Single-channel EEG Research Articles

Automatic sleep stage classification using time–frequency images of CWT and transfer learning using convolution neural network

For automatic sleep stage classification, the existing methods mostly rely on hand-crafted features selected from polysomnographic records. In this paper, the goal is to develop a deep learning-based method by using single channel electroencephalogram (EEG) that automatically exploits the time–frequency spectrum of EEG signal, removing the need for manual feature extraction. The time–frequency RGB color images for EEG signal are extracted using continuous wavelet transform (CWT). The transfer learning of a pre-trained convolution neural network, squeezenet is employed to classify these CWT images into its sleep stages. The proposed method is evaluated using a publicly available Physionet sleep EDFx dataset using single-channel EEG Fpz-Cz channel. Evaluation results show that this method can achieve near state of the art accuracy even using single channel EEG signal.

Automatic Detection and Correction of Blink Artifacts in Single Channel EEG Signals

Ocular Artifacts (OAs) are inevitable during EEG acquisition and make the signal analysis critical. Detection and correction of these artifacts is a major problem now a day’s. In this paper an energy detection method is used to detect the artifacts and performed wavelet thresholding within the researched zones to protect neural data at non blink regions. Various sets of Wavelet Transform (WT) techniques and threshold functions are collated and identification of the optimum combination for OA’s separation is indicated in many research areas including Technology & Management. The output of these methods at blink regions is compared interns of various standard metrics using established techniques of Supply Chains. Results of this study demonstrate that the SWT+HT has better in rejecting the artifacts than other methods in this paradigm.

Pair-Wise Matching of EEG Signals for Epileptic Identification via Convolutional Neural Network

Electroencephalogram (EEG) have been extensively analyzed to identify the characteristics of epileptic seizures in the literature. However, most of these studies focus on the properties of single channel EEG data while neglecting the association between signals from diverse channels. To bridge this gap, we propose an EEG instance matching-based epilepsy classification approach by introducing one convolutional neural network (CNN). First of all, each pair of EEG signals are exploited to form one 2 dimensional matrix, which could be used to reveal the interaction between them. Secondly, the generated matrices are fed into the proposed CNN that would discriminate the input representations. To evaluate the performance of the presented approach, the comparison experiments between the state-of-the-art techniques and our work are conducted on publicly available epilepsy EEG benchmark database. Experimental results indicate that the proposed algorithm could yield the performance with an average accuracy of 99.3%, average sensitivity of 99.5%, and average specificity 99.6%.

Automatic Sleep Stage Classification With Single Channel EEG Signal Based on Two-Layer Stacked Ensemble Model

Sleep stage classification, including wakefulness (W), rapid eye movement (REM), and non- rapid eye movement (NREM) which includes three sleep stages that describe the depth of sleep, is one of the most critical steps in effective diagnosis and treatment of sleep-related disorders. Clinically, sleep staging is performed by domain experts through visual inspection of polysomnography (PSG) recordings, which is time-consuming, labor-intensive and often subjective in nature. Therefore, this study develops an automatic sleep staging system, which uses single channel electroencephalogram (EEG) signal, for convenience of wearing and less interference in the sleep, to do automatic identification of various sleep stages. To achieve the automatic sleep staging system, this study proposes a two-layer stacked ensemble model, which combines the advantages of random forest (RF) and LightGBM (LGB), where RF focuses on reducing the variance of the proposed model while LGB focuses on reducing the bias of the proposed model. Particularly, the proposed model introduces a class balance strategy to improve the N1 stage recognition rate. In order to evaluate the performance of the proposed model, experiments are performed on two datasets, including Sleep-EDF database (SEDFDB) and Sleep-EDF Expanded database (SEDFEDB). In the SEDFDB, the overall accuracy (ACC), weight F1-score (WF1), Cohen’s Kappa coefficient (Kappa), sensitivity of N1 (SEN-N1) obtained by proposed model are 91.2%, 0.916, 0.864 and 72.52% respectively using subject-non-independent test (SNT). In parallel, the ACC, WF1, Kappa, SEN-N1 obtained by proposed model are 82.4%, 0.751, 0.719 and 27.15% respectively using subject-independent test (SIT). Experimental results show that the performance of the proposed model are competitive with the state-of-the-art methods and results, and the recognition rate of N1 stage is significantly improved. Moreover, in the SEDFEDB, the experimental results indicate the robustness and generality of the proposed model.

