Biomedical Signal Processing and Control 39 (2018) 23–33 Contents lists available at ScienceDirect Biomedical Signal Processing and Control journal homepage: www.elsevier.com/locate/bspc Research Paper A novel approach for noise removal and distinction of EEG recordings Nader Alharbi a,b,∗ a Bournemouth University, Bournemouth, UK b King Saud bin Abdulaziz University for Health Sciences, Riyadh, Saudi Arabia a r t i c l e i n f o Article history: Received 20 June 2016 Received in revised form 21 March 2017 Accepted 20 July 2017 Keywords: EEG Noise removal Distinction Epileptic Eigenvalues a b s t r a c t This paper presents a novel approach for the analysis of electroencephalography (EEG) signals. It is based on the distribution of the eigenvalues of a scaled Hankel matrix, which can enable us to determine the number of eigenvalues required for noise removal and signal extraction in singular spectrum analysis. This paper examines the applicability of the approach to discriminate between epileptic seizure and normal EEG signals, the extraction of attractive patterns, the filtering of EEG signals and the elimination of the noise included in the signals. Various criteria are used as features to distinguish between epileptic and normal EEG segments. The results indicate the capability of the approach for noise removal in both EEG signals, and for discrimination between them. © 2017 Elsevier Ltd. All rights reserved. 1. Introduction The EEG signal is the record of electrical brain activity. It is a complex signal, and one of the most frequently used to study and investigate neurological disorders. EEG signals show not only the brain function, but also the state of all the body systems [1]. Fur- thermore, the EEG biosignal records play a significant role in the detection and treatment of brain diseases such as epilepsy and brain tumors. Moreover, the analysis of EEG signals can be used to diagnose brain death [2]. The EEG signal is a valuable tool to study the function of the brain and neurobiological disorders; however, its recording is con- taminated by diverse types of noises and artifacts, which can cause problems in the accurate analysis of brain signals. These types of noise can be electrical, or can be made by our bodies, since the signal records have small amplitudes [2]. Furthermore, different artifacts such as ocular ones, blinking of the eyes, or muscle activities make noise in EEG recording and detecting such noise becomes a complex task. Although the signals can be affected by an internal or external noise, which often have unknown characteristics, they can be iden- tified if the signal and noise subspaces are accurately separated. It is known that removing artifacts or noises from the EEG signal is necessary for analysing any kind of brain diseases, and is helpful in decomposing the signal in a proper manner. Various methods can be implemented for denoising or removing noise from EEG signals; for instance, independent component analysis (ICA) [3–5], princi- ∗ Correspondence to: King Saud bin Abdulaziz University for Health Sciences, Riyadh, Saudi Arabia. pal component analysis (PCA) [6–8] and wavelet transform (WT) [9]. Chaos theory is a complicated mathematical theory that can describe complex dynamical systems. Chaotic behaviour can be found in many nonlinear dynamical systems in nature [10]. For example, an EEG signal is considered as a nonlinear time series, particularly during an epileptic seizure, and can also be character- ized as being chaotic [10,11]. Therefore, there have been critical attempts to differentiate strange attractors in brain signals [12]. Furthermore, as decomposing brain signal is a necessary tool for detecting epileptic activity, the extracted information from EEG recordings or the detection of epileptic seizure is helpful for diag- nosing and treating epilepsy patients. A major challenge in analysing EEG is that its data is often non- stationary, particularly when an abnormal event is observed within the signals [13]. Several methods have been applied to the analy- sis and discrimination of different categories of EEG signals [13]. However, many of these depend on restrictive assumptions of the normality and linearity of the observed data. Thus, development of a technique which is robust for analysing non-stationary time series is of paramount importance in accurate diagnosis of brain diseases. The singular spectrum analysis (SSA) technique, for instance, is not based on these assumptions, and can be helpful for analysing and modelling biomedical data. SSA is a new method, which has been developed and used to solve several practical medical problems (for more information about SSA we refer to [14,15]). For example, it has been used in the extraction of a weak fetal heart signal from a noisy maternal ECG [16]; separation of biomedical data such as electromyography http://dx.doi.org/10.1016/j.bspc.2017.07.011 1746-8094/© 2017 Elsevier Ltd. All rights reserved.