Real-time ocular artifact suppression using recurrent neural network for electro-encephalogram based brain-computer interface

Enhancing EEG Artifact Removal Efficiency by Introducing Dense Skip Connections to IC-U-Net

Electroencephalographic (EEG) data is considered contaminated with various types of artifacts. Deep learning has been successfully applied to developing EEG artifact removal techniques to increase the signal-to-noise ratio (SNR) and enhance brain-computer interface performance. Recently, our research team has proposed an end-to-end UNet-based EEG artifact removal technique, IC-U-Net, which can reconstruct signals against various artifacts. However, this model suffers from being prone to overfitting with a limited training dataset size and demanding a high computational cost. To address these issues, this study attempted to leverage the architecture of UNet++ to improve the practicability of IC-U-Net by introducing dense skip connections in the encoder-decoder architecture. Results showed that this proposed model obtained superior SNR to the original model with half the number of parameters. Also, this proposed model achieved comparable convergency using a quarter of the training data size.

Channel-independent recreation of artefactual signals in chronically recorded local field potentials using machine learning

Acquisition of neuronal signals involves a wide range of devices with specific electrical properties. Combined with other physiological sources within the body, the signals sensed by the devices are often distorted. Sometimes these distortions are visually identifiable, other times, they overlay with the signal characteristics making them very difficult to detect. To remove these distortions, the recordings are visually inspected and manually processed. However, this manual annotation process is time-consuming and automatic computational methods are needed to identify and remove these artefacts. Most of the existing artefact removal approaches rely on additional information from other recorded channels and fail when global artefacts are present or the affected channels constitute the majority of the recording system. Addressing this issue, this paper reports a novel channel-independent machine learning model to accurately identify and replace the artefactual segments present in the signals. Discarding these artifactual segments by the existing approaches causes discontinuities in the reproduced signals which may introduce errors in subsequent analyses. To avoid this, the proposed method predicts multiple values of the artefactual region using long–short term memory network to recreate the temporal and spectral properties of the recorded signal. The method has been tested on two open-access data sets and incorporated into the open-access SANTIA (SigMate Advanced: a Novel Tool for Identification of Artefacts in Neuronal Signals) toolbox for community use.

Predictions of ocular changes caused by diabetes in glaucoma patients

BACKGROUND AND OBJECTIVE

This paper builds different neural network models with simple topologies, having one or two hidden layers which were subsequently employed in the prediction of ocular changes progression in patients with diabetes associated with primer open-angle glaucoma.

MATERIAL AND METHODS

For attempting to indicate whether there is a relationship between glaucoma and diabetes, a simulation method, based on artificial neural networks (ANN), Jordan Elman networks (JEN) type, in particular, was applied in conjunction with clinical observation. The study was conducted on a sample of 101 eyes with open angle glaucoma included and, in each case, the patients had associated diabetes mellitus. A high degree of accuracy was exhibited by the models, demonstrating the potential effectiveness of this artificial intelligence technique for predicting ocular changes associated with diabetes. The parameters considered in this study for modelling purpose were: glaucoma age, diabetes age, C/D ratio (cup/disk size), glycated haemoglobin level (HbA1c), intraocular pressure (IOP), patient age, mean deviation (MD) and LENS appearance.

RESULTS

Relatively simple models, feed-forward neural networks with one or two intermediate layers, provided clinically meaningful data in direct modelling, the probability of correct answers being of 95%. Inverse modelling was also performed, in which MD depreciation was the output parameter. High accuracy was exhibited, in this case, with Jordan Elman networks, with the confidence interval of ±15%.

CONCLUSIONS

The neural models have demonstrated the possibility of their use in successfully predicting the relationship between glaucoma and diabetes in a real clinical environment.

Brain Dynamics in Predicting Driving Fatigue Using a Recurrent Self-Evolving Fuzzy Neural Network

This paper proposes a generalized prediction system called a recurrent self-evolving fuzzy neural network (RSEFNN) that employs an on-line gradient descent learning rule to address the electroencephalography (EEG) regression problem in brain dynamics for driving fatigue. The cognitive states of drivers significantly affect driving safety; in particular, fatigue driving, or drowsy driving, endangers both the individual and the public. For this reason, the development of brain-computer interfaces (BCIs) that can identify drowsy driving states is a crucial and urgent topic of study. Many EEG-based BCIs have been developed as artificial auxiliary systems for use in various practical applications because of the benefits of measuring EEG signals. In the literature, the efficacy of EEG-based BCIs in recognition tasks has been limited by low resolutions. The system proposed in this paper represents the first attempt to use the recurrent fuzzy neural network (RFNN) architecture to increase adaptability in realistic EEG applications to overcome this bottleneck. This paper further analyzes brain dynamics in a simulated car driving task in a virtual-reality environment. The proposed RSEFNN model is evaluated using the generalized cross-subject approach, and the results indicate that the RSEFNN is superior to competing models regardless of the use of recurrent or nonrecurrent structures.

