Automatic minimization of ocular artifacts from electroencephalogram: A novel approach by combining Complete EEMD with Adaptive Noise and Renyi's Entropy

EPIC: Annotated epileptic EEG independent components for artifact reduction

Scalp electroencephalogram is a non-invasive multi-channel biosignal that records the brain’s electrical activity. It is highly susceptible to noise that might overshadow important data. Independent component analysis is one of the most used artifact removal methods. Independent component analysis separates data into different components, although it can not automatically reject the noisy ones. Therefore, experts are needed to decide which components must be removed before reconstructing the data. To automate this method, researchers have developed classifiers to identify noisy components. However, to build these classifiers, they need annotated data. Manually classifying independent components is a time-consuming task. Furthermore, few labelled data are publicly available. This paper presents a source of annotated electroencephalogram independent components acquired from patients with epilepsy (EPIC Dataset). This dataset contains 77,426 independent components obtained from approximately 613 hours of electroencephalogram, visually inspected by two experts, which was already successfully utilised to develop independent component classifiers.

Measurement(s)	Electroencephalogram Independent Components
Technology Type(s)	MATLAB and Python

Intelligent Extraction of Salient Feature From Electroencephalogram Using Redundant Discrete Wavelet Transform

At present, electroencephalogram (EEG) signals play an irreplaceable role in the diagnosis and treatment of human diseases and medical research. EEG signals need to be processed in order to reduce the adverse effects of irrelevant physiological process interference and measurement noise. Wavelet transform (WT) can provide a time-frequency representation of a dynamic process, and it has been widely utilized in salient feature analysis of EEG. In this paper, we investigate the problem of translation variability (TV) in discrete wavelet transform (DWT), which causes degradation of time-frequency localization. It will be verified through numerical simulations that TV is caused by downsampling operations in decomposition process of DWT. The presence of TV may cause severe distortions of features in wavelet subspaces. However, this phenomenon has not attracted much attention in the scientific community. Redundant discrete wavelet transform (RDWT) is derived by eliminating the downsampling operation. RDWT enjoys the attractive merit of translation invariance. RDWT shares the same time-frequency pattern with that of DWT. The discrete delta impulse function is used to test the time-frequency response of DWT and RDWT in wavelet subspaces. The results show that DWT is very sensitive to the translation of delta impulse function, while RDWT keeps the decomposition results unchanged. This conclusion has also been verified again in decomposition of actual EEG signals. In conclusion, to avoid possible distortions of features caused by translation sensitivity in DWT, we recommend the use of RDWT with more stable performance in BCI research and clinical applications.

Ensemble Deep Neural Network for Automatic Classification of EEG Independent Components

OBJECTIVE

Independent component analysis (ICA) is commonly used to remove noisy artifacts from multi-channel scalp electroencephalogram (EEG) signals. ICA decomposes EEG into different independent components (ICs) and then, experts remove the noisy ones. This process is highly time-consuming and experts are not always available. To surpass this drawback, research is going on to develop models to automatically conduct IC classification. Current state-of-the-art models use power spectrum densities (PSDs) and topoplots to classify ICs. The performance of these methods may be limited by disregarding the IC time-series that would contribute to fully simulate the visual inspection performed by experts.

METHODS

We present a novel ensemble deep neural network that combines time-series, PSDs, and topoplots to classify ICs. Moreover, we study the ability to use our model in transfer learning approaches.

RESULTS

Experimental results showed that using time-series improves IC classification. Results also indicated that transfer learning obtained higher performance than simply training a new model from scratch.

CONCLUSION

Researchers should develop IC classifiers using the three sources of information. Moreover, transfer learning approaches should be considered when producing new deep learning models.

SIGNIFICANCE

This work improves IC classification, enhancing the automatic removal of EEG artifacts. Additionally, since labelled ICs are generally not publicly available, the possibility of using our model in transfer learning studies may motivate other researchers to develop their own classifiers.

Objective Assessments of Mental Fatigue During a Continuous Long-Term Stress Condition

Prolonged periods of cognitive workload will cause mental fatigue, but objective, quantitative, and sensitive measurements that reflect long-term, stress-induced mental fatigue have yet to be elucidated. This study aims to apply a potential marker of Rényi entropy to investigate the mental fatigue changes in a long-term, high-level stress condition and compare three different instruments for assessment of mental fatigue: EEG, the oddball task, and self-scoring. We recruited nine individuals who participated in a 5-day intellectually challenging competition. The participants were assessed for mental fatigue each day of the competition using prefrontal cortex electroencephalogram (EEG). Reaction time in an oddball task and self-rated scoring were used comparatively to evaluate the performance of the EEG. Repeated measures ANOVA was utilized to analyze the differences among score, reaction time, and wavelet Rényi entropy. The results demonstrated that both wavelet Rényi entropy extracted from EEG and self-rated scoring revealed significant increases in mental fatigue during the 5 days of competition (P < 0.001). The reaction time of the oddball task did not show significant changes during the five-day competition (P = 0.066). Moreover, the wavelet Rényi entropy analysis of EEG showed greater sensitivity than the self-rated scoring and reaction time of the oddball task for measuring mental fatigue changes. In conclusion, this study shows that mental fatigue accumulates during long-term, high-level stress situations. The study also indicates that EEG wavelet Rényi entropy is an efficient metric to reflect the change of mental fatigue under a long-term stress condition and that EEG is a better method to assess long-term mental fatigue.

