Motor imagery based brain-computer interface: improving the EEG classification using Delta rhythm and LightGBM algorithm

https://doi.org/10.1016/j.bspc.2021.103102Get rights and content

Highlights

Abstract

This article contains a new method to improving the EEG motor imagery classification system quality with an application on BCI competition IV 2a, 2b, and PhysioNet EEG-MI datasets. This work uses a bandpass filter to eliminates all unused signals and then increases the prediction accuracy from 50% to more than 96% in both binary and multi-class cases, knowing that applying PSO optimizer on the parameters of the LightGBM classifier allows to find the best and stable status of EEG signals classification, also decision tree algorithm (DT) allows to get the importance degree of all acquisition electrodes used in the classification stage. This work also uses the correlation matrix to determined all artifacts between different electrodes, in such a way the prediction accuracy value increases from 50% and 60% to higher values of 96% and 98% in binary and multi-class classification, and high prediction speed remains more than 63703 and 2395 samples per second in binary and multi-class cases respectively. A comparison at the end of related works found a maximum accuracy value of around 85.5%.

Introduction

In the human brain, billions of neurons communicate with one another via tiny electrical signals [11], [38], [73]. These electrical signals mean the sum of the inhibition and excitation potential of the dendrites of brain neuron ensemble with parallel geometric orientation [14], [27], [50], these signals are measured from the scalp using non-invasive electrodes, and these signals are called the electroencephalograms (EEG) [28], [32], [73]. BCI allows the human brain to communicate directly with an external device [1]. The brain-computer interface (BCI) collects signals from the brain and translates them into one or more actions. The outcomes of these events have not only provided patients with disabilities and injured cerebrospinal fluid with an important channel for communication with the environment, but they have also helped to improve their quality of life by allowing them to take control of the living facilities around them [12]. There are more and more research studies around the world as the low cost of electronic devices and technological advances in the recording of brain signals [12], [17]. BCI uses many methods of recording signals. These include electroencephalography, functional magnetic resonance imaging (fMRI), the magnetic activity of the brain (MEG), electrocorticogram (ECOG), and near-infrared spectroscopy (NIRS). EEG is the non-invasive and most cost-effective of these methods [1], [12], [66], [70]. The other advantage of EEG is its better temporal resolution than in any other way. In recent years, the BCI system, based on EEG signals, has been extensively investigated and implemented [21], [41]. Between different types of EEG signals, motor imagery signal (MI) [8], [62], recent research has generated a great deal of interest, as it is a fairly flexible EEG technology that allows us to discriminate against various brain activities. MI-EEG signals are recording of brain activity when the subject presents or wants to perform tasks such as hand or foot movement. MIEEG signals are generated in the brain sensory cortex area in response to these imaginative or thought actions [49]. These MI signals were used by researchers to distinguish between different brain activities for different tasks. The use of computer-aided technology and EEG signals has been widely used, especially in the medical diagnosis of complex diseases [32], [59], [76], major depressive disorder (MDD) [4], [45], and dementia [72], [74] such as Alzheimer, alcohol use disorder (AUD) [9], [44], Parkinson and dementia with Lewy bodies (DLB), mild cognitive impairment (MCI). The application of Motor Imagery based on EEG has opened up a new door in the realm of neuroprosthesis [32], [42], [46]. In addition, other areas of research, such as authentication [18], [75], emotion recognition [15], [57], sleep stage classification [65], drowsiness monitoring [19], [32], and eye state detection [1], [2], [3], [31], [54] with the use of physiological data such as EEG, these results are becoming more and more obvious. Various machine learning and deep learning algorithms were used to conduct semi-automated classification [8], [67]. In recent years, a large number of EEG classification studies have been carried out using different classification techniques and a collection of execution and evaluation of different characteristics has been collected, including the k-nearest neighbor (k-NN), support vector machine (SVM), linear discriminant analysis (LDA) and artificial neural network (ANN) [48]. Deep Learning (DL) [37], [53] can greatly simplify processing channels, allowing automatic end-to-end training of preprocessing, retrieval, and classification modules, while ensuring competitive performance in targets. Deep neural networks (DNN) inspired by earlier models such as multi-layer perceptrons (MLP) [10], [52], [53] In this study, we are interested in using the EEG signals bandwidth between 1/10 and 1 Hz of Delta rhythms to minimize the noise effect and to reserve all characteristics of each task and the employing of a common spatial pattern filter (CSP) makes large independence between several channels, even also the use of correlation matrix evaluate the acquisition system quality when determining all links existing between the different electrodes. To minimize the number of channels from 22 to just 4 electrodes without degrading the prediction quality, this work uses the decision tree (DT) algorithm. Thus the EEG classification was done using the LGBM algorithm because of the best results that give compared to other classifiers. Also, this work uses a PSO optimizer to find good classification parameters Fig. 1, then the accuracy values are more than 95% of EEG motor imagery classification in binary and multi-class states, such that the maximum value in the literature remains at the order of 85.5% [2], knowing that selecting better channels and parameters optimization are using for every single subject [3]. The rest of this article is organized as follows, Section 2 presents the related work for this paper, Section 3 presents all datasets used, feature extraction, classification, and optimization algorithms proposed for the prediction system, the results obtained and discussed are presented in Section 4, while Section 5 provides conclusions and an overview of future work.