Performance Evaluation and Implementation of FPGA Based SGSF in Smart Diagnostic Applications

The main objective of the paper is to implement Savitzky Golay Smoothing Filter (SGSF) so as to apply in pre-processing of real time smart medical diagnostic systems. As very important information of EEG and ECG waveforms lies in the peak of the signal, hence it becomes absolutely necessary to filter noise and artifacts from the signal. The implemented filter should be able to reject the noise efficiently along with the least distortion from the original signal. The shape preserving characteristics of the filter are determined by introducing different noise levels in the signal. The designed filter is tested on synthetic signals of EEG and ECG by adding different types of noise and the performance is analysed on various parameters, i.e., SNR, SSNR, SNRI, MSE, COR and signal distortion of the final output. The smoothing performance comparison of SGSF with the most commonly used Moving Average Filter (MAF) proves that SGSF is more efficient. Hence it is suggested that MAF can be replaced by SGSF. For real time issues, it is further implemented on reconfigurable architectures so as to achieve high speed, low cost, low power consumption and less area. Therefore SGSF is realized on FPGA platform to combine the advantages of both. Real time EEG and ECG signals are also considered for experimentation. The experimental results show that the proposed methodology (FPGA-SGSF) significantly reduces the processing time and preserves the actual features of the signal.

Classification of brain signals associated with imagination of hand grasping, opening and reaching by means of wavelet-based common spatial pattern and mutual information

An important issue in designing a practical brain-computer interface (BCI) is the selection of mental tasks to be imagined. Different types of mental tasks have been used in BCI including left, right, foot, and tongue motor imageries. However, the mental tasks are different from the actions to be controlled by the BCI. It is desirable to select a mental task to be consistent with the desired action to be performed by BCI. In this paper, we investigated the detecting the imagination of the hand grasping, hand opening, and hand reaching in one hand using electroencephalographic (EEG) signals. The results show that the ERD/ERS patterns, associated with the imagination of hand grasping, opening, and reaching are different. For classification of brain signals associated with these mental tasks and feature extraction, a method based on wavelet packet, regularized common spatial pattern (CSP), and mutual information is proposed. The results of an offline analysis on five subjects show that the two-class mental tasks can be classified with an average accuracy of 77.6% using proposed method. In addition, we examine the proposed method on datasets IVa from BCI Competition III and IIa from BCI Competition IV.

An online EEG-based brain-computer interface for controlling hand grasp using an adaptive probabilistic neural network

This paper presents a new online single-trial EEG-based brain-computer interface (BCI) for controlling hand holding and sequence of hand grasping and opening in an interactive virtual reality environment. The goal of this research is to develop an interaction technique that will allow the BCI to be effective in real-world scenarios for hand grasp control. One of the major challenges in the BCI research is the subject training. Currently, in most online BCI systems, the classifier was trained offline using the data obtained during the experiments without feedback, and used in the next sessions in which the subjects receive feedback. We investigated whether the subject could achieve satisfactory online performance without offline training while the subjects receive feedback from the beginning of the experiments during hand movement imagination. Another important issue in designing an online BCI system is the machine learning to classify the brain signal which is characterized by significant day-to-day and subject-to-subject variations and time-varying probability distributions. Due to these variabilities, we introduce the use of an adaptive probabilistic neural network (APNN) working in a time-varying environment for classification of EEG signals. The experimental evaluation on ten naïve subjects demonstrated that an average classification accuracy of 75.4% was obtained during the first experiment session (day) after about 3 min of online training without offline training, and 81.4% during the second session (day). The average rates during third and eighth sessions are 79.0% and 84.0%, respectively, using previously calculated classifier during the first sessions, without online training and without the need to calibrate. The results obtained from more than 5000 trials on ten subjects showed that the method could provide a robust performance over different experiment sessions and different subjects.

Removing ocular movement artefacts by a joint smoothened subspace estimator

To cope with the severe masking of background cerebral activity in the electroencephalogram (EEG) by ocular movement artefacts, we present a method which combines lower-order, short-term and higher-order, long-term statistics. The joint smoothened subspace estimator (JSSE) calculates the joint information in both statistical models, subject to the constraint that the resulting estimated source should be sufficiently smooth in the time domain (i.e., has a large autocorrelation or self predictive power). It is shown that the JSSE is able to estimate a component from simulated data that is superior with respect to methodological artefact suppression to those of FastICA, SOBI, pSVD, or JADE/COM1 algorithms used for blind source separation (BSS). Interference and distortion suppression are of comparable order when compared with the above-mentioned methods. Results on patient data demonstrate that the method is able to suppress blinking and saccade artefacts in a fully automated way.