An Effective and Robust Framework for Ocular Artifact Removal From Single-Channel EEG Signal Based on Variational Mode Decomposition

Removal of ocular artifacts (OAs) from electroencephalogram (EEG) signal is crucial for accurate and effective EEG analysis and brain-computer interface research. The elimination of OAs is quite challenging in absence of reference electro-oculogram and in single-channel EEG signal using existing independent component analysis based OA removal techniques. Though few of the recent OAs removal techniques suppress the OAs in the single-channel significantly, these techniques introduce distortion in clinical features of the EEG signal during artifact removal process. To address these issues, in this paper, we propose a robust framework for the detection and removal of OAs based on variational mode decomposition (VMD) and turning point count. The proposed framework exploits the effectiveness of VMD in two stages denoted as VMD-I and VMD-II respectively. The proposed framework has four components: EEG signal decomposition into two modes using VMD-I; rejection of low-frequency baseline components; processed EEG signal decomposition into three modes using VMD-II; rejection of mode containing OAs based on turning point count based threshold criteria. We evaluate the effectiveness of the proposed framework using the EEG signals in presence of various ocular artifacts with different amplitudes and shapes taken from three standard databases including, Mendeley database, MIT-BIH Polysmnographic database and EEG during mental arithmetic tasks database. Evaluation results demonstrate that proposed framework eliminates OAs with minimal loss in valuable clinical features in both reconstructed EEG signal and in all local rhythms. Furthermore, subjective and objective comparative analysis demonstrate that our framework outperforms few existing OAs removal techniques in removing OAs from single-channel EEG signal.

Channel Projection-Based CCA Target Identification Method for an SSVEP-Based BCI System of Quadrotor Helicopter Control.

The brain-computer interface (BCI) plays an important role in assisting patients with amyotrophic lateral sclerosis (ALS) to enable them to participate in communication and entertainment. In this study, a novel channel projection-based canonical correlation analysis (CP-CCA) target identification method for steady-state visual evoked potential- (SSVEP-) based BCI system was proposed. The single-channel electroencephalography (EEG) signals of multiple trials were recorded when the subject is under the same stimulus frequency. The CCAs between single-channel EEG signals of multiple trials and sine-cosine reference signals were obtained. Then, the optimal reference signal of each channel was utilized to estimate the test EEG signal. To validate the proposed method, we acquired the training dataset with two testing conditions including the optimal time window length and the number of the trial of training data. The offline experiments conducted a comparison of the proposed method with the traditional canonical correlation analysis (CCA) and power spectrum density analysis (PSDA) method using a 5-class SSVEP dataset that was recorded from 10 subjects. Based on the training dataset, the online 3D-helicopter control experiment was carried out. The offline experimental results showed that the proposed method outperformed the CCA and the PSDA methods in terms of classification accuracy and information transfer rate (ITR). Furthermore, the online experiments of 3-DOF helicopter control achieved an average accuracy of 87.94 ± 5.93% with an ITR of 21.07 ± 4.42 bit/min.

Sleep autoscoring based on a single EEG Channel: comparison with visual scoring in patients

Sleep autoscoring based on a single EEG Channel: comparison with visual scoring in patients

Deep learning enables accurate sleep staging based on a single frontal EEG channel

Deep learning enables accurate sleep staging based on a single frontal EEG channel

Decreased electrocortical temporal complexity distinguishes sleep from wakefulness

