Wyniki wyszukiwania - BazTech

1

A dynamic spectrum driven network for enhanced multimodal emotion recognition with EEG and ECG signals

Samal Priyadarsini, Hashmi Mohammad Farukh

Biocybernetics and Biomedical Engineering

|

2026

|

Vol. 46, iss. 1

139--161

EN

For effective human-machine interaction, utilizing various physiological cues to recognize emotions is crucial. Using many physiological signals yields more accurate outcomes when recognizing human emotional states. This study introduces a new approach called DSDNet (Dynamic spectrum driven network) to emotion recognition using Electroencephalogram (EEG) and Electrocardiogram (ECG) signals. The method involves a dynamic time frequency analysis technique that combines synchrosqueezed transform with short time fast fractional Fourier transform. The signals are divided into segments, and the corresponding time-frequency spectrograms from EEG and ECG signals are combined for additional assessment and the importance of these spectrogram features are visualized by using SHAP deep explainer. Subsequently, these spectrogram features are provided to a simple efficient convolutional neural network for classification. The proposed approach utilized the DREAMER and AMIGOS datasets for development and comparison with several high-performance algorithms. This approach surpassed the most notable results in the existing literature, with an accuracy of 98.6%, 98.9%, and 99.2% for the valence, arousal, and dominance categories respectively, when applied to the DREAMER dataset. Similarly, when applied to the AMIGOS dataset, it achieved accuracies of 98.8%, 99.5%, and 99.4% for all three categories. Therefore, the findings of this research indicate that by incorporating various physiological signals and modern approaches in the field of human-machine interaction, it is possible to greatly enhance the accuracy of emotion detection results.

2

Decoding human movement intentions from EEG signals using recurrent neural networks

Kleczyński Daniel, Kluska Jacek

Zeszyty Naukowe Politechniki Rzeszowskiej. Elektrotechnika

|

2025

|

Vol. 31

43--53

EN

The present study develops a model for recognizing movement intentions from Electroencephalography (EEG) signals using various Recurrent Neural Network (RNN) configurations, with a focus on Long Short-Term Memory (LSTM) networks. Experiments demonstrate that, with proper sample allocation and class balance, the model achieves an average accuracy of 0.9815±0.0025 and an average Receiver Operating Characteristic area (AUC) of 0.9989±0.0004 when training and test data include the same subjects. The best-performing LSTM model - augmented with a fully connected layer - was configured with a hidden layer size of 233, learning rate of 3.872 × 10−4, 3 layers, dropout of 0.3773, and sequence length of 457. However, when test subjects were completely excluded from training, the model’s accuracy did not exceed 50%, suggesting significant inter-subject variability or limitations in generalization. This work contributes to advancing Brain-Computer Interfaces for applications such as prosthetic control and provides insights into the prerequisites for effective EEG signal utilization.

3

Seafarer Mental Workload Assessment Using a Hybrid Deep Learning Model

Jiang Ruonan, Fan Shiqi

TransNav : International Journal on Marine Navigation and Safety of Sea Transportation

|

2025

|

Vol. 19 No. 1

55--60

EN

Human errors in maritime operations are closely linked to seafarers' mental workload; however, traditional assessment methods lack real-time neurocognitive resolution. This study introduces a novel psychophysiological framework that integrates electroencephalography (EEG) analysis with deep learning to objectively quantify seafarers' mental workload during onboard operations. A high-fidelity bridge simulator was utilized to generate critical maritime scenarios, including ship encounters, narrow channel navigation, poor visibility, and emergency responses. High-density EEG signals were analyzed to extract spectral features (Gamma, Beta, Alpha, Theta, Delta). A hybrid Convolutional Neural Network-Bidirectional Long Short-Term Memory (CNN-BiLSTM) model was proposed to classify workload states of seafarers, combining Convolutional Neural Network (CNN)-extracted frequency patterns with Bidirectional Long Short-Term Memory (Bi-LSTM)-captured temporal dynamics, which achieves 96% accuracy. Furthermore, SHAP interpretability analysis indicated that Theta and Alpha frequencies are key indicators in distinguishing between high and low workloads for seafarers. These results provide a quantitative tool for cognitive assessment of seafarers in maritime training and serve as a guideline for workload allocation in ship bridge teams for shipping companies and maritime authorities.