Automated polysomnogram artifact compensation using the generalized singular value decomposition algorithm

Manual/visual polysomnogram (psg) analysis is a standard and commonly implemented procedure utilized in the diagnosis and treatment of sleep related human pathologies. Current technological trends in psg analysis focus upon translating manual psg analysis into automated/computerized approaches. A necessary first step in establishing efficient automated human sleep analysis systems is the development of reliable pre-processing tools to discriminate between outlier/artifact instances and data of interest. This paper investigates the application of an automated approach, using the generalized singular value decomposition algorithm, to compensate for specific psg artifacts.

Application of paraconsistent artificial neural networks as a method of aid in the diagnosis of Alzheimer disease

The visual analysis of EEG has shown useful in helping the diagnosis of Alzheimer disease (AD) when the diagnosis remains uncertain, being used in some clinical protocols. However, such analysis is subject to the inherent equipment imprecision, patient movement, electrical records, and physician interpretation of the visual analysis variation. The Artificial Neural Network (ANN) could be a helpful tool, appropriate to address problems such as prediction and pattern recognition. In this work, it has use a new class of ANN, the Paraconsistent Artificial Neural Network (PANN), which is capable of handling uncertain, inconsistent, and paracomplete information, for recognizing predetermined patterns of EEG and to assess its value as a possible auxiliary method for AD diagnosis. Thirty three patients with Alzheimer's disease and 34 controls patients of EEG records were obtained during relaxed wakefulness. It was considered as normal patient pattern, the background EEG activity between 8.0 and 12.0 Hz (with an average frequency of 10 Hz), allowing a range of 0.5 Hz. The PANN was able to recognize waves that belonging to their respective bands of clinical use (theta, delta, alpha, and beta), leading to an agreement with the clinical diagnosis at 82% of sensitivity and at 61% of specificity. Supported with these results, the PANN could be a promising tool to manipulate EEG analysis, bearing in mind the following considerations: the growing interest of specialists in EEG analysis visual and the ability of the PANN to deal directly imprecise, inconsistent and paracomplete data, providing an interesting quantitative and qualitative analysis.

An on-line BCI for control of hand grasp sequence and holding using adaptive probabilistic neural network

This paper presents a new EEG-based Brain-Computer Interface (BCI) for on-line controlling the sequence of hand grasping and holding in a virtual reality environment. The goal of this research is to develop an interaction technique that will allow the BCI to be effective in real-world scenarios for hand grasp control. Moreover, for consistency of man-machine interface, it is desirable the intended movement to be what the subject imagines. For this purpose, we developed an on-line BCI which was based on the classification of EEG associated with imagination of the movement of hand grasping and resting state. A classifier based on probabilistic neural network (PNN) was introduced for classifying the EEG. The PNN is a feedforward neural network that realizes the Bayes decision discriminant function by estimating probability density function using mixtures of Gaussian kernels. Two types of classification schemes were considered here for on-line hand control: adaptive and static. In contrast to static classification, the adaptive classifier was continuously updated on-line during recording. The experimental evaluation on six subjects on different days demonstrated that by using the static scheme, a classification accuracy as high as the rate obtained by the adaptive scheme can be achieved. At the best case, an average classification accuracy of 93.0% and 85.8% was obtained using adaptive and static scheme, respectively. The results obtained from more than 1500 trials on six subjects showed that interactive virtual reality environment can be used as an effective tool for subject training in BCI.

Signal quality of simultaneously recorded invasive and non-invasive EEG

Both invasive and non-invasive electroencephalographic (EEG) recordings from the human brain have an increasingly important role in neuroscience research and are candidate modalities for medical brain-machine interfacing. It is often assumed that the major artifacts that compromise non-invasive EEG, such as caused by blinks and eye movement, are absent in invasive EEG recordings. Quantitative investigations on the signal quality of simultaneously recorded invasive and non-invasive EEG in terms of artifact contamination are, however, lacking. Here we compared blink related artifacts in non-invasive and invasive EEG, simultaneously recorded from prefrontal and motor cortical regions using an approach suitable for detection of small artifact contamination. As expected, we find blinks to cause pronounced artifacts in non-invasive EEG both above prefrontal and motor cortical regions. Unexpectedly, significant blink related artifacts were also found in the invasive recordings, in particular in the prefrontal region. Computing a ratio of artifact amplitude to the amplitude of ongoing brain activity, we find that the signal quality of invasive EEG is 20 to above 100 times better than that of simultaneously obtained non-invasive EEG. Thus, while our findings indicate that ocular artifacts do exist in invasive recordings, they also highlight the much better signal quality of invasive compared to non-invasive EEG data. Our findings suggest that blinks should be taken into account in the experimental design of ECoG studies, particularly when event related potentials in fronto-anterior brain regions are analyzed. Moreover, our results encourage the application of techniques for reducing ocular artifacts to further optimize the signal quality of invasive EEG.