In most mammals, the sleep-wake cycle is constituted by three behavioral states: wakefulness (W), non-REM (NREM) sleep, and REM sleep. These states are associated with drastic changes in cognitive capacities, mostly determined by the function of the thalamo-cortical system. The intra-cranial electroencephalogram or electocorticogram (ECoG), is an important tool for measuring the changes in the thalamo-cortical activity during W and sleep. In the present study we analyzed broad-band ECoG recordings of the rat by means of a time-series complexity measure that is easy to implement and robust to noise: the Permutation Entropy (PeEn). We found that PeEn is maximal during W and decreases during sleep. These results bring to light the different thalamo-cortical dynamics emerging during sleep-wake states, which are associated with the well-known spectral changes that occur when passing from W to sleep. Moreover, the PeEn analysis allows us to determine behavioral states independently of the electrodes’ cortical location, which points to an underlying global pattern in the signal that differs among the cycle states that is missed by classical methods. Consequently, our data suggest that PeEn analysis of a single EEG channel could allow for cheap, easy, and efficient sleep monitoring.

GACNN SleepTuneNet: a genetic algorithm designing the convolutional neuralnetwork architecture for optimal classification ofsleep stages from a single EEG channel

This study presents a method for designing--by a genetic algorithm, without manual intervention--the feature learning architecture for classification of sleep stages from a single EEG channel, when using a convolutional neural network called GACNN SleepTuneNet. Two EEG electrode positions were selected, namely FP2-F4 and FPz-Cz, from two available datasets. Twenty-five generations were involved in diagnosis without hand-crafted features, to learn the architecture for classification of sleep stages based on AASM standard. Based on the results, our model not only achieved the highest classification accuracy, but it also distinguished the sleep stages based on either of the two EEG electrode signals, in both datasets. The results show that our model performed the best with highest overall accuracy rates and kappa statistic (CAP sleep: 95.61 % and 0.94; Sleep EDF: 92.51 % and 0.90) among other state-of-the-art methods that require no manual intervention. Our model could automatically learn the features for classification of sleep stages, for different raw EEG electrode positions in different datasets, without user-assisted feature extraction.

Novel Approaches for the Removal of Motion Artifact From EEG Recordings

The electroencephalogram (EEG) signal is contaminated with various noises or artifacts during recording. For the automated detection of neurological disorders, it is a vital task to filter out these artifacts from the EEG signal. In this paper, we propose two novel approaches for the removal of motion artifact from the single channel EEG signal. These methods are based on the multiresolution total variation (MTV) and multiresolution weighted total variation (MWTV) filtering schemes. The multiresolution analysis using the discrete wavelet transform (DWT) helps to segregate the EEG signal into various subband signals. The total variation (TV) and weighted TV (WTV) are applied to the approximation subband signal. The filtered approximation subband signal is evaluated based on the difference between the noisy approximation subband signal and the output of the TV or WTV filter. The processed EEG signal is obtained using the multiresolution wavelet-based reconstruction. The difference in the signal to noise ratio ( $\Delta $ SNR) and the percentage of reduction in correlation coefficients ( $\eta $ ) is used for evaluating the diagnostic quality of the processed EEG signal. The experimental results demonstrate that the proposed MTV and MWTV approaches have better denoising performance with (average $\Delta $ SNR, and average $\eta $ ) values of (29.12 dB and 68.56%) and (29.29 dB and 67.51%), respectively, as compared to the existing techniques.

Predicting functional outcomes after stroke: an observational study of acute single-channel EEG