4

Toward the integration of mixed reality and brain-computer interfaces based on code-modulated visual evoked potentials

Moreno-Calderón Selene, Martínez-Cagigal Víctor, Martín-Fernández Ana, Santamaría-Vázquez Eduardo, Hornero Roberto

Biocybernetics and Biomedical Engineering

|

2025

|

Vol. 45, iss. 3

528--538

EN

Background and objective: Brain-computer interface (BCI) systems can assist individuals with severe motor disabilities by enabling communication through their brain signals using spellers, which allow selecting commands from a set of options. For this technology, accuracy, speed and user comfort are essential. Code-modulated visual evoked potentials (c-VEPs) have demonstrated promising performance in BCI control. Integrating BCI systems with mixed reality (MR) could provide portability and autonomy. However, to the best of our knowledge, no existing studies have explored the feasibility of combining MR with c-VEP-based BCIs. This study aims to: (1) evaluate the performance of integrating MR with c-VEP-based BCIs and (2) study the visual fatigue induced by c-VEPs compared to traditional screen. Methods: Twenty participants used a 36-character speller to select words in both MR and traditional screen conditions. Metrics like accuracy and information transfer rate (ITR) were measured. Usability and eyestrain were evaluated through questionnaires. Results: The integration of MR with c-VEPs achieved an accuracy of 96.71% and an ITR of 27.55 bits/min, compared to 95.98% accuracy and 27.10 bits/min for the conventional screen condition. The questionnaires revealed minimal levels of visual fatigue in both conditions and high usability. No significant differences were observed between conditions in terms of performance or visual fatigue. Conclusions: The c-VEP-based speller with MR-BCI technology proved feasible, achieving performance levels similar to the conventional setup, with high accuracy in both conditions. The study also found comparable visual fatigue between MR and traditional screens, supporting the practicality of MR integration in BCI systems.

5

Reevaluating performance in c-VEP BCIs: The impact of calibration time

Martínez-Cagigal Víctor, Santamaría-Vázquez Eduardo, Pérez-Velasco Sergio, Martín-Fernández Ana, Hornero Roberto

Biocybernetics and Biomedical Engineering

|

2025

|

Vol. 45, iss. 4

685--696

EN

Code-modulated visual evoked potentials (c-VEP) have demonstrated high performance in non-invasive brain-computer interfaces (BCIs). Recently, research has begun to consider practical aspects such as visual comfort, where non-binary sequences and variations in the spatial frequency of stimuli play significant roles. However, calibration requirements remain underexplored in performance comparisons. This study aims to analyze a multi-variable tradeoff crucial to the practical application of c-VEP-based BCIs: decoding accuracy, decoding speed, and calibration time. Visual comfort is retrospectively evaluated using two pre-recorded datasets. Models were trained with increasing calibration cycles and tested across varying decoding times, depicting learning and decoding curves. The datasets comprised 32 healthy subjects, and featured different stimulus paradigms: plain non-binary stimuli and checkerboard-like binary stimuli with spatial frequency variations. Results showed that all conditions achieved over 97% grand-averaged accuracy with sufficient calibration. However, a clear tradeoff emerged between calibration duration and performance. Achieving 95% average accuracy within a 2 s decoding window required mean calibration durations of 28.7±19.0 s for binary stimuli, or 148.7±72.3 s for non-binary stimuli. The binary checkerboard-based condition with a spatial frequency of 1.2 c/º (C016) proved to be particularly effective, achieving over 95% accuracy within 2 s decoding window using only 7.3 s of calibration, and reporting a significant improvement in visual comfort. A minimum calibration time of 1 min was considered essential to adequately estimate the brain response, critical in template-matching paradigms. In conclusion, achieving optimal c-VEP performance requires balancing calibration duration, decoding speed and accuracy, and visual comfort.

6

EEG-based analysis of ASMR Stimuli: a pilot study of neuropsychological responses through conventional vs. bone-conduction headphones

Seifzadeh Sahar, Kostek Bożena, Poerio Giulia L.

