Removal of EOG artifacts and separation of different cerebral activity components from single channel EEG—An efficient approach combining SSA–ICA with wavelet thresholding for BCI applications

Highlights

• Highly efficient EOG artifact removal model.

• Retrieval of EEG information in $\alpha$ band with high accuracy.

• Along with removal of EOG artifacts it also provide separation of various cerebral activity components.

Abstract

The electroencephalogram (EEG) signals are usually interfered by many sources of noise like electrooculogram (EOG), which degrades the signals of interest. It causes the poor performance of the Brain–Computer Interface (BCI) systems. In this work, the problem of removal of EOG artifacts and separation of different cerebral activities found in a single-channel contaminated EEG is addressed. For this purpose, a novel model, based on the combined use of Singular Spectrum Analysis (SSA) and Independent Component Analysis (ICA) with a Stationary Wavelet Transform (SWT) is presented. ICA technique is a highly efficient method, which deals with the multichannel EEG signals. But it is difficult to apply the ICA on single channel EEG. Hence, using SSA, the single channel contaminated EEG signals are converted into multivariate information. Then, the multivariate information is fed to ICA, which separates the source signals as different independent components (ICs). Despite the fact that the ICA method performs excellent source separation, still, some required EEG signal content is present in the IC representing itself as an artifact, and thus dropping it would cause loss of EEG signal content. To avoid this problem, SWT is applied on the artifact IC, which performs the thresholding, to separate the actual artifact and preserve the EEG signal content. Matlab simulations have been done on both synthetically generated and real-life EEG signals and the proposed model is compared with the existing works. It is demonstrated that the proposed model has the best artifact separation performance than all the existing techniques, which is shown in terms of the metrics, RRMSE (Relative Root Mean Square Error) and MAE (Mean Absolute Error).

Introduction

Electroencephalogram signals are the neurophysiologic estimation of the electrical activity of the cerebrum, and these are mostly interfered by unwanted signals or noise called artifacts caused due to eye movement, eye squints, movement due to the head or electrodes, activity of muscles, breathing, heartbeat, electrical line noise, etc. It is highly essential, not only to remove the artifacts from the contaminated EEG signals but also to separate the different cerebral activity components present in the EEG signals for further investigation in clinical practice. Several EOG artifact removal methods have been proposed in [1], [2], [3], [4], [5], [6], [7], [8], [9], [10], [11], [12]. These methods are able to separate EOG artifacts only and fail to separate cerebral activity components present within the EEG signals. Independent Component Analysis (ICA), which is a Blind Source Separation (BSS) technique, is generally applied on multi-channel EEG information to separate concealed sources [13]. But it is difficult to apply the ICA directly on single channel EEG. In [14], ICA on single channel signals is proposed based on two assumptions, namely, (i) the source signals of interest should be stationary and (ii) it should be disassociated in the frequency spectrum. Anyway, such assumptions are not suitable for electroencephalogram applications, and subsequently, leads to degradation in performance.

To obtain better performance from ICA upon single-channel signal, the signal needs to be converted to multidimensional information by utilizing decomposition methods such as Ensemble Empirical Mode Decomposition (EEMD) or Wavelet transform [15]. Initially, the joint use of EEMD and ICA is proposed for single-channel EEG signals in [16] and it has been extended to several medical applications in [17], [18], [19]. In this, using the EEMD technique, the single-channel EEG is split to several IMFs (intrinsic mode functions). Then, ICA is applied to the IMF group to separate the different sources. But, its performance degrades due to edge effect problem [20], and also the method depends on the noise parameter (np) and number of ensembles (ne). To avoid the edge effect problem in EEMD–ICA, the combined use of Local Mean Decomposition and ICA (LMD–ICA) is proposed in [21], but in this the parameters adjustment in local mean decomposition is difficult. The combined use of DWT (Discrete Wavelet Transform) and ICA (w-ICA) is proposed in [22] and applied in several medical science applications in [23], [24]. In these works, the single-channel signal is decomposed into a certain number of wavelet components (WCs) using wavelet transform and ICA is applied on the WCs to separate hidden sources. But, the overall performance of w-ICA relies upon the choice of the mother wavelet. Moreover, it requires prior information about sources to be removed. Recently, the combination of Singular Spectrum Analysis (SSA) and ICA, namely SSA–ICA is presented in [25], in which the EEG is converted to multivariate information by using SSA with several decomposition stages. In each stage, the EEG is splits to lower, higher frequency bands and ICA is applied on these bands, which separates various sources. But this method has the problem of removing signal along with noise.

Thus, from study of existing works, it is observed that most of the source separation techniques use ICA and have some disadvantages. It is to be noted that in many EEG based biomedical applications, the artifacts are frequently present in a very narrow band of frequency. ICA works in the time-space which implies that, despite the ICA method being successful in source separation, yet some useful EEG signal content present in the IC gets misrepresented as artifact, and thus dropping it would cause a loss of EEG signal content which is not acceptable [23]. Hence, in this paper, to avoid such loss of needful EEG information, a novel method is proposed. This is based on the combined use of SSA and ICA with SWT. In this proposed work, the EEG signal is first converted into multivariate information using similar procedure presented in [25]. Then, ICA is employed on the multivariate information, which separates the hidden sources as different ICs. After that, the SWT is applied on the artifact IC, which performs the thresholding to separate the actual artifact from the artifact IC and thus preserves the EEG information content.

Matlab simulations are done on both synthetically generated and real-life EEG signals and the proposed model is compared with existing w-ICA, w-JADE, LMD–ICA, EEMD–ICA and SSA–ICA methods. It is inferred from the simulations that the proposed model has the best EOG artifact separation performance than the existing techniques, which is shown in terms of Relative Root Mean Square Error (RRMSE) and Mean Absolute Error (MAE). In addition, the proposed method is capable of retrieving the information in the $\alpha$ band with high accuracy as well as separating various cerebral activity components in single channel EEG for further clinical investigation. The rest of the paper is organized as follows. The existing techniques are discussed in Section 2. The proposed SSA–ICA with SWT technique is presented in Section 3. The experimental results are discussed in Section 4. Finally, Section 5 provides the conclusion.