ABSTRACTBackground: Early and objective prediction of functional outcome after stroke is an important issue in rehabilitation. Electroencephalography (EEG) has long been utilized to describe and monitor brain function following neuro-trauma, and technological advances have improved usability in the acute setting. However, skepticism persists whether EEG can provide the same prognostic value as neurological examination.Objective: The current cohort study examined the relationship between acute single-channel EEG and functional outcomes after stroke.Methods: Resting-state EEG recorded at a single left pre-frontal EEG channel (FP1) was obtained from 16 adults within 72 h of first stroke. At 30 and 90 days, measures of disability (modified Rankin Scale; mRS) and involvement in daily activities (modified Barthel Index; mBI) were obtained. Acute EEG measures were correlated with functional outcomes and compared to an early neurological examination of stroke severity using the National Institute of Health Stroke Scale (NIHSS). Classification of good outcomes (mRS ≤1 or mBI ≥95) was also examined using Receiver Operator Curve analyses.Results: One-third to one-half of participants experienced incomplete post-stroke recovery, depending on the time point and measure. Functional outcomes correlated with acute theta values (rs 0.45–0.60), with the strength of associations equivalent to previously reported values obtained from conventional multi-channel systems. Acute theta values ≥0.25 were associated with good outcomes, with positive (67-83%) and negative predictive values (70-90%) comparable to those obtained using the NIHSS.Conclusions: Acute, single-channel EEG can provide unique, non-overlapping clinical information, which may facilitate objective prediction of functional outcome after stroke.

A nature-inspired biomarker for mental concentration using a single-channel EEG

We developed a system for measuring the attentional process during the performance of specific activities. The proposed biomarker device is able to estimate the mental concentration using a single-channel EEG. The system captures the EEG signal and several brain waves located in the left orbitofrontal brain region. Furthermore, we extended the input features of the system applying spectrum analysis. We applied two well-known evolutionary algorithms for selecting the best combination of input features: simulated annealing and geometric particle swarm optimization. Besides, we solved the binary classification problem (concentration vs. relaxation) using support vector machines and neural networks. Support vector machines are among the most common instruments for solving binary classification problems. On the other hand, we selected to study a family of neural networks named echo state networks, because the model is ideal for embedded systems and has shown good accuracy in real-world applications. The training and execution are fast, robust, and reliable. The developed system is autonomous, portable, reliable, non-invasive and has a low economic cost. Besides, it can be easily adjusted for each person and for each problem.

Diffuse to fuse EEG spectra – Intrinsic geometry of sleep dynamics for classification

We propose a novel algorithm for sleep dynamics visualization and automatic annotation by applying diffusion geometry based sensor fusion algorithm to fuse spectral information from two electroencephalograms (EEG). The diffusion geometry approach helps organize the nonlinear dynamical structure hidden in the EEG signal. The visualization is achieved by the nonlinear dimension reduction capability of the chosen diffusion geometry algorithms. For the automatic annotation purpose, the support vector machine is trained to predict the sleep stage. The prediction performance is validated on a publicly available benchmark database, Physionet Sleep-EDF [extended] SC* (SC = Sleep Cassette) and ST* (ST = Sleep Telemetry), with the leave-one-subject-out cross validation. When we have a single EEG channel (Fpz-Cz), the overall accuracy, macro F1 and Cohen's kappa achieve 82.72%, 75.91% and 76.1% respectively in Sleep-EDF SC* and 78.63%, 73.58% and 69.48% in Sleep-EDF ST*. This performance is compatible with the state-of-the-art results. When we have two EEG channels (Fpz-Cz and Pz-Oz), the overall accuracy, macro F1 and Cohen's kappa achieve 84.44%, 78.25% and 78.36% respectively in Sleep-EDF SC* and 79.05%, 74.73% and 70.31% in Sleep-EDF ST*. The results suggest the potential of the proposed algorithm in practical applications.

EEG sleep stages identification based on weighted undirected complex networks

Background and ObjectiveSleep scoring is important in sleep research because any errors in the scoring of the patient's sleep electroencephalography (EEG) recordings can cause serious problems such as incorrect diagnosis, medication errors, and misinterpretations of patient's EEG recordings. The aim of this research is to develop a new automatic method for EEG sleep stages classification based on a statistical model and weighted brain networks. MethodsEach EEG segment is partitioned into a number of blocks using a sliding window technique. A set of statistical features are extracted from each block. As a result, a vector of features is obtained to represent each EEG segment. Then, the vector of features is mapped into a weighted undirected network. Different structural and spectral attributes of the networks are extracted and forwarded to a least square support vector machine (LS-SVM) classifier. At the same time the network's attributes are also thoroughly investigated. It is found that the network's characteristics vary with their sleep stages. Each sleep stage is best represented using the key features of their networks. ResultsIn this paper, the proposed method is evaluated using two datasets acquired from different channels of EEG (Pz-Oz and C3-A2) according to the R&K and the AASM without pre-processing the original EEG data. The obtained results by the LS-SVM are compared with those by Naïve, k-nearest and a multi-class-SVM. The proposed method is also compared with other benchmark sleep stages classification methods. The comparison results demonstrate that the proposed method has an advantage in scoring sleep stages based on single channel EEG signals. ConclusionsAn average accuracy of 96.74% is obtained with the C3-A2 channel according to the AASM standard, and 96% with the Pz-Oz channel based on the R&K standard.