Vibrations in Physical Systems

|

2024

|

Vol. 35, nr 3

art. no. 2024307

EN

In this study, the impact of ASMR (Autonomous Sensory Meridian Response) experiences delivered through different types of headphones was evaluated with respect to neural responses and anxiety levels. The EEG data of a 24-year-old participant was recorded while he underwent ASMR stimulation using conventional and bone-conduction headphones. The State-Trait Anxiety Inventory (STAI) assessed anxiety levels before and after ASMR stimulation, showing decreased state anxiety following intervention. Based on spectral analysis of Electroencephalography (EEG) data, significant differences were found between headphone types and cognitive tasks (Mathematical calculations). Using conventional headphones, gamma activity was evident in the posterior brain regions, suggesting that headphone type may influence ASMR-induced neural activity. The pilot study findings emphasize the importance of refining auditory delivery methods for clinical applications to maximize ASMR efficacy and therapeutic outcomes.

7

Klasyfikacja sygnałów EEG w interfejsach mózg-komputer z wykorzystaniem sieci neuronowych

Galas Krzysztof

Studia i Materiały Informatyki Stosowanej

|

2024

|

T. 16, nr 3

18--23

PL

Celem pracy jest zaprezentowanie metod klasyfikacji sygnałów EEG w interfejsach mózg-komputer (BCI) z wykorzystaniem sieci neuronowych. Dzięki ich zdolności do modelowania złożonych zależności w danych, możliwe jest skuteczniejsze rozpoznawanie wzorców aktywności mózgowej, co przyczynia się do poprawy dokładności i szybkości działania systemów BCI. W pracy omówiono architektury sieci neuronowych wykorzystywane do analizy sygnałów EEG, takie jak sieci konwolucyjne (CNN) czy rekurencyjne (RNN). Badania pokazują, że te metody mają ogromny potencjał w zastosowaniach takich jak sterowanie urządzeniami wspomagającymi, komunikacja oraz rozrywka.

EN

The aim of this paper is to present methods for classifying EEG signals in brain-computer interfaces (BCIs) using neural networks. Thanks to their ability to model complex relationships in the data, it is possible to recognise patterns of brain activity more effectively, which contributes to improving the accuracy and speed of BCI systems. This paper discusses neural network architectures used to analyse EEG signals, such as convolutional networks (CNNs) or recurrent networks (RNNs). The research shows that these methods have immense potentialin applications such as assistive device control, communication,and entertainment.

8

Entropy-based feature extraction for classification of EEG signal using Lifting Wavelet Transform

Ananthi A., Subathra M.S.P., George S. Thomas, Sairamya N.J.

Przegląd Elektrotechniczny

|

2024

|

R. 100, nr 9

146--150

EN

In the realm of Brain-Computer Interface (BCI), a crucial hurdle lies in effectively classifying Motor Imagery (MI) signals. Numerous techniques have been developed for Electroencephalogram (EEG) signal-based MI classification. The proposed system transforms EEG signals into various representations through Lifting Wavelet Transform (LWT). Long Short Term Memory (LSTM) is employed for classifying the extracted feature vectors in each line. The performance of this method is evaluated on the PhysioNet database, specifically for distinguishing between right and left hand imagery move. The strategy,resulting in 100% accuracy in 19 out of 72 wavelet families of LWT. This combination proves to be a highly efficient tool for BCI-based EEG analysis, showcasing its potential as a resourceful solution in this domain.

PL

W obszarze interfejsu mózg-komputer (BCI) kluczową przeszkodą jest skuteczna klasyfikacja sygnałów obrazowania motorycznego (MI). Opracowano liczne techniki klasyfikacji MI na podstawie sygnału elektroencefalogramu (EEG). Proponowany system przekształca sygnały EEG na różne reprezentacje za pomocą transformacji falkowej Lifting Wavelet Transform (LWT). Pamięć długoterminowa Long Short Term Memory (LSTM) jest wykorzystywana do klasyfikowania wyodrębnionych wektorów cech w każdej linii. Wydajność tej metody jest oceniana w bazie danych PhysioNet, w szczególności w celu rozróżnienia ruchu obrazowania prawej i lewej ręki. Strategia ta zapewnia 100% dokładność w 19 z 72 rodzin falek LWT. Ta kombinacja okazuje się wysoce wydajnym narzędziem do analizy EEG opartej na BCI, pokazując swój potencjał jako zasobnego rozwiązania w tej dziedzinie.

