EEG-based Classification Research Articles

Classification of mindfulness experiences from gamma-band effective connectivity: Application of machine-learning algorithms on resting, breathing, and body scan

Background and ObjectivePracticing mindfulness is a mental process toward interoceptive awareness, achieving stress reduction and emotion regulation through brain-function alteration. Literature has shown that electroencephalography (EEG)-derived connectivity possesses the potential to differentiate brain functions between mindfulness naïve and mindfulness experienced, where such quantitative differentiation could benefit telediagnosis for mental health. However, there is no prior guidance in model selection targeting on the mindfulness-experience prediction. Here we hypothesized that the EEG effective connectivity could reach a good prediction performance in mindfulness experiences with brain interpretability. MethodsWe aimed at probing direct Directed Transfer Function (dDTF) to classify the participants’ history of mindfulness-based stress reduction (MBSR), and aimed at optimizing the prediction accuracy by comparing multiple machine learning (ML) algorithms. Targeting the gamma-band effective connectivity, we evaluated the EEG-based prediction of the mindfulness experiences across 7 machine learning (ML) algorithms and 3 sessions (i.e., resting, focus-breathing, and body-scan). ResultsThe support vector machine and naïve Bayes classifiers exhibited significant accuracies above the chance level across all three sessions, and the decision tree algorithm reached the highest prediction accuracy of 91.7 % with the resting state, compared to the classification accuracies with the other two mindful states. We further conducted the analysis on essential EEG channels to preserve the classification accuracy, revealing that preserving just four channels (F7, F8, T7, and P7) out of 19 yielded the accuracy of 83.3 %. Delving into the contribution of connectivity features, specific connectivity features predominantly located in the frontal lobe contributed more to classifier construction, which aligned well with the existing mindfulness literature. ConclusionIn the present study, we initiated a milestone of developing an EEG-based classifier to detect a person's mindfulness experience objectively. The prediction accuracy of the decision tree was optimal to differentiate the mindfulness experiences using the local resting-state EEG data. The suggested algorithm and key channels on the mindfulness-experience prediction may provide guidance for predicting mindfulness experiences using the EEG-based classification embedded in future wearable neurofeedback systems or plausible digital therapeutics.

EEG-based classification of Alzheimer’s disease and frontotemporal dementia: a comprehensive analysis of discriminative features

EEG-based classification of Alzheimer’s disease and frontotemporal dementia: a comprehensive analysis of discriminative features

Achieving EEG-based depression recognition using Decentralized-Centralized structure

Currently, electroencephalography (EEG)-based depression recognition using deep learning methods has become an important approach for early diagnosis and treatment. Due to the functional differences and activation differences among various brain regions, the channels in EEG data may potentially exhibit a certain degree of local correlation. Furthermore, several studies have indicated that there are significant differences in the synchronization for multiple brain regions between patients with depression and healthy controls. However, few researchers have designed depression recognition model architectures according to the spatial distribution characteristics of EEG data. In multiple existing EEG-based classification models, the channels of EEG data are directly merged. This approach hinders the model from effectively capturing the spatial information of EEG data and leads to a partial loss of synchronized information of several brain regions. In light of this, we propose a novel attention network named Decentralized-Centralized Structure Transformer (DCST), which can achieve effective depression recognition utilizing the spatial distribution characteristics of EEG data. The leave-one-subject-out cross-validation is employed to evaluate the classification performance for all models. The experimental results show that our proposed model achieves the highest classification accuracy, 89.8%.

Multimodal decoding of error processing in a virtual reality flight simulation

Technological advances in head-mounted displays (HMDs) facilitate the acquisition of physiological data of the user, such as gaze, pupil size, or heart rate. Still, interactions with such systems can be prone to errors, including unintended behavior or unexpected changes in the presented virtual environments. In this study, we investigated if multimodal physiological data can be used to decode error processing, which has been studied, to date, with brain signals only. We examined the feasibility of decoding errors solely with pupil size data and proposed a hybrid decoding approach combining electroencephalographic (EEG) and pupillometric signals. Moreover, we analyzed if hybrid approaches can improve existing EEG-based classification approaches and focused on setups that offer increased usability for practical applications, such as the presented game-like virtual reality flight simulation. Our results indicate that classifiers trained with pupil size data can decode errors above chance. Moreover, hybrid approaches yielded improved performance compared to EEG-based decoders in setups with a reduced number of channels, which is crucial for many out-of-the-lab scenarios. These findings contribute to the development of hybrid brain-computer interfaces, particularly in combination with wearable devices, which allow for easy acquisition of additional physiological data.