Orthogonal convolutional neural networks for automatic sleep stage classification based on single-channel EEG

Background and objectiveIn recent years, several automatic sleep stage classification methods based on convolutional neural networks (CNN) by learning hierarchical feature representation automatically from raw EEG data have been proposed. However, the state-of-the-art of such methods are quite complex. Using a simple CNN architecture to classify sleep stages is important for portable sleep devices. In addition, employing CNNs to learn rich and diverse representations remains a challenge. Therefore, we propose a novel CNN model for sleep stage classification. MethodsGenerally, EEG signals are better described in the frequency domain; thus, we convert EEG data to a time–frequency representation via Hilbert–Huang transform. To learn rich and effective feature representations, we propose an orthogonal convolutional neural network (OCNN). First, we construct an orthogonal initialization of weights. Second, to avoid destroying the orthogonality of the weights in the training process, orthogonality regularizations are proposed to maintain the orthogonality of weights. Simultaneously, a squeeze-and-excitation (SE) block is employed to perform feature recalibration across different channels. ResultsThe proposed method achieved a total classification accuracy of 88.4% and 87.6% on two public datasets, respectively. The classification performances of different convolutional neural networks models were compared to that of the proposed method. The experiment results demonstrated that the proposed method is effective for sleep stage classification. ConclusionsExperiment results indicate that the proposed OCNN can learn rich and diverse feature representations from time-frequency images of EEG data, which is important for deep learning. In addition, the proposed orthogonality regularization is simple and can be easily adapted to other architectures.

Single-Channel EEG-Based Machine Learning Method for Prescreening Major Depressive Disorder

Many studies developed the machine learning method for discriminating Major Depressive Disorder (MDD) and normal control based on multi-channel electroencephalogram (EEG) data, less concerned about using single channel EEG collected from forehead scalp to discriminate the MDD. The EEG dataset is collected by the Fp1 and Fp2 electrode of a 32-channel EEG system. The result demonstrates that the classification performance based on the EEG of Fp1 location exceeds the performance based on the EEG of Fp2 location, and shows that single-channel EEG analysis can provide discrimination of MDD at the level of multi-channel EEG analysis. Furthermore, a portable EEG device collecting the signal from Fp1 location is used to collect the second dataset. The Classification and Regression Tree combining genetic algorithm (GA) achieves the highest accuracy of 86.67% based on leave-one-participant-out cross validation, which shows that the single-channel EEG-based machine learning method is promising to support MDD prescreening application.

A validation study of a consumer wearable sleep tracker compared to a portable EEG system in naturalistic conditions

ObjectiveTo compare a wearable device, the Fitbit Versa (FV), to a validated portable single-channel EEG system across multiple nights in a naturalistic environment. MethodsTwenty participants (10 men and 10 women) aged 25–67 years were recruited for the present study. Study duration was 14 days during which participants were asked to wear the FV daily and nightly. The study intended to reproduce free-living conditions; thus, no guidelines for sleep or activity were imposed on the participants. A total of 138 person-nights, equivalent to 76,539 epochs, were used in the validation process. Sleep measures were compared between the FV and portable EEG using Bland-Altman plots, paired t-tests and epoch-by-epoch (EBE) analyses. ResultsThe FV showed no significant bias with the EEG for the global sleep measures time in bed (TIB) and total sleep time (TST), and for calculated sleep efficiency (cSE = [TST/TIB] x 100). The FV had 92.1% sensitivity, 54.1% specificity, and 88.5% accuracy with a Cohen's kappa of 0.41, but a prevalence- and bias adjusted kappa of 0.77. The predictive values for sleep (PVS; positive predictive value) and wakefulness (PVW; negative predictive value) were 95.0% and 42.0%, respectively. The FV showed significant bias compared to the portable EEG for time spent in specific sleep stages, for SE as provided by FV, for sleep onset latency, sleep period time, and wake after sleep onset. ConclusionsThe consumer sleep tracker could be a useful tool for measuring sleep duration in longitudinal epidemiologic naturalistic studies albeit with some limitations in specificity.