9

Analysis of Event-Related potentials for emotion recognition

Majkowski Andrzej, Kołodziej Marcin, Rak Remigiusz Jan

Przegląd Elektrotechniczny

|

2024

|

R. 100, nr 5

290--293

EN

The primary objective of this study was to determine the feasibility of classifying emotions into three categories (positive, negative, and neutral) using event-related potentials (ERPs) for individual users. Visual stimuli from the International Affective Picture System (IAPS) database were utilized. Various features, such as signal samples, discrete wavelet transform, discrete Fourier transform, and discrete cosine transform, were computed from one-second electroencephalographic signal (EEG) segments following the presentation of the stimulus. For the classification task, a one-nearest neighbor classifier (1-NN) was employed. The research yielded a system for preprocessing and classifying emotions. The study involved eight participants. The experiments presented in this paper demonstrate the possibility of distinguishing emotions into three categories (pleasant, unpleasant, and neutral) for a single user, achieving an average accuracy level of 87%. However, when considering all users collectively, we achieved a classification accuracy of 96%.

PL

Głównym celem artykułu było określenie możliwości klasyfikacji emocji w podziale na trzy kategorie (pozytywne, negatywne i neutralne) przy użyciu potencjałów wywołanych (ERPs) dla poszczególnych użytkowników. Wykorzystano bodźce wizualne z bazy danych International Affective Picture System (IAPS). Jako cechy zastosowano: próbki sygnału, dyskretna transformacja falkowa, dyskretna transformacja Fouriera oraz dyskretna transformacja kosinusowa, uzyskane z jednosekundowych segmentów sygnału elektroencefalograficznego (EEG) po prezentacji bodźca. Do zadania klasyfikacji zastosowano klasyfikator najbliższego sąsiada (1-NN). W wyniku prac powstał system do klasyfikowania emocji. W badaniu uczestniczyło ośmioro uczestników. Eksperymenty przedstawione w tym artykule pokazują możliwość rozróżniania emocji na trzy kategorie (przyjemne, nieprzyjemne i neutralne) dla jednego użytkownika, osiągając średni poziom dokładności 87%. Jednakże, biorąc pod uwagę wszystkich użytkowników łącznie, osiągnięto dokładność klasyfikacji na poziomie 96%.

10

EEG based emotion analysis using reinforced spatio-temporal attentive graph neural and contextnet techniques

Devi C. Akalya, Renuka D. Karthika

Journal of Automation Mobile Robotics and Intelligent Systems

|

2024

|

Vol. 18, No. 3

61--68

EN

EEG-based emotion classification is considered to separate and observe the mental state or emotions. Emotion classification using EEG is used for medical, security and other purposes. Several deep learning and machine learning strategies are employed to classify the EEG emotion signals. They do not provide sufficient accuracy and have higher complexity and high error rate. In this manuscript, a novel Reinforced Spatio-Temporal Attentive Graph Neural Networks (RSTAGNN) and ContextNet for emotion classification with EEG signals is proposed (RSTAGNN-ContextNet-GWOA-EEG-EA). Here, the input EEG signals are taken from two benchmark datasets,namely DEAP and K-EmoCon datasets. Then, the input EEG signals are pre-processed,and the fea- tures are extracted utilizing ContextNet with Global Principal Component Analysis (GPCA). After that, the EEG signal emotions are classified using Reinforced Spatio- Temporal Attentive Graph Neural Networks method. RSTAGNN weight parameters are optimized under the Glowworm Swarm Optimization Algorithm (GWOA). The proposed model classifies the EEG signal emotions with high accuracy. The efficacy of the proposed method using the DEAP dataset attains higher accuracy by 24.05%, 12.64% related to existing systems, like Multi-domain feature fusion for emotion classification (DWT-SVM-EEG- EA-DEAP), EEG emotion finding utilizing fusion mode of graph CNN with LSTM (GCNN-LSTM-EEG-EA-DEAP) respectively. The efficiency of the proposed method using the K-EmoCon dataset attains higher accuracy 32.64%, 15.65% related to existing systems, like Toward Robust Wearable Emotion Realization along Contrastive Repre- sentation Learning (CAT-EEG-EA-K-EmoCon) and Human Emotion Recognition using Physiological Signals (CAT- EEG-EA-K-EmoCon) respectively.

