Intelligent approach for artifacts removal from EEG signal using heuristic-based convolutional neural network
Introduction
EEG records the electrical activity of the cerebral cortex by electrodes, which are generally installed on the scalp. This approach is commonly employed for the medical diagnosis of sleep disorders and epilepsy [1]. Nowadays, EEG attracts more medical practitioners in the field of BCI applications [2]. Though, the EEG signals are generally distorted due to the environmental and biological artifacts for practical settings [3]. Undesired signals [4] are considered artifacts that include non-cerebral origins recorded by EEG electrodes [5]. The artifacts are categorized into environmental [6] artifacts and biological artifacts [7]. Environmental artifacts are generally originated from the external part of the human body owing to interference or electrode movement from external devices like electric motor or power main [8] whereas the signals from non-cerebral sources in the human body like muscle or ocular and cardiac activity are considered as biological artifacts [9], [10]. However, both environmental and biological artifacts degrade EEG signals, and thus BCI applications or medical diagnosis [11] is obstructed through the distortion of the considered power spectrum [12].
The consistent interpretation of the EEG recordings is influenced by the limitations of muscle artifacts, and hence a new emerging case is developed by adopting the wearable few-channel EEG [13]. Though, the low robustness and high computational load of the conventional approaches restrict their performance and broader applications in artifacts removal. When the EEG data is collected from the recording systems, the most important scenario is signal artifacts [14], [15], which may also contaminate the quality of EEG data. Moreover, complete knowledge of the artifacts types is required for efficiently removing the noise or artifacts. The unwanted signals or artifacts are mostly derived from physiological artifacts, experimental error, and environment noise [16]. In addition, the external factors such as experiment error and environment artifacts are categorized as extrinsic artifacts, where the intrinsic artifacts are taken from physiological activities like muscle activity, heartbeat, eye blink, etc from the body [17], [18]. The most important physiological artifacts such as extrinsic artifacts, cardiac artifacts, muscle artifacts, and ocular artifacts affect the EEG data [19].
Recent researchers have used different Time-Frequency Representation (TFR)-based signal decomposition approaches for removing the EEG artifacts [20]. Some of the TFR techniques like Wavelet Transform (WT), Short Time Fourier Transform (STFT), etc are proposed for extracting the fine-scale fluctuations in EEG signals [21]. The pure EEG activity is attained when the inverse WT of threshold wavelet coefficients is applied, which is completely based on EEG de-noising approaches. Various methods are developed for the removal of EMG artifacts, where the popular methods like signal decomposition approaches and Blind Source Separation (BSS) techniques such as ICA and Canonical Correlation Analysis (CCA) are chosen. Here, both BSS methods are efficiently used for removing the artifacts from the experimentation results, which are validated regarding computation time and accuracy [22]. Ensemble Empirical Mode Decomposition (EEMD) [18] attains more benefits concerned with their better capability of processing the non-stationary signals while considering the signal decomposition methods that were also avoided to select the wavelet basic in WT based on the data-driven principle [22]. The major issue of the EEG artifacts removal is the selection of the threshold level because it will never maintain the artifacts signals as the original ones, and it does not eliminate the coefficients of the original EEG signal [23]. Therefore, there is the necessity of designing a novel artifacts removal model.
The major contribution of the developed model is given here.
- •
To propose a new artifacts removal model from EEG signal by proposing a heuristic-based 1D-CNN using a hybrid optimization algorithm.
- •
To propose a new hybrid SM-EFO algorithm for efficient artifacts removal model by developing an improved 1D-CNN through the optimization of the number of hidden neurons and number of epochs to maximize the multi-objective function.
- •
To propose a new improved 1D-CNN using hybrid SM-EFO algorithm for attaining the clean signal from the 400 × 1 window of the noisy signal by minimizing the multi-objective function concerned with PSNR and RMSE.
- •
To analyze the performance of the developed artifacts removal model of EEG signal with different ECG, EMG, and EOG artifacts in terms of different measures while comparing with some meta-heuristic-based algorithms.
The remaining sections of the developed model are given here. Section II explains the related works. Section III proposes an enhanced deep learning-based artifacts removal from EEG signal. Section IV describes the improved 1D-CNN for artifacts removal from EEG signal. Section V implements the SM-EFO algorithm for improving the 1D-CNN architecture. Section VI analyzes the results. Section VII concludes this paper.
Section snippets
Related works
In 2017, Qazi and Kahalekar et al. [24] have developed a new optimization-based learning approach using Firefly and Levenberg Marquardt algorithms termed FLM, which has also adopted a Neural Network (NN)-enhanced adaptive filtering model for eliminating the artifacts from the EEG. The considered EEG signal was given to the adaptive filter to yield the optimal weights by using FLM. This was further given to the NN for attaining the optimal weights. Moreover, an enhanced model was developed based
Proposed model for EEG artifacts removal
EEG is more vulnerable to different biomedical or physiological signals that create more complexities in the analysis of EEG signals. Thus, removing noise or artifacts from the EEG signals is a significant parameter to analyze EEG signals in the medical field. Because, EEG signals play a vital role in the diagnosis and treatment of brain-related diseases like brain tumors and epilepsy, which prevents brain death. Though, this EEG signal consists of non-stationary and non-linear signals with
1-Dimensional CNN for artifacts removal
The 1D-CNN [36] is one of the categories of deep learning that has attained more attention in recent decades. It has been employed in diverse areas for raw continuous signals by efficient artifacts removal of different signals. The 1D-CNN architecture consists of different features like better mining structure of Spatio-temporal features, automatic learning of features for attaining adaptive design, and increased classification accuracy at a faster rate. It also shows feasible performance in
Objective model
The proposed artifacts removal model of EEG signals using the SM-EFO algorithm considers the major phase as an improved 1D-CNN model, where the multi-objective function regarding the minimization of PSNR and RMSE improves the performance. The objective function of the developed model is formulated in Eq. (4).