Automatic and tunable algorithm for EEG artifact removal using wavelet decomposition with applications in predictive modeling during auditory tasks

Brain–computer interface (BCI) systems are becoming increasingly popular nowadays. Electroencephalogram (EEG) signals recorded by BCI systems are however frequently contaminated by artifacts and while applying any artifact removal algorithm, precautions should be taken not to remove useful information. Widely and most popular approaches to remove artifacts from EEG are based on independent component analysis (ICA), which relies on the multichannel EEG signal and needs an expert to manually pick the artifactual component to remove it or needs reference signals of artifacts. Recently, wavelet-based approaches have been proposed and demonstrated as well-suited for single channel EEG signal. However, control over the loss of information still remains an issue. Therefore, in this paper, we propose an algorithm based on wavelet packet decomposition (WPD) that allows controlling the suppression or removal of presumed artifacts, by tuning intuitive parameters. The proposed algorithm has three operating modes and two tuning parameters. We study the performance of the proposed algorithm and compare it with ICA-based approaches and a comparative wavelet-based approach on an EEG dataset collected for a study of auditory tasks. In addition to visual inspection, spectral response and distribution, the results of predictive tasks show that the proposed approach performs better than ICA-based approach and performance can be further improved by properly tuning the parameters for an individual predictive model. The proposed approach also performs better in terms of retention of neural information and removal of artifactual noise measured by mutual information and correlation coefficient, as compared to the comparative wavelet-based approach.

Deep Learning With EEG Spectrograms in Rapid Eye Movement Behavior Disorder.

REM Behavior Disorder (RBD) is now recognized as the prodromal stage of α-synucleinopathies such as Parkinson's disease (PD). In this paper, we describe deep learning models for diagnosis/prognosis derived from a few minutes of eyes-closed resting electroencephalography data (EEG) collected at baseline from idiopathic RBD patients (n = 121) and healthy controls (HC, n = 91). A few years after the EEG acquisition (4±2 years), a subset of the RBD patients were eventually diagnosed with either PD (n = 14) or Dementia with Lewy bodies (DLB, n = 13), while the rest remained idiopathic RBD. We describe first a simple deep convolutional neural network (DCNN) with a five-layer architecture combining filtering and pooling, which we train using stacked multi-channel EEG spectrograms from idiopathic patients and healthy controls. We treat the data as in audio or image classification problems where deep networks have proven successful by exploiting invariances and compositional features in the data. For comparison, we study a simple deep recurrent neural network (RNN) model using three stacked Long Short Term Memory network (LSTM) cells or gated-recurrent unit (GRU) cells—with very similar results. The performance of these networks typically reaches 80% (±1%) classification accuracy in the balanced HC vs. PD-conversion classification problem. In particular, using data from the best single EEG channel, we obtain an area under the curve (AUC) of 87% (±1%)—while avoiding spectral feature selection. The trained classifier can also be used to generate synthetic spectrograms using the DeepDream algorithm to study what time-frequency features are relevant for classification. We find these to be bursts in the theta band together with a decrease of bursting in the alpha band in future RBD converters (i.e., converting to PD or DLB in the follow up) relative to HCs. From this first study, we conclude that deep networks may provide a useful tool for the analysis of EEG dynamics even from relatively small datasets, offering physiological insights and enabling the identification of clinically relevant biomarkers.