A spike detection method in EEG based on improved morphological filter

In this paper, a spike detection method is introduced. Traditional morphological filter is improved for extracting spikes from epileptic EEG signals and two key problems are addressed: morphological operation design and structure elements optimization. An average weighted combination of open-closing and close-opening operation, which can eliminate statistical deflection of amplitude, is utilized to separate background EEG and spikes. Then, according to the characteristic of spike component, the structure elements are constructed with two parabolas and a new criterion is put forward to optimize the structure elements. The proposed method is evaluated using normal and epileptic EEG data recorded from 12 test subjects. A comparison between the improved morphological filter, traditional morphological filter and wavelet analysis with Mexican hat function is presented, which indicates that the improved morphological filter is superior in restraining background activities. We demonstrate that the average detection rate of the improved morphological filter is much higher than that of the other two methods, and there is no false detection for normal EEG signals with the proposed method.

Paraconsistent artificial neural networks and Alzheimer disease: a preliminary study

EEG visual analysis has proved useful in aiding AD diagnosis, being indicated in some clinical protocols. However, such analysis is subject to the inherent imprecision of equipment, patient movements, electric registers, and individual variability of physician visual analysis.

Removal of ocular artifacts from the EEG: a comparison between time-domain regression method and adaptive filtering method using simulated data

We recently proposed an adaptive filtering (AF) method for removing ocular artifacts from EEG recordings. The method employs two parameters: the forgetting factor lambda and the filter length M. In this paper, we first show that when lambda = M = 1, the adaptive filtering method becomes equivalent to the widely used time-domain regression method. The role of lambda (when less than one) is to deal with the possible non-stationary relationship between the reference EOG and the EOG component in the EEG. To demonstrate the role of M, a simulation study is carried out that quantitatively evaluates the accuracy of the adaptive filtering method under different conditions and comparing with the accuracy of the regression method. The results show that when there is a shape difference or a misalignment between the reference EOG and the EOG artifact in the EEG, the adaptive filtering method can be more accurate in recovering the true EEG by using an M larger than one (e.g. M = 2 or 3).

Electro-encephalogram based brain-computer interface: improved performance by mental practice and concentration skills

Mental imagination is the essential part of the most EEG-based communication systems. Thus, the quality of mental rehearsal, the degree of imagined effort, and mind controllability should have a major effect on the performance of electro-encephalogram (EEG) based brain-computer interface (BCI). It is now well established that mental practice using motor imagery improves motor skills. The effects of mental practice on motor skill learning are the result of practice on central motor programming. According to this view, it seems logical that mental practice should modify the neuronal activity in the primary sensorimotor areas and consequently change the performance of EEG-based BCI. For developing a practical BCI system, recognizing the resting state with eyes opened and the imagined voluntary movement is important. For this purpose, the mind should be able to focus on a single goal for a period of time, without deviation to another context. In this work, we are going to examine the role of mental practice and concentration skills on the EEG control during imaginative hand movements. The results show that the mental practice and concentration can generally improve the classification accuracy of the EEG patterns. It is found that mental training has a significant effect on the classification accuracy over the primary motor cortex and frontal area.

A fully automatic method for ocular artifact suppression from EEG data using wavelet transform and independent component analysis

Contamination of electroencephalographic (EEG) recordings with different kinds of artifacts is the main obstacle to the analysis of EEG data. Independent component analysis (ICA) is a general accepted tool for isolating artifactual components. One major challenge to artifact removal using ICA is the automatic identification of the artifactual components. However there is still little consensus on criteria for automatic rejection of undesired components. In this paper we present a new identification procedure based on an efficient combination of statistical and wavelet-based measures for ocular artifact suppression. The results on 420 4-s EEG epochs indicate that the artifact components can be identified correctly with 96.4%

Real-time ocular artifacts suppression from EEG signals using an unsupervised adaptive blind source separation

Independent component analysis (ICA) has been shown to be a powerful tool for artifactual suppression from electroencephalogram (EEG) recordings. However, the real-time application of this method for artifact rejection has not been considered so far. This article presents a method based on an unsupervised, self-normalizing, adaptive learning algorithm for on-line blind source separation. Simulation results are provided to show the validity and effectiveness of the technique with different distributions. The results from real-data demonstrate that the proposed scheme removes perfectly eye blink and eye movement artifacts from the EEG signals and is suitable for use during on-line EEG monitoring such as EEG-based brain computer interface.