MSLTE: multiple self-supervised learning tasks for enhancing EEG emotion recognition

Objective. The instability of the EEG acquisition devices may lead to information loss in the channels or frequency bands of the collected EEG. This phenomenon may be ignored in available models, which leads to the overfitting and low generalization of the model. Approach. Multiple self-supervised learning tasks are introduced in the proposed model to enhance the generalization of EEG emotion recognition and reduce the overfitting problem to some extent. Firstly, channel masking and frequency masking are introduced to simulate the information loss in certain channels and frequency bands resulting from the instability of EEG, and two self-supervised learning-based feature reconstruction tasks combining masked graph autoencoders (GAE) are constructed to enhance the generalization of the shared encoder. Secondly, to take full advantage of the complementary information contained in these two self-supervised learning tasks to ensure the reliability of feature reconstruction, a weight sharing (WS) mechanism is introduced between the two graph decoders. Thirdly, an adaptive weight multi-task loss (AWML) strategy based on homoscedastic uncertainty is adopted to combine the supervised learning loss and the two self-supervised learning losses to enhance the performance further. Main results. Experimental results on SEED, SEED-V, and DEAP datasets demonstrate that: (i) Generally, the proposed model achieves higher averaged emotion classification accuracy than various baselines included in both subject-dependent and subject-independent scenarios. (ii) Each key module contributes to the performance enhancement of the proposed model. (iii) It achieves higher training efficiency, and significantly lower model size and computational complexity than the state-of-the-art (SOTA) multi-task-based model. (iv) The performances of the proposed model are less influenced by the key parameters. Significance. The introduction of the self-supervised learning task helps to enhance the generalization of the EEG emotion recognition model and eliminate overfitting to some extent, which can be modified to be applied in other EEG-based classification tasks.

A Case of Electroencephalography and Machine Learning in Early Diagnosis of Psychotic and Affective Disorders

AbstractElectroencephalography (EEG) serves as a non-invasive, cost-effective, and robust tool, directly measuring in-vivo neuronal mass activity with high temporal resolution. Using state-of-the-art machine learning techniques, EEG recordings have the potential to generate in silico biomarkers for severe mental disorders. In this study, we developed EEG-based classification models for schizophrenia and depression taking into account physiological and pathological aging processes. From a cohort (N=735, 51.6% male) that is acquired in LMU Hospital, Department of Psychiatry and Psychotherapy, comprising healthy control individuals (HC, N=245) and patients with schizophrenia (SCZ, N=250) or major depressive disorder (MDD, N=240), we extracted power spectrum density and connectivity measures based on 60 second resting-state EEG recordings with 19 channels. The support vector machine models were trained to 1) classify patients with SCZ or MDD and HC individuals, and 2) predict age in HC individuals using ten-by-ten repeated nested-cross validation. The age-predicting model was applied to patient groups to calculate EphysAGE (Electrophysiological Age Gap Estimation) by subtracting chronological age from chronological age. The links between diagnosis, medication, and EphysAGE, i.e., accelerated aging, were then further explored with univariate analyses. The EphysAGE Model had an explained variance of 46% (MAE=8.7 years, T=14.31, P1000<0.001). The patients with SCZ had a significantly higher EphysAGE (mean[SD]=0.61[10.32]) than the patients with MDD (mean[SD]=-1.10[10.49], p=0.04). The classification models discriminated SCZ from HC (Balanced Accuracy, BAC=72.7%, p<0.001), MDD from HC (BAC=67.0%, p<0.001), and SCZ from MDD individuals (BAC=63.2%, p<0.001). Higher EphysAGE was associated with an increased likelihood of being misclassified as SCZ in HC and MDD (ρHC=0.23, p<0.001; ρMDD=0.17, p=0.01) based on percentile rank scores from the SCZ Model. Moreover, in the Differential Diagnostic Model, higher EphysAGE is positively correlated with being misclassified as SCZ in patients with MDD (ρMDD=0.14, p=0.03). Machine learning models can extract electrophysiological signatures of MDD and SCZ for potential clinical use. However, the impact of aging processes on diagnostic separability calls for timely application of such models, possibly in early recognition settings.Disclosure of InterestNone Declared

Enhancing Adaboost performance in the presence of class-label noise: A comparative study on EEG-based classification of schizophrenic patients and benchmark datasets

