A functional source separation algorithm to enhance error-related potentials monitoring in noninvasive brain-computer interface

https://doi.org/10.1016/j.cmpb.2020.105419Get rights and content

Highlights

  • Semi Blind functional source separation (FSS) identify optimal spatial filter for BCI.

  • FSS algorithm is able to enhance error-related potential (ErrPs) monitoring in non-invasive BCI.

  • Bayesian linear classification shows higher accuracy for FSS respect to single EEG electrode.

  • Bayesian linear classification shows higher accuracy for FSS respect to xDAWN spatial filter.

Abstract

Background and objectives

An Error related Potential (ErrP) can be noninvasively and directly measured from the scalp through electroencephalography (EEG), as response, when a person realizes they are making an error during a task (as a consequence of a cognitive error performed from the user). It has been shown that ErrPs can be automatically detected with time-discrete feedback tasks, which are widely applied in the Brain-Computer Interface (BCI) field for error correction or adaptation. In this work, a semi-supervised algorithm, namely the Functional Source Separation (FSS), is proposed to estimate a spatial filter for learning the ErrPs and to enhance the evoked potentials.

Methods

EEG data recorded on six subjects were used to evaluate the proposed method based on FFS algorithm in comparison with the xDAWN algorithm. FSS- and xDAWN-based methods were compared also to the Cz and FCz single channel. Single-trial classification was considered to evaluate the performances of the approaches. (Both the approaches were evaluated on single-trial classification of EEGs.)

Results

The results presented using the Bayesian Linear Discriminant Analysis (BLDA) classifier, show that FSS (accuracy 0.92, sensitivity 0.95, specificity 0.81, F1-score 0.95) overcomes the other methods (Cz - accuracy 0.72, sensitivity 0.74, specificity 0.63, F1-score 0.74; FCz - accuracy 0.72, sensitivity 0.75, specificity 0.61, F1-score 0.75; xDAWN - accuracy 0.75, sensitivity 0.79, specificity 0.61, F1-score 0.79) in terms of single-trial classification.

Conclusions

The proposed FSS-based method increases the single-trial detection accuracy of ErrPs with respect to both single channel (Cz, FCz) and xDAWN spatial filter.

Introduction

Brain-Computer Interface (BCI) is a noninvasive technology that enables communication between the user's brain and a digital device (e.g. smart wheelchairs, computers, or prosthesis), usually named agent. BCI allows the recognition of the user's intention by decoding the neural activity through electroencephalography (EEG) in order to control the agent and improve its performances. Reaching this goal implies high cognitive attention and effort, since the user is asked a considerable concentration on the stimuli provided while operating a BCI.

In literature, many works analyzed the capability of the BCI to recognize erroneous behavior of agents directly from the user's brain signals [1], [2], [3], [4], [5], [6], [7]. Errors in the recognition of the user's intention elicit potentials called evoked Error-related EEG Potentials (ErrPs). ErrPs were analyzed for the first time in 1990, in a study about choice-reaction tasks [8]; in the same work, the typical ErrPs waveform was defined. In this paradigm, the user monitors the agent's actions providing a feedback that can be used to improve the overall performance of the agent. ErrP is physiologically defined as a two components brain signal, consisting in negative and positive peaks, associated with the response monitoring and error detection processes. Both peaks originate in the anterior cingulate cortex, a frontal brain structure involved in the cognitive and affective brain processes [9]. Typically, the signal is generated within 500 ms from the erroneous agent decision, where the first component is a negative peak at almost 50–100 ms. After the negative peak, a positive peak is generated, further divided into fronto-central and centroparietal components [10].

The works [1,2] provide and explain some examples where ErrP signals are generated when a user monitors the performance of an agent, without performing a direct control. Unlike traditional BCI systems, the user does not provide continuous commands, but only monitors the agent's performance, thus making possible to tailor the agent's behavior to the user's needs and preferences [3]. Furthermore, in the experimental protocol proposed in [1], the user tried to move a cursor towards a target location (either using a keyboard or mental commands). Moreover, it showed the possibility to recognize and correct an erroneous decision of the agent exploiting the EEG signals.