11

Probabilistic and explainable modeling of Phase-Phase Cross-Frequency Coupling patterns in EEG. Application to dyslexia diagnosis

Castillo-Barnes Diego, Gallego-Molina Nicolás J., Formoso Marco A., Ortiz Andrés, Figueiredo Patrícia, Luque Juan L.

Biocybernetics and Biomedical Engineering

|

2024

|

Vol. 44, no. 4

814--823

EN

This work explores the intricate neural dynamics associated with dyslexia through the lens of Cross-Frequency Coupling (CFC) analysis applied to electroencephalography (EEG) signals evaluated from 48 seven-year-old Spanish readers from the LEEDUCA research platform. The analysis focuses on CFS (Cross-Frequency phase Synchronization) maps, capturing the interaction between different frequency bands during low-level auditory processing stimuli. Then, making use of Gaussian Mixture Models (GMMs), CFS activations are quantified and classified, offering a compressed representation of EEG activation maps. The study unveils promising results specially at the Theta-Gamma coupling (Area Under the Curve = 0.821), demonstrating the method’s sensitivity to dyslexia-related neural patterns and highlighting potential applications in the early identification of dyslexic individuals.

12

Detection of Attention Deficit Hyperactivity Disorder based on EEG feature maps and deep learning

Karabiber Cura Ozlem, Akan Aydin, Kocaaslan Atli Sibel

Biocybernetics and Biomedical Engineering

|

2024

|

Vol. 44, no. 3

450--460

EN

Attention Deficit Hyperactivity Disorder (ADHD) is a neurological condition, typically manifesting in childhood. Behavioral studies are used to treat the illness, but there is no conclusive way to diagnose it. To comprehend changes in the brain, electroencephalography (EEG) signals of ADHD patients are frequently examined. In the proposed study, we introduce EEG feature map (EEG-FM)-based image construction to input deep learning architectures for classifying ADHD. To demonstrate the effectiveness of the proposed method, EEG data of 15 ADHD patients and 18 control subjects are analyzed and detection performance is presented. EEG-FMbased images are obtained using both traditional time domain features used in EEG analysis, such as Hjorth parameters (activity, mobility, complexity), skewness, kurtosis, and peak-to-peak, and nonlinear features such as the largest Lyapunov Exponent, correlation dimension, Hurst exponent, Katz fractal dimension, Higuchi fractal dimension, and approximation entropy. EEG-FM-based images are used to train DarkNet19 architecture and deep features are extracted for each image dataset. Fewer deep features are chosen for each image dataset using the Minimum Redundancy Maximum Relevance (mRMR) feature selection method, and the concatenated deep feature set is created by merging the selected features. Finally, various machine learning methods are used to classify the concatenated deep features. Our EEG-FM and DarkNet19-based approach yields classification accuracies for ADHD between 96.6% and 99.9%. Experimental results indicate that the use of EEG-FM-based images as input to DarkNet19 architecture gives significant advantages in the detection of ADHD.

13

Classification of motor imagery EEG signals using wavelet scattering transform and Bi-directional long short-term memory networks

Zhang Hongyuan, Zhao Zijian, Liu Chong, Duan Miao, Lu Zhiguo, Wang Hong

Biocybernetics and Biomedical Engineering

|

2024

|

Vol. 44, no. 4

874--884

EN

A brain-computer interface (BCI) is a technology that creates a communication path between the brain and external devices. Raw EEG data in BCI contain a large amount of complex information, but only some of it needs to be focused on in research. So Feature extraction and classification play an important role in BCI by reducing the data dimensionality and improving the accuracy of subsequent classification. Wavelet scattering transform is an emerging feature extraction method that generates time-shift invariant representations of EEG signals. We applied the wavelet scattering transform to extract features from motor imagery EEG signals, and utilized these features for classification purposes. To achieve this, we proposed a new method that combines wavelet scattering transform with a bidirectional long short-term memory (BiLSTM) network in a fusion deep learning network. Wavelet scattering transform can deeply mine the feature information in EEG signals. In the classification stage, multiple time window features obtained in the scattering transform are sent to the BiLSTM network for classification. The final result will be determined by a vote. In addition, for the processing of raw EEG data, we proposed a time-step based time window strategy that can better utilize the small dataset. This operation can obtain EEG data of multiple time steps. The proposed method was validated using BCI competition II dataset III and BCI competition IV dataset 2b. The results show that the proposed method in this paper can effectively improve the accuracy of motor imagery EEG and provide a new idea for the feature extraction and classification research of motor imagery brain-computer interface.