The performance of Adaboost is highly sensitive to noisy and outlier samples. This is therefore the weights of these samples are exponentially increased in successive rounds. In this paper, three novel schemes are proposed to hunt the corrupted samples and eliminate them through the training process. The methods are: I) a hybrid method based on K-means clustering and K-nearest neighbor, II) a two-layer Adaboost, and III) soft margin support vector machines. All of these solutions are compared to the standard Adaboost on thirteen Gunnar Raetsch’s datasets under three levels of class-label noise. To test the proposed method on a real application, electroencephalography (EEG) signals of 20 schizophrenic patients and 20 age-matched control subjects, are recorded via 20 channels in the idle state. Several features including autoregressive coefficients, band power and fractal dimension are extracted from EEG signals of all participants. Sequential feature subset selection technique is adopted to select the discriminative EEG features. Experimental results imply that exploiting the proposed hunting techniques enhance the Adaboost performance as well as alleviating its robustness against unconfident and noisy samples over Raetsch benchmark and EEG features of the two groups.

Advanced EEG-Based Classification of Periodic Leg Movements and Bruxism Using Complexity and Entropy Features

Periodic leg movements (PLM) and bruxism are prevalent sleep disorders that significantly impact sleep quality and overall health. Accurate and timely detection of these disorders is pivotal for effective treatment and improved patient well-being. In this research paper, we present a novel approach for the identification of Periodic Leg Movements (PLM) and bruxism. Our methodology leverages sophisticated complexity and entropy features extracted from electroencephalogram (EEG) signals recorded across various sleep stages. The implementation involves a diverse set of artificial intelligence (AI) classifiers, encompassing K-nearest neighbours (KNN), artificial neural networks, support vector machines (SVM) with radial basis function, SVM with polynomial kernel, and SVM with sigmoid kernel. These classifiers are employed to categorize EEG signals into three distinct classes: PLM, bruxism, or normal sleep patterns. To underscore the efficacy and superiority of our proposed approach, we conduct a thorough comparison with existing publications in the field. Our approach consistently outperforms previous methods and attains remarkable accuracies of 98.74% for PLM and 94.74% for bruxism, particularly when utilizing the SVM Sigmoid classifier. Furthermore, our study highlights the potential of EEG signals in elucidating alterations in brain activity associated with PLM and bruxism, underscoring the reliability of complexity and entropy features as robust biomarkers for these sleep disorders. Our research significantly advances the contemporary pinnacle in EEG-based detection of Periodic Leg Movements (PLM) and bruxism signifies a significant advancement, holding potential for impactful clinical applications in the realm of sleep disorder diagnosis and treatment.

Emotion recognition based on convolutional gated recurrent units with attention

Studying brain activity and deciphering the information in electroencephalogram (EEG) signals has become an emerging research field, and substantial advances have been made in the EEG-based classification of emotions. However, using different EEG features and complementarity to discriminate other emotions is still challenging. Most existing models extract a single temporal feature from the EEG signal while ignoring the crucial temporal dynamic information, which, to a certain extent, constrains the classification capability of the model. To address this issue, we propose an Attention-Based Depthwise Parameterized Convolutional Gated Recurrent Unit (AB-DPCGRU) model and validate it with the mixed experiment on the SEED and SEED-IV datasets. The experimental outcomes revealed that the accuracy of the model outperforms the existing state-of-the-art methods, which confirmed the superiority of our approach over currently popular emotion recognition models.

EEG-Based Classification of Spoken Words Using Machine Learning Approaches

Amyotrophic lateral sclerosis (ALS) is a neurodegenerative disease that affects the nerve cells in the brain and spinal cord. This condition leads to the loss of motor skills and, in many cases, the inability to speak. Decoding spoken words from electroencephalography (EEG) signals emerges as an essential tool to enhance the quality of life for these patients. This study compares two classification techniques: (1) the extraction of spectral power features across various frequency bands combined with support vector machines (PSD + SVM) and (2) EEGNet, a convolutional neural network specifically designed for EEG-based brain–computer interfaces. An EEG dataset was acquired from 32 electrodes in 28 healthy participants pronouncing five words in Spanish. Average accuracy rates of 91.04 ± 5.82% for Attention vs. Pronunciation, 73.91 ± 10.04% for Short words vs. Long words, 81.23 ± 10.47% for Word vs. Word, and 54.87 ± 14.51% in the multiclass scenario (All words) were achieved. EEGNet outperformed the PSD + SVM method in three of the four classification scenarios. These findings demonstrate the potential of EEGNet for decoding words from EEG signals, laying the groundwork for future research in ALS patients using non-invasive methods.

