An efficient multi-scale CNN model with intrinsic feature integration for motor imagery EEG subject classification in brain-machine interfaces

Biomedical Signal Processing and Control, Journal Year: 2022, Volume and Issue: 74, P. 103496 - 103496

Published: Jan. 22, 2022

Language: Английский

---

Editorial: Multimodal approaches to investigating neural dynamics in cognition and related clinical conditions: integrating EEG, MEG, and fMRI data

Frontiers in Systems Neuroscience, Journal Year: 2025, Volume and Issue: 19

Published: Feb. 11, 2025

neuroanatomical information, behavioral data including performance evaluations on a working memory task, were analyzed as key indicator of the effects neurofeedback and associated cerebral hemodynamic changes over time. The findings indicated that training was effective in enhancing prefrontal activity maintaining cognitive function acute stroke patients. multidimensional approach enabled thorough examination relationship between brain structure, function, behavior.Jones, Keith G., et al. conducted detailed EEG patterns during propofol-induced burst suppression patients with treatment-resistant depression. Their study identifies distinct types activity, highlighting variability neural responses to anesthesia. Participants randomized receive either high-or low-dose propofol infusions. identified three 1-broadband bursts, 2-spindles, 3-low-frequency each varying significantly across subjects terms occurrence, spectral power, spatial scalp. Combining clinical (e.g., drug dosages patient characteristics) provided valuable insights into individualization anesthetic dosing diverse propofol, potentially leading more tailored treatment strategies practice.In another study, Mertiens, Sean, investigated how Parkinson's disease (PD) affects resting state networks (RSNs) using MEG measure phase-amplitude coupling (PAC). This combined T1-MRI scans. captured dynamic oscillations PAC within RSNs, while accurate source reconstruction signals. integration allows for precise mapping RSNs revealed significant alterations these PD compared healthy controls, particularly sensorimotor network (SMN). Levodopa medication normalized SMN toward similar though no observed optimal frequencies, suggesting levodopa primarily motor symptoms without influencing all same way.The results above-mentioned studies demonstrate combining different modalities can enhance process provide deeper diseases disorders. However, previously discussed, integrating multiple present challenges, some are focused improving addressing technical issues. Karittevlis, Christodoulos, introduce novel method analyzing evoked somatosensory stimulation, aimed at identifying early thalamus cortex. EEG-MEG often face challenges accurately determining due reliance complex models many assumptions. These be inaccurate because skull distorts electrical Additionally, stimulus trials complicate achievement consistent results. proposes straightforward bypasses modeling uses called virtual sensors (VS), which combines directly capture activity. improves accuracy elicited trial-to-trial variability, offering clearer reliable communication regions. By simplifying computational processes, authors potential applications real-time brain-computer interfaces, practical benefits multimodal approaches applied neuroscience.The by Dayarian, N, Khadem, A leverages recording MRI-based head new algorithm localization. Accurate localization is crucial diagnosing planning treatments neurological disorders such epilepsy, precisely location abnormal It also plays role surgical planning, helping minimize damage critical regions, aids monitoring progression tailoring individual patients, thus personalized medicine (Yang al., 2023).However, faces several geometry brain, complicates modeling. Traditional methods struggle heterogeneous properties tissues low resolution EEG, making it difficult differentiate closely situated sources. signals susceptible noise artifacts, muscle or eye movements, obscure true hinder (Michel Brunet, 2019;Hirata 2024). Many techniques rely assumptions about models, may not reflect anatomical variations, further impacting (Hirata co-registration model critical, any misalignment affect results.Dayarian, propose hybrid strengths boundary element (BEM) finite (FEM), address challenges. BEM effectively isotropic regions dipolar sources, FEM handles complex, anisotropic greater accuracy. validating this realistic achieves improvements forward problem solutions. demonstrates enhanced error metrics alone, value incorporating information from MRI improve precision neuroimaging analyses.Each imaging has its own advantages limitations, outlined Table 1. To fast dynamics, high temporal essential rapid Among non-invasive techniques, stand out their millisecondlevel resolution, them ideal studying processes. In contrast, fMRI fNIRS offer order seconds, limits applicability investigating dynamics.In most sensitive fMRI, followed fNIRS, MEG, EEG. High mapping. offers highest (in millimeters), allowing