Section snippets

Related work

Over the past decade, EEG modeling and classification has attracted a lot of attention from most researchers, such as that Monesi and Sardouie [43] introduces a new method of special extraction called extended common spatial pattern (ECSP) analysis, which uses already available data knowledge to produce a wider range of characteristics than traditional PSC analysis. Similarly, they presented the extended common temporal pattern (ECTP) analysis. To utilize spatial and time information, he

BCI competition IV dataset 2b

We used BCI Competition IV 2b dataset available on the BCI competition website.2 This competition aimed to find out whether a test was related to the motor imaging of the left hand (Class 1) or the motor imaging of the right hand (Class 2). EEG data were collected from nine people using three bipolar electrodes at a sampling rate of 250 Hz. Since only 3 channels are available depending on the time model [16], [43].

BCI competition IV dataset 2a

We also used the BCI

Results and discussion

The experiments were conducted on the 2.4 GHz desktop and 6 GB of RAM with four Intel®Core (TM) i5 CPUs and 64-bit/Windows 10 operating system.

Conclusion

In conclusion, the prediction system carried out in this work is applied on the BCI competition IV and PhysioNet-MI datasets show its effectiveness compared to the literature, which shows a great leap towards a good future of technology based on BCI systems for diagnosing complex disease and control by thought. Moreover, this method allows for adjustment and updates of the classification parameters for any acquisition system and subject, with the aim of improving the prediction of signals

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

References (76)

  • A. Saghafi et al.

    Random eye state change detection in real-time using eeg signals

    Exp. Syst. Appl.

    (2017)
  • A. Subasi

    Eeg signal classification using wavelet feature extraction and a mixture of expert model

    Exp. Syst. Appl.

    (2007)
  • H. Xiong et al.

    Two-dimensional FLD for face recognition

    (2005)
  • J. Yang et al.

    Two-dimensional discriminant transform for face recognition

    (2005)
  • S. Abenna et al.

    Brain-computer interface: Rhythm alpha analysis for eyes signals

  • S. Abenna, M. Nahid, A. Bajit, BCI: Classifiers optimization for EEG signals acquiring in real-time, in: 2020 6th IEEE...
  • S. Abenna, M. Nahid, and A. Bajit. Brain-computer interface: A novel EEG classification for baseline eye states using...
  • U.R. Acharya et al.

    Automated eeg-based screening of depression using deep convolutional neural network

    Comput. Methods Programs Biomed.

    (2018)
  • M.H. Alomari et al.

    Eeg mouse: A machine learning-based brain computer interface

    Int. J. Adv. Comput. Sci. Appl.

    (2014)
  • S.U. Amin et al.

    Multilevel weighted feature fusion using convolutional neural networks for eeg motor imagery classification

    IEEE Access

    (2019)
  • S.U. Amin et al.

    Deep learning for eeg motor imagery classification based on multi-layer cnns feature fusion

    Fut. Generat. Comput. Syst.

    (2019)
  • K. Ang et al.

    Filter bank common spatial pattern algorithm on bci competition iv datasets 2a and 2b

    Front. Neurosci.

    (2012)
  • A. Anuragi et al.

    Alcohol use disorder detection using eeg signal features and flexible analytical wavelet transform

    Biomed. Signal Process. Contr.

    (2019)
  • A.R. Aslam, T. Iqbal, M. Aftab, W. Saadeh, M. Awais, B. Altaf. A 10.13uJ/classification 2-channel Deep Neural...
  • E. Aydemir et al.