14

A linear-attention-combined convolutional neural network for EEG-based visual stimulus recognition

Huang Junjie, Chen Wanzhong, Zhang Tao

Biocybernetics and Biomedical Engineering

|

2024

|

Vol. 44, no. 2

369--379

EN

The recognition task of visual stimuli based on EEG (Electroencephalogram) has become a major and important topic in the field of Brain-Computer Interfaces (BCI) research. Although the underlying spatial features of EEG can effectively represent visual stimulus information, it still remains a highly challenging task to explore the local-global information of the underlying EEG to achieve better decoding performance. Therefore, in this paper we propose a deep learning architecture called Linear-Attention-combined Convolutional Neural Network (LACNN) for visual stimuli EEG-based classification task. The proposed architecture combines the modules of Convolutional Neural Networks (CNN) and Linear Attention, effectively extracting local and global features of EEG for decoding while maintaining low computational complexity and model parameters. We conducted extensive experiments on a public EEG dataset from the Stanford Digital Repository. The experimental results demonstrate that LACNN achieves an average decoding accuracy of 54.13% and 29.83% in 6-category and 72-exemplar classification tasks respectively, outperforming the state-of-the-art methods, which indicates that our method can effectively decode visual stimuli from EEG. Further analysis of LACNN shows that the Linear Attention module improves the separability between different category features and localizes key brain region information that aligns with the paradigm principles.

15

Predicting states of epilepsy patients using deep learning models

Sidaoui Boutkhil

Applied Computer Science

|

2024

|

Vol. 20, no. 2

109--125

EN

In this study, the authors present and scrutinize two deep learning models designed for predicting the states of epilepsy patients by utilizing extracted data from their brain's electrical activities recorded in electroencephalography (EEG) signals. The proposed models leverage deep learning networks, with the first being a recurrent neural network known as Long Short-Term Memory (LSTM), and the second a non-recurrent network in the form of a Deep Feedforward Network (DFN) architecture. To construct and execute the DFN and LSTM architectures, the authors rely on 22 characteristics extracted from diverse EEG signals, forming a comprehensive dataset from five patients. The primary goal is to forecast impending epilepsy seizures and categorize three distinct states of brain activity in epilepsy patients. The models put forward yield promising results, particularly in terms of classification rates, across various preceding seizure timeframes ranging from 5 to 50 minutes.

16

Using high-tech tools for consumer buying decisions of FMCG

Wiścicka-Fernando Małgorzata, Biercewicz Konrad

Zeszyty Naukowe. Organizacja i Zarządzanie / Politechnika Śląska

|

2023

|

z. 174

387--402

EN

Purpose: The main aim of this paper is to explore consumer decisions and emotions during shopping at the self-service store with fast-moving consumer goods (FMCG). Design/methodology/approach: The subject of the study is to assess the impact of emotions during the choice-making process on consumers' buying decisions. The respondents are citizens of the West Pomeranian region, Poland. The survey was conducted using state-of-the-art data acquisition technologies, i.e., Virtual Reality and EEG. An interview was also used as a complementary form. The research was both qualitative and quantitative, with a research sample of 34 respondents and took place in the virtual world. The researchers used primary data. The results presented here are part of a broader research project that used a triangulation of research methods to allow a deeper analysis of the conscious and unconscious aspects of the subjects. Findings: The research provided independent data on consumer emotions. The authors identified 4 groups of emotions that appeared during the selection of a product and were highly differentiated and strongly dependent on such characteristics as consumer type and gender. It has also been noticed that the longer a product is held, the lower emotional “sleepiness’. Research limitations/implications: One of the main limitations is the data collection process, which is relatively expensive, so the sample size is limited. The results obtained can be a signpost for a researcher who would like to use this new technology for further research. Practical implications: The results obtained can be used by shop managers in planning the sales activities or shop space to help the customer decide. Originality/value: In the research was used an innovative combination of virtual reality (VR) equipment and an electroencephalogram (EEG). To the best of the authors' knowledge, the results of a study from the FMCG industry using both devices simultaneously have never been published.