The application of ErrPs in BCI technology has increased during the last years, especially for the correction of the system behavior through what is called reinforcement learning. Precisely, the most common application of ErrP was done in BCI spellers, where during the spelling of a word, a character can be discarded if wrong [4]. ErrPs can be involved as a suitable alternative or a complementary signal for BCI systems, especially as supervision or feedback signal during the execution of the task [7]. ErrPs have turned out to be used for fixing this kind of problems, as demonstrated in an experiment carried in [5], where a biofeedback based on ErrPs is applied to a closed-loop system for the behavioral correction of a robot. The feasibility in using the ErrPs combined to BCI signals for correcting the erroneous commands was investigated in [6]; in addition, the authors in [11] found out that ErrPs are also elicited as a misinterpretation of user's intent. Another research [12] proposed the classification of error related potentials from EEG during a real-world driving task. While the subject was driving, a directional cue was shown before reaching an intersection and the proposed system infers whether the cued direction coincided with the subject's intention. Other works investigated the co-adaptation of human-agent using ErrPs and decoding of ErrPs in tasks with continuous feedback [13,14].

The most important challenge of non-invasive BCI applications relates to the performances, since the control cannot offer a constant level of assistance due to the weakness and noisy of EEG signals [15], [16], [17]. Consequently, the application of spatial filters to improve the Signal-to-Noise Ratio (SNR) and the single-trial classification is worthy of investigation. Spatial filters are proposed in the literature with the aim to increase the SNR by using a weighted sum of all electrodes rather than relying on a single, or a small sub-set, of EEG channels. Some examples of spatial filters are the so-called xDAWN and the Common Spatial Pattern (CSP) [15,18]. Variants and extensions of CSP are proposed in [18], [19], [20], trying to overcome the drawbacks of CSP and improve the classification of single-trial EEG. In [21,22], the authors proposed adaptive spatial filters, the former based on ensembles of CSP patches whereas the latter based on a combination of blind source separation and regression analysis.

In this work, Functional Source Separation (FSS) has been considered to estimate a spatial filter for learning the ErrPs in BCI context. In order to enhance evoked ErrPs, FSS algorithm is designed by considering the ErrPs as a functional constraints [23], [24], [25], [26]. A direct comparison between the FSS [23] and the xDAWN algorithm [15,27] is presented to show the capability of the spatial filters to enhance the evoked ErrPs. Moreover, a single-trial classification was reported to assess the performances of FSS with respect to xDAWN. Moreover, the FSS and xDAWN based methods are also compared with the single channels Cz and FCz, usually selected to monitor ErrPs [1], in terms of single-trial classification.

Section 1 of this paper presents the experimental protocol, the spatial filters and the Bayesian Linear Discriminant Analysis (BLDA) classification algorithm used in the study. Section 2 exposes the idea of using the FSS as spatial filter for learning the ErrPs in BCI. Experimental results are presented in Section 3 including a qualitative evaluation of the spatial filters and the single-trial classification. In Section 4, we discuss the overall results and draw conclusion. Finally, in Section 5, we highlight the future works.

Section snippets

Materials and methods

In this section the experimental details are presented and the FSS and xDAWN algorithms, together with the BLDA classification algorithm, are described. The Fig. 1 shows the proposed procedure which consist of a training and a testing phase. A detailed description of each part is reported in the following sections.

Results

In this section, a qualitative and quantitative comparison among single channel (FCz and Cz), xDAWN and FSS spatial filter is presented. Afterwards, classification results among single channel (FCz and Cz), xDAWN and FSS are reported as well.

Discussion

In this study, a semi-supervised algorithm named FSS was proposed to improve the SNR at the single-trial level with the aim to better classify ErrPs, as response, when a person realizes they are making an error during a task. Accurate classification of those responses become crucially important in a BCI framework. The data used in this study, were presented in [1] and the results obtained by FSS were compared with those obtained by xDAWN [15,27] and single channels Cz and FCz [1,23], which are

Future works

As future works it could be possible to put the algorithm online for the testing phase after an offline training phase, removing BCI signal artifacts and allowing the possibility to online detect the ErrPs. In this way the BCI system can be employed as an interface in an online task, during which a person realizes they are making an error as a consequence of a cognitive mistake.