Here, PSNR is “defined as the ratio of the signal power to the noise power”. PSNR is employed to evaluate the perceptual quality of audio based on the
Experimental setup
The proposed artifacts removal model from EEG signal was implemented in MATLAB 2020a, and the experimental analysis was done by considering the maximum number of iterations as 25 and number of population as 10 with different performance measures in terms of Source-to-Distortion Ratio (SDR), Cumulative Squared Euclidean Distance (CSED), correlation coefficient and Mean Absolute Error (MAE). The analysis was performed by comparison with diverse optimization-based approaches like PSO [41], BSO [36]
Conclusion
This paper has developed a new artifacts removal model from EEG signals by proposing a new hybrid meta-heuristic-based deep learning. The artifacts considered in the EEG signals were ECG, EMG and EOG. A new improved 1D-CNN was developed using the proposed SM-EFO algorithm to show the complex non-linear correlation among noisy EEG signals and pure EEG signals. Here, the multi-objective function was considered in the improved 1D-CNN by minimizing the RMSE and PSNR, in which the optimization of
CRediT authorship contribution statement
Mariyadasu Mathe: Data curation. M. Padmaj: Project administration. Battula Tirumala Krishna: Visualization.
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 (41)
- et al.
FASTER: Fully Automated Statistical Thresholding for EEG artifact Rejection
Journal of Neuroscience Methods
(2010) - et al.
Effectiveness of Wavelet Denoising on Electroencephalogram Signals
Journal of Applied Research and Technology
(2013) - et al.
Recovering EEG brain signals: Artifact suppression with wavelet enhanced independent component analysis
Journal of Neuroscience Methods
(2006) - et al.
Automatic correction of artifact from single-trial event-related potentials by blind source separation using second order statistics only
Med. Eng. Phys
(2006) - et al.
Artifacts extraction from EEG data using the infomax approach
Biocybern Biomed Eng
(2011) - et al.
Trends in EEG-BCI for daily-life: requirements for artifacts removal
Biomed. Signal Process. Control
(2017) - et al.
Employing spatially constrained ICA and wavelet denoising, for automatic removal of artifacts from multichannel EEG data
Signal Processing
(2012) - et al.
A Comparative study of feature rankingtechniques for epileptic seizure detection using Wavelet Transform
Comput. Electr. Eng
(2016) - et al.
Removal of ocular artifact from the EEG: a review
Clin. Neurophysiol
(2000) - et al.
Artifacts removal from EEG signal: FLM optimization-based learning algorithm for neural network-enhanced adaptive filtering
Biocybernetics and Biomedical Engineering
(2017)
Mohammad Ali Badamchizadeh, “Artifacts removal in EEG signal using a new neural network enhanced adaptive filter,”
Neurocomputing
Application of hybrid GLCT-PICA de-noising method in automated EEG artifact removal
Biomedical Signal Processing and Control
A hybrid method for artifact removal of visual evoked EEG
Journal of Neuroscience Methods
Xiaojun Wu “A novel end-to-end 1D-rescnn model to remove artifact from EEG signals,”
Neurocomputing
Jinlong Wang “Automatic ocular artifacts removal in EEG using deep learning,”
Biomedical Signal Processing and Control
Artifact removal from electroencephalograms using a hybrid BSS-SVM algorithm
IEEE Signal Processing Letters
Unsupervised Eye Blink Artifact Denoising of EEG Data with Modified Multiscale Sample Entropy, Kurtosis, and Wavelet-ICA
IEEE Journal of Biomedical and Health Informatics
Automatic removal of ocular artefacts using adaptive filtering and independent component analysis for electroencephalogram data
IET Signal Processing
Cited by (12)
Less complexity-aware intelligent 1-dimensional CNN with weighted deep features for artifacts removal in EEG signals
2024, Digital Signal Processing: A Review JournalImplementation of artificial intelligence and machine learning-based methods in brain–computer interaction
2023, Computers in Biology and MedicineHigh performance clean versus artifact dry electrode EEG data classification using Convolutional Neural Network transfer learning
2023, Clinical Neurophysiology PracticePareto Front Estimation in Multi-Objective Optimization for Signal Separation
2023, 2023 Innovations in Intelligent Systems and Applications Conference, ASYU 2023Evolutionary Algorithm-Based Optimal Wiener-Adaptive Filter Design: An Application on EEG Noise Mitigation
2023, IEEE Transactions on Instrumentation and Measurement