    A tunable-q wavelet transform and quadruple symmetric pattern based eeg signal classification method

    Medical Hypotheses

    (2019)
  • M.A. Bakhshali et al.

    Eeg signal classification of imagined speech based on riemannian distance of correntropy spectral density

    Biomed. Signal Process. Contr.

    (Jun 2019)
  • A. Barachant et al.

    Common spatial pattern revisited by riemannian geometry

  • M. Biophysics. Cell Physiology Sourcebook. ISBN...
  • J.A. Blanco et al.

    Single-trial cognitive stress classification using portable wireless electroencephalography

    Sensors

    (2019)
  • C. Brunner, R. Leeb, G.R. Müller-Putz, A. Schlögl, and G. Pfurtscheller. Bci competition iv, data set iib....
  • K. Casimo et al.

    Bci use and its relation to adaptation in cortical networks

    IEEE Trans. Neural Syst. Rehabil. Eng.

    (2017)
  • J. Chen et al.

    Eeg-based biometric identification with convolutional neural network

    Multimedia Tools Appl.

    (2019)
  • L. Chen, Y. Zhao, J. Zhang, and J. Zou. Automatic detection of alertness/drowsiness from physiological signals using...
  • E. Combrisson, K. Jerbi, Exceeding chance level by chance: The caveat of theoretical chance levels in brain signal...
  • E. Donchin et al.

    The mental prosthesis: assessing the speed of a p300-based brain-computer interface

    IEEE Trans. Rehabil. Eng.

    (2000)
  • J. Dong et al.

    Estimation of daily dew point temperature by using bat algorithm optimization based extreme learning machine

    Appl. Therm. Eng.

    (2019)
  • H. Dose et al.

    An end-to-end deep learning approach to mi-eeg signal classification for bcis

    Exp. Syst. Appl.

    (2018)
  • R. Eberhart et al.

    A new optimizer using particle swarm theory

  • P. Gaur et al.

    A multi-class eeg-based bci classification using multivariate empirical mode decomposition based filtering and riemannian geometry

    Exp. Syst. Appl.

    (2018)
  • M. Gong, Y. Bai, J. Qin, J. Wang, P. Yang, S. Wang, Gradient boosting machine for predicting return temperature of...
  • M. Hallett, The Bereitschaftspotential Movement –Related Marjan J ahanshahi, 2003. ISBN...
  • A. Hassan et al.

    Automatic identification of epileptic seizures from eeg signals using linear programming boosting

    Comput. Methods Programs Biomed.

    (2016)
  • H. Jiajun et al.

    Ultra-short term wind prediction with wavelet transform, deep belief network and ensemble learning

    Energy Conversion and Management

    (2020)
  • X.-y. Jing, H.-s. Wong, D. Zhang, Face recognition based on 2D Fisherface approach. 39:707-710, 2006....
  • B. Kaur, D. Singh, P.P. Roy, Eyes open and eyes close activity recognition using eeg signals,...
  • A. Khosla et al.

    A comparative analysis of signal processing and classification methods for different applications based on eeg signals

    Biocybernetics and Biomedical Engineering

    (2020)
  • Y. Kim et al.

    Motor imagery classification using mu and beta rhythms of eeg with strong uncorrelating transform based complex common spatial patterns

    Computational Intelligence and Neuroscience

    (2016)
  • S. Kumar et al.

    Csp-tsm: Optimizing the performance of riemannian tangent space mapping using common spatial pattern for mi-bci

    Computers in Biology and Medicine

    (2017)
  • Cited by (16)

    • EEG-based BCI: A novel improvement for EEG signals classification based on real-time preprocessing

      2022, Computers in Biology and Medicine
      Citation Excerpt :

      Moreover, this work uses a CSP filter to minimize the correlation between all EEG channels. Therefore, all extracted features have been well stabilized in the prediction system, and LGBM algorithm is used for the classification and feature selection steps [6,24,33,40]. As a result, the entire system was applied to EEG data to detect baseline eye states such as the SAES acquired using our EEG device, UCI dataset [1,41,42], and motor imagery tasks using PhysioNet-MI dataset as illustrated in Fig. 1, to find that average accuracy values are greater than the related work’s results.

    View all citing articles on Scopus
    1

    ORCID: 0000-0002-4172-7439

    View full text