Voluntary finger movement in man: cerebral potentials and theory

    Biol Cybern

    (1976)
  • M. Duff et al.

    Adaptive mixed reality rehabilitation improves quality of reaching movements more than traditional reaching therapy following stroke

    Neurorehabil Neural Repair

    (2013)
  • Gangadhar G, Chavarriaga R, Mill R. Anticipation Based Brain-Computer Interfacing (aBCI). 4th Int. IEEE EMBS Conf....
  • Cited by (110)

    • Towards a Gait Planning Training Strategy Using Lokomat

      2024, Lecture Notes in Networks and Systems
    View all citing articles on Scopus
    View full text