Multimodal analysis of electroencephalographic and electrooculographic signals

Electrooculography (EOG) is a method to concurrently obtain electrophysiological signals accompanying an Electroencephalography (EEG), where both methods have a common cerebral pattern and imply a similar medical significance. The most common electrophysiological signal source is EOG that contaminated the EEG signal and thereby decreases the accuracy of measurement and the predicated signal strength. In this study, we introduce a method to improve the correction efficiency for EOG artifacts (EOAs) on raw EEG recordings: We retrieve cerebral information from three EEG signals with high system performance and accuracy by applying feature engineering and a novel machine-learning (ML) procedure. To this end, we use two adaptive algorithms for signal decomposition to remove EOAs from multichannel EEG signals: empirical mode decomposition (EMD) and complete ensemble empirical mode decomposition (CEEMD), both using the Hilbert-Huang transform. First, the signal components are decomposed into multiple intrinsic mode functions. Next, statistical feature extraction and dimension reduction using principal component analysis are employed to select optimal feature sets for the ML procedure that is based on classification and regression models. The proposed CEEMD algorithm enhances the accuracy compared to the EMD algorithm and considerably improves the multi-sensory classification of EEG signals. Models of three different categories are applied, and the classification is based on a K-nearest neighbor (k-NN) algorithm, a decision tree (DT) algorithm, and a support vector machine (SVM) algorithm with accuracies of 94% for K-NN, 75% for DT, and 69% for SVM. For each classification model, a regression learner is used to assist as an evidence rule for the proposed artificial system and to influence the learning process from classification and regression models. The regression learning algorithms applied include algorithms based on an ensemble of trees (ET), a DT, and a SVM. We find that the ET-based regression model exhibits a determination coefficient R² = 1.00 outperforming the other two approaches with R² = 0.80 for DT and R² = 0.76 for SVM.

Complete Ensemble Empirical Mode Decomposition on FPGA for Condition Monitoring of Broken Bars in Induction Motors

Empirical mode decomposition (EMD)-based methods are powerful digital signal processing techniques because they do not need a priori information of the target signal due to their intrinsic adaptive behavior. Moreover, they can deal with non-linear and non-stationary signals. This paper presents the field programmable gate array (FPGA) implementation for the complete ensemble empirical mode decomposition (CEEMD) method, which is applied to the condition monitoring of an induction motor. The CEEMD method is chosen since it overcomes the performance of EMD and EEMD (ensemble empirical mode decomposition) methods. As a first application of the proposed FPGA-based system, the proposal is used as a processing technique for feature extraction in order to detect and classify broken rotor bar faults in induction motors. In order to obtain a complete online monitoring system, the feature extraction and classification modules are also implemented on the FPGA. Results show that an average effectiveness of 96% is obtained during the fault detection.

Comparison of a Pragmatic and Regression Approach for Wearable EEG Signal Quality Assessment

Wearable electroencephalogram (EEG) solutions allow portability and real-time measurements in uncontrolled conditions. For reliable and reproducible interpretation of the EEG data, it is essential to accurately identify EEG segments contaminated by artefacts. Two data quality indicator approaches are proposed: pragmatic and regression based. The former extracts statistical features and applies data-driven thresholding, while the latter uses a regression model on the same set of statistical features to predict data quality. The performance of the approaches is validated against EEG data recorded during uncontrolled laboratory and free-living conditions, and compared to a validated approach. The proposed approaches achieve average accuracy of over [Formula: see text] in detecting artefactual data, which is higher than the FORCe signal quality estimation method ([Formula: see text]). The main strength of the proposed algorithms is in the significant increase of specificity over the state-of-the-art. The two models perform equally across different databases. Training of the two approaches on free-living conditions data showed better generalization when tested on different types of databases, i.e., uncontrolled laboratory and free-living. Although the accuracy in determining artefact-contaminated data is highest when using a window size of 8 s, the accuracy drop is minor when using shorter window size, demonstrating another advantage over existing methods. Given low complexity of both pragmatic and regression approach, it facilitates a real-time implementation, which is demonstrated using a wearable EEG headset system available at IMEC.

Time-domain exponential energy for epileptic EEG signal classification

Automatic classification and prediction of epileptic electroencephalogram (EEG) signal are of great concern to the research community due to its non-stationary and non-linear properties. Features with minimal computation cost are highly needed for the rapid real-time precise diagnosis and implementation in the EEG scanning devices. Even though energy is a well-known feature for the analysis of signals, it is very rarely used in EEG analysis. An exponential energy feature in the time domain is proposed in this study. The proposed exponential energy feature provides a classification accuracy of 89% in the Bern-Barcelona EEG dataset and 99.5% in the Ralph Andrzejak EEG dataset. The promising results open a wide applicability of exponential energy in biomedical signal analysis.