Incorporating hand-crafted features into deep learning models for motor imagery EEG-based classification

Incorporating hand-crafted features into deep learning models for motor imagery EEG-based classification

Energy-based features and bi-LSTM neural network for EEG-based music and voice classification

AbstractThe human brain receives stimuli in multiple ways; among them, audio constitutes an important source of relevant stimuli for the brain regarding communication, amusement, warning, etc. In this context, the aim of this manuscript is to advance in the classification of brain responses to music of diverse genres and to sounds of different nature: speech and music. For this purpose, two different experiments have been designed to acquire EEG signals from subjects listening to songs of different musical genres and sentences in various languages. With this, a novel scheme is proposed to characterize brain signals for their classification; this scheme is based on the construction of a feature matrix built on relations between energy measured at the different EEG channels and the usage of a bi-LSTM neural network. With the data obtained, evaluations regarding EEG-based classification between speech and music, different musical genres, and whether the subject likes the song listened to or not are carried out. The experiments unveil satisfactory performance to the proposed scheme. The results obtained for binary audio type classification attain 98.66% of success. In multi-class classification between 4 musical genres, the accuracy attained is 61.59%, and results for binary classification of musical taste rise to 96.96%.

Multi-channel EEG-based classification of consumer preferences using multitaper spectral analysis and deep learning model

Multi-channel EEG-based classification of consumer preferences using multitaper spectral analysis and deep learning model

Regional Contribution in Electrophysiological-Based Classifications of Attention Deficit Hyperactive Disorder (ADHD) Using Machine Learning

Attention deficit hyperactivity disorder (ADHD) is a common neurodevelopmental condition in children and is characterized by challenges in maintaining attention, hyperactivity, and impulsive behaviors. Despite ongoing research, we still do not fully understand what causes ADHD. Electroencephalography (EEG) has emerged as a valuable tool for investigating ADHD-related neural patterns due to its high temporal resolution and non-invasiveness. This study aims to contribute to diagnostic accuracy by leveraging EEG data to classify children with ADHD and healthy controls. We used a dataset containing EEG recordings from 60 children with ADHD and 60 healthy controls. The EEG data were captured during cognitive tasks and comprised signals from 19 channels across the scalp. Our primary objective was to develop a machine learning model capable of distinguishing ADHD subjects from controls using EEG data as discriminatory features. We employed several well-known classifiers, including a support vector machine, random forest, decision tree, AdaBoost, Naive Bayes, and linear discriminant analysis, to discern distinctive EEG patterns. To further enhance classification accuracy, we explored the impact of regional data on the classification outcomes. We arranged the EEG data according to the brain regions from which they were derived (namely frontal, temporal, central, parietal, and occipital) and examined their collective effects on the accuracy of our classifications. Notably, we considered combinations of three regions at a time and found that certain combinations led to enhanced accuracy. Our findings underscore the potential of EEG-based classification in distinguishing children with ADHD from healthy controls. The Naive Bayes classifier yielded the highest accuracy (84%) when applied to specific region combinations. Moreover, we evaluated the classification performance based on hemisphere-specific EEG data and found promising results, particularly when using the right hemisphere region channels.

EEG-based classification combining Bayesian convolutional neural networks with recurrence plot for motor movement/imagery

Electroencephalogram (EEG)-based Motor imagery (MI) is a key topic in the brain-computer interface (BCI). The EEG-based real execution and motor imagery multi-class classification tasks are also crucial, but only a few kinds of literature research it. In addition, classification accuracy still has room for improvement, and the inter-individual variability problems in BCI applications need to be solved. To address these issues, we developed a novel model (RP-BCNNs) that combines the recurrence plot (RP) and Bayesian Convolutional Neural Networks (BCNNs). First, we employ an RP computation for preprocessed EEG signals of each channel and merge all RPs of all channels into one based on the weighted average method. Then, we feed the RP features into BCNNs to classify 2-class, 3-class, 4-class, and 5-class on real/imaginary movements classification tasks. The results show that the RP-BCNNs model outperforms the state-of-the-art methods, achieving average accuracies of 92.86%, 94.12%, 91.37%, 92.61% for real movements and 94.07%, 93.77%, 90.54%, 91.85% for imaginary movements. Our findings suggest that combining complex network methods with deep learning can improve the classification performance of EEG-based BCI systems (e.g., motor imagery, emotion recognition, and epileptic seizure classification).