17

Patient specific channel optimization using entropy and CNN deep learning for epileptic seizure prediction

Cherouati Brahim, Senouci Mohamed

Przegląd Elektrotechniczny

|

2023

|

R. 99, nr 7

106--110

EN

Predicting epileptic seizures in advance improves greatly the life of epileptic patients. In this paper we present a new approach based on patient specific channel optimization using four different features namely entropy, variance, kurtosis and skewness. After selecting three best channels for each method, we then use Convolutional Neural Network (CNN) to classify raw EEG signal in order to discriminate between interictal and preictal state. With entropy, our method achieves a good degree of prediction in terms of accuracy 97.09%, sensitivity 97.67% and specificity 96.51% for patient 01 using channels 4, 8 and 20.

PL

Przewidywanie napadów padaczkowych z wyprzedzeniem znacznie poprawia życie chorych na padaczkę. W tym artykule prezentujemy nowe podejście oparte na optymalizacji kanałów specyficznych dla pacjenta przy użyciu czterech różnych metod, a mianowicie entropii, wariancji, kurtozy i skośności. Po wybraniu trzech najlepszych kanałów dla każdej z metod, wykorzystujemy Neuronową Sieć Konwolucyjną (CNN) do klasyfikacji surowego sygnału EEG w celu rozróżnienia pomiędzy stanem międzynapadowym i przednapadowym. Dzięki entropii nasza metoda osiąga dobry stopień predykcji w zakresie dokładności 97,09%, czułości 97,67% i specyficzności 96,51% dla pacjenta 01 przy użyciu kanałów 4, 8 i 20.

18

Outlier detection in EEG signals

Duraj Agnieszka, Chomątek Łukasz

Przegląd Elektrotechniczny

|

2023

|

R. 99, nr 1

237--240

EN

In this paper, the topic of detection of outliers in EEG signals was discussed, which facilitates making decisions about the diagnosis of a patient based on this study. We used two methods to detect outliers: the support vector machine and the k nearest neighbors method. The experiments were performed on a publicly available dataset containing EEG test results for 500 patients. The obtained results showed that the methods we used allow for the outlier detection efficiency at the level of 93%.

PL

W niniejszej pracy podjęto temat detekcji wyjątków w sygnałach EEG, co pozwala na ułatwienie podejmowania decyzji co do diagnozy pacjenta na podstawie tego badania. Do detekcji wyjątków wykorzystaliśmy dwie metody: maszynę wektorów nośnych i metodę k najblizszych sąsiadów. Eksperymenty zostały przeprowadzone na ogólnodostępnym zbiorze danych zawieraj ącym wyniki badania EEG dla 500 pacjentów. Uzyskane wyniki pokazały, że u żyte przez nas metody pozwalają na uzyskanie skuteczności detekcji wyjątków na poziomie 93%.

19

Effect of input channel reduction on EEG seizure detection

Kabir Md Shaikh Abrar, Farhan Faisal, Siddiquee Adnan Amin, Baroi Oli Lowna, Marium Taniza, Rahimi Md. Jakaria

Przegląd Elektrotechniczny

|

2023

|

R. 99, nr 12

195--200

EN

In this study, the effectiveness of six machine learning and eight deep learning algorithms in analyzing electroencephalogram (EEG) signals for detecting epileptic seizures has been investigated. The study utilizes 14 channels in the EMOTIV EPOC+ device which is based on international 10-20 system. To find out the most informative and sensitive channel, one of the 14 channels has been dropped one at a time. The accuracy values were determined for all the methods using two different publicly available datasets: the Guinea-Bissau epilepsy dataset and the Nigeria epilepsy dataset. In case of machine learning models, the performance of SVM classifier performs best with maximum accuracy of 83.2% (Guinea-Bissau) and 77% (Nigeria) without excluding any channels. No significant performance degradation has been observed for single channel exclusion of this classifier. Among the deep learning models, the four best performing models in terms of accuracy are CNN-LSTM (92.5%), IC-RNN (91.8%), ChronoNet (91.1%) and C-DRNN (88.6%). After excluding one channel at a time and investigating their effect on the performance of the four DL models, it has been observed that the most significant and most sensitive channels lie within the frontal and parietal zone. This finding will be very useful in practice as it indicates that the electrodes in the frontal and parietal zone should be placed with absolute precision for accurate diagnosis of the diseases. In addition, this study also explore the effectiveness of the selected classifiers in detecting seizure in case of failure of any particular EEG signal channel.