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Declaration of Competing Interest

All authors have no financial and personal relationships with other people or organizations that could inappropriately bias the work.

References (56)

  • M. Spüler et al.

    Online use of error-related potentials in healthy users and people with severe motor impairment increases performance of a P300-BCI

    Clin. Neurophysiol.

    (2012)
  • F. Barcelo

    Detection of change: event-related potential and fMRI findings

    Clin. Neurophysiol.

    (2004)
  • J. Polich et al.

    An integrative theory of P3a and P3b

    Clin. Neurophysiol.

    (2007)
  • R. Chavarriaga et al.

    Learning from eeg error-related potentials in noninvasive brain-computer interfaces

    IEEE Trans. Neural Syst. Rehabil. Eng.

    (2010)
  • P.W. Ferrez et al.

    You are wrong! - Automatic detection of interaction errors from brain waves

  • R. Chavarriaga et al.

    Errare machinale est: the use of error-related potentials in brain-machine interfaces

    Front. Neurosci.

    (2014)
  • B. Dal Seno et al.

    Online detection of P300 and error potentials in a BCI speller

    Comput. Intell. Neurosci.

    (2010)
  • A.F. Salazar-Gomez et al.

    Correcting robot mistakes in real time using EEG signals

  • X. Artusi et al.

    Performance of a simulated adaptive BCI based on experimental classification of movement-related and error potentials

    IEEE J. Emerg. Sel. Top. Circuits Syst.

    (2011)
  • I. Iturrate et al.

    Shared-control brain-computer interface for a two dimensional reaching task using EEG error-related potentials

  • M. Falkenstein et al.

    Effects of errors in choice reactiontasks on the ERP under focused and divided attention

  • G. Hajcak et al.

    To err is autonomic: error-related brain potentials, ANS activity, and post-error compensatory behavior

    Psychophysiology

    (2003)
  • M. Spüler et al.

    Error-related potentials during continuous feedback: using EEG to detect errors of different type and severity

    Front. Hum. Neurosci.

    (2015)
  • P.W. Ferrez et al.

    Error-related EEG potentials generated during simulated brain-computer interaction

    IEEE Trans. Biomed. Eng.

    (2008)
  • R. Chavarriaga et al.

    EEG-based decoding of error-related brain activity in a real-world driving task

    J. Neural Eng.

    (2015)
  • S.K. Ehrlich et al.

    Human-agent co-adaptation using error-related potentials

    J. Neural Eng.

    (2018)
  • C. Lopes Dias et al.

    Masked and unmasked error-related potentials during continuous control and feedback

    J. Neural Eng.

    (2018)
  • B. Rivet et al.

    xDAWN algorithm to enhance evoked potentials: application to brain-computer interface

    IEEE Trans. Biomed. Eng.

    (2009)
  • Cited by (17)

    • A CNN-transformer hybrid approach for decoding visual neural activity into text

      2022, Computer Methods and Programs in Biomedicine
      Citation Excerpt :

      Intuitively, only with a good understanding of the human brain can we make machines more intelligent. As one of the critical and challenging topics in cognitive neuroscience, brain decoding has been extensively studied for disease diagnosis [35,36] or brain-machine interfaces [37,38] recently. Language decoding aims to transform neural activities in the human brain into the natural language in the form of words or sentences [10], which will help the application of brain-machine interfaces.

    • A survey of quantum computing hybrid applications with brain-computer interface

      2022, Cognitive Robotics
      Citation Excerpt :

      This paper mainly discusses the research progress of hBCI and summarizes three types of hBCI, namely, hBCI based on a multi-brain model, multi-sensory hBCI, and hBCI based on multi-modal signals. By analyzing the general principles, paradigm design, experimental results, advantages, and applications of the latest hBCI system, we found that using hBCI technology can improve the detection performance of BCI and achieve multi-degree/multi-function control, which is significantly better than single-mode BCI [68]. Based on the above questions, in the following content, we briefly describe the application of hybrid BCI technology by researchers.

    • A new approach to feature extraction in MI-based BCI systems

      2022, Artificial Intelligence-Based Brain-Computer Interface
    View all citing articles on Scopus
    View full text