Impacts of Cortical Regions on EEG-based Classification of Lexical Tones and Vowels in Spoken Speech.

Speech impairment is one of the most serious problems for patients with communication disorders, e.g., stroke survivors. The brain-computer interface (BCI) systems have shown the potential to alternatively control or rehabilitate the neurological damages in speech production. The effects of different cortical regions in speech-based BCI systems are essential to be studied, which are favorable for improving the performance of speech-based BCI systems. This work aimed to explore the impacts of different speech-related cortical regions in the electroencephalogram (EEG) based classification of seventy spoken Mandarin monosyllables carrying four vowels and four lexical tones. Seven audible speech production-related cortical regions were studied, involving Broca's and Wernicke's areas, auditory cortex, motor cortex, prefrontal cortex, sensory cortex, left brain, right brain, and whole brain. Following the previous studies in which EEG signals were collected from ten subjects during Mandarin speech production, the features of EEG signals were extracted by the Riemannian manifold method, and a linear discriminant analysis (LDA) was regarded as a classifier to classify different vowels and lexical tones. The results showed that when using electrodes from whole brain, the classifier reached the best performances, which were 48.5% for lexical tones and 70.0% for vowels, respectively. The vowel classification results under Broca's and Wernicke's areas, auditory cortex, or prefrontal cortex were higher than those under the motor cortex or sensory cortex. No such differences were observed in the lexical tone classification task.

A novel framework of MOPSO-GDM in recognition of Alzheimer's EEG-based functional network.

Most patients with Alzheimer's disease (AD) have an insidious onset and frequently atypical clinical symptoms, which are considered a normal consequence of aging, making it difficult to diagnose AD medically. But then again, accurate diagnosis is critical to prevent degeneration and provide early treatment for AD patients. This study aims to establish a novel EEG-based classification framework with deep learning methods for AD recognition. First, considering the network interactions in different frequency bands (δ, θ, α, β, and γ), multiplex networks are reconstructed by the phase synchronization index (PSI) method, and fourteen topology features are extracted subsequently, forming a high-dimensional feature vector. However, in feature combination, not all features can provide effective information for recognition. Moreover, combining features by manual selection is time-consuming and laborious. Thus, a feature selection optimization algorithm called MOPSO-GDM was proposed by combining multi-objective particle swarm optimization (MOPSO) algorithm with Gaussian differential mutation (GDM) algorithm. In addition to considering the classification error rates of support vector machine, naive bayes, and discriminant analysis classifiers, our algorithm also considers distance measure as an optimization objective. Finally, this method proposed achieves an excellent classification error rate of 0.0531 (5.31%) with the feature vector size of 8, by a ten-fold cross-validation strategy. These findings show that our framework can adaptively combine the best brain network features to explore network synchronization, functional interactions, and characterize brain functional abnormalities, which can improve the recognition efficiency of diseases. While improving the classification accuracy of application algorithms, we aim to expand our understanding of the brain function of patients with neurological disorders through the analysis of brain networks.

Identifying the Depression in Patients through EEG-Based Classification Using Spatial Information

Identifying the Depression in Patients through EEG-Based Classification Using Spatial Information

Review of EEG-based Classification of Depression Patients

Review of EEG-based Classification of Depression Patients

EEG-based classification of imagined digits using a recurrent neural network

Objective. In recent years, imagined speech brain–computer (machine) interface applications have been an important field of study that can improve the lives of patients with speech problems through alternative verbal communication. This study aims to classify the imagined speech of numerical digits from electroencephalography (EEG) signals by exploiting the past and future temporal characteristics of the signal using several deep learning models. Approach. This study proposes a methodological combination of EEG signal processing techniques and deep learning models for the recognition of imagined speech signals. EEG signals were filtered and preprocessed using the discrete wavelet transform to remove artifacts and retrieve feature information. To classify the preprocessed imagined speech neural signals, multiple versions of multilayer bidirectional recurrent neural networks were used. Main results. The method is examined by leveraging MUSE and EPOC signals from MNIST imagined digits in the MindBigData open-access database. The presented methodology’s classification performance accuracy was noteworthy, with the model’s multiclass overall classification accuracy reaching a maximum of 96.18% on MUSE signals and 71.60% on EPOC signals. Significance. This study shows that the proposed signal preprocessing approach and the stacked bidirectional recurrent network model are suitable for extracting the high temporal resolution of EEG signals in order to classify imagined digits, indicating the unique neural identity of each imagined digit class that distinguishes it from the others.