PL

W tym badaniu zbadano skuteczność sześciu algorytmów uczenia maszynowego i ośmiu algorytmów głębokiego uczenia się w analizie sygnałów elektroencefalogramu (EEG) w celu wykrywania napadów padaczkowych. W badaniu wykorzystano 14 kanałów w urządzeniu EMOTIV EPOC+ opartym na międzynarodowym systemie 10-20. Aby znaleźć najbardziej pouczający i wrażliwy kanał, usuwano jeden z 14 kanałów na raz. Wartości dokładności określono dla wszystkich metod przy użyciu dwóch różnych publicznie dostępnych zbiorów danych: zbioru danych dotyczących padaczki w Gwinei Bissau i zbioru danych dotyczących padaczki w Nigerii. W przypadku modeli uczenia maszynowego wydajność klasyfikatora SVM jest najlepsza przy maksymalnej dokładności 83,2% (Gwinea Bissau) i 77% (Nigeria) bez wykluczania jakichkolwiek kanałów. Nie zaobserwowano znaczącego pogorszenia wydajności w przypadku wykluczenia pojedynczego kanału tego klasyfikatora. Wśród modeli głębokiego uczenia się cztery modele o najlepszych wynikach pod względem dokładności to CNN-LSTM (92,5%), IC-RNN (91,8%), ChronoNet (91,1%) i C DRNN (88,6%). Po wykluczeniu jednego kanału na raz i zbadaniu ich wpływu na działanie czterech modeli DL zaobserwowano, że najważniejsze i najbardziej czułe kanały znajdują się w strefie czołowej i ciemieniowej. Odkrycie to będzie bardzo przydatne w praktyce, gdyż wskazuje, że elektrody w strefie czołowej i ciemieniowej powinny być umieszczone z absolutną precyzją, aby umożliwić trafną diagnostykę schorzeń. Ponadto w badaniu tym zbadano również skuteczność wybranych klasyfikatorów w wykrywaniu napadów w przypadku awarii dowolnego konkretnego kanału sygnałowego EEG.

20

An improved performance of support vector machine to classify EEG motor imagery based on differential asymmetry

Putranto Yulianto Tejo, Putra Oddy Virgantara, Hafidz Isa, Sardjono Arief, Hariadi Mochamad, Putranto Yulianto Tejo

Przegląd Elektrotechniczny

|

2023

|

R. 99, nr 6

196--203

EN

One challenge in EEG motor imaging is th e low signal-to-noise ratio of brain signals. Its emergence in the accurate rendition of brain signals varies significantly from person to person. Here, we propose a framework to classify tasks based on fusion features using a Support Vector Machine. Our features are acquired from Discrete Wavelet Transform and Empirical Mode Decomposition. Subsequently, the disparity between measurements of left and right brain signals was calculated. Our proposed work significantly improves accuracy from 83.29 % to 93.16 % compared to previous work.

PL

Jednym z wyzwań w obrazowaniu motorycznym EEG jest niski stosunek sygnału do szumu sygnałów mózgowych. Jego pojawienie się w dokładnym przekazywaniu sygnałów mózgowych różni się znacznie w zależności od osoby. Tutaj proponujemy ramy do klasyfikowania zadań w oparciu o funkcje fuzji przy użyciu maszyny wektorów nośnych. Nasze funkcje są uzyskiwane z dyskretnej transformacji falkowej i dekompozycji trybu empirycznego. Następnie obliczono rozbieżność między pomiarami sygnałów lewego i prawego mózgu. Nasza proponowana praca znacznie poprawia dokładność z 83,29% do 93,16% w porównaniu z poprzednią pracą.