visualization active layers. lowest resolution. Although have been developed mitigate limitation, they volume conduction require density electrodes determine (Liu 2023).Apart factors usage complexity, portability, cost considerations when se Apart selecting method. leverage various limitations technique, recommended. simultaneous use two presents following sections.The primary advantage EEG-fMRI co-recording ability both (through data) data), resulting comprehensive understanding One issue need MR-compatible hardware, increase costs. systems designed non-magnetic induced voltages currents could arise radiofrequency gradient fields scans (Mele 2019;Scrivener, 2021;Warbrick, 2022). If adequately controlled, cause heating ECG leads, posing risks participants (Kugel 2003).To risks, careful design wires materials resistant built-in resistors necessary. addition, magnetic environment quality signals, reduce signal-to-noise ratio, post-processing influence MRI's fields, effect generally minimal disregarded. typically lying down, suitable requiring positions, sitting, standing, walking. Moreover, environment's cold temperatures intense uncomfortable especially children elderly, confounding Lastly, well-suited sleep disorder discomfort impracticality extended scanning sessions (Duyn, 2012;Mele 2019;Ebrahimzadeh 2022).Analyzing recordings unique requires specific methods. Common analysis include (Huster 2012;Abreu 2018):1. Symmetrical Approaches  Model-Based Techniques: involve creating mathematical understand data. complexities ensuring Despite mentioned frequently used than other approaches. preference likely maturity technologies relatively lower systems.While notable lack inability record sitting standing restricted duration, an noisy. Functional near-infrared spectroscopy (fNIRS) addresses measuring metabolic similarly but flexibility. influenced differences skin color hair type, positions (sitting, walking) periods. Consequently, EEG-fNIRS provides solution concerns portability systems. smaller maintenance requirements 2021;Uchitel 2021).One challenge configuring related interference. Ongoing technological development needed (Ahn Jun, 2017;Chen 2020;Li 2022).Pre-processing processing modality steps, introduces additional negative impact one modalities. solutions (Hossain 2022;Li 2022;Mughal 2022) facilitates data, thereby overall function.Simultaneous achieve necessity systems, adverse recordings. researchers employ (Im 2005;Plis 2010;Hall 2014;Cichy 2016):1. Minimize Temporal Overlap: Reduce concurrent time lessen 2. Use Precise Timing Triggers: Ensure alignment timing triggers. 3. Shield sensors: Protect external electromagnetic 4. Apply ICA: ICA separate signals.These help difficulties facilitating data.To MEG-fMRI co-recording, replacing advantageous. substitution reducing MEG's higher superior ratio common MEG-fNIRS MEG-fNIRS. difficulty achieving arises differing physical modality. For instance, configuration optodes necessitates placement align optimally measurement points (Ru 2022).To analyze integrate preprocessing fusion recommended:1. Alignment: Synchronize reference. MEG: filtering remove drifts high-frequency noise, artifacts. fNIRS: Perform baseline correction detrending manage slow wavelet adaptive handle motion physiological noise.3. Spatial Align coordinates reference.4. Data Standardization: comparability standardizing data.Feature Extraction Fusion: Extract relevant features datasets combine advanced canonical correlation (CCA), joint independent component (jICA), algorithms.These enable interpretation datasets.Along methods, functions Behavioral includes person's psychological tasks, response time, correlate behaviors, providing processed brain. electrophysiological reveal brain-behavioral interactions view functions.Thus, choice depends research type objectives. 2 summary, continue advance, pivotal developing diagnostic therapeutic psychiatric Future should focus analysis, optimizing sensor configurations, robust Such advancements deepen our pathology, opportunities neuroscience applications. Notably, review articles cover combination separately, collection original empirical value.

Language: Английский

---

Hybrid EEG-fNIRS brain-computer interface based on the non-linear features extraction and stacking ensemble learning

Journal of Applied Biomedicine, Journal Year: 2023, Volume and Issue: 43(2), P. 463 - 475

Published: April 1, 2023

The Brain-computer interface (BCI) is used to enhance the human capabilities. hybrid-BCI (hBCI) a novel concept for subtly hybridizing multiple monitoring schemes maximize advantages of each while minimizing drawbacks individual methods. Recently, researchers have started focusing on Electroencephalogram (EEG) and "Functional Near-Infrared Spectroscopy" (fNIRS) based hBCI. main reason due development artificial intelligence (AI) algorithms such as machine learning approaches better process brain signals. An original EEG-fNIRS hBCI system devised by using non-linear features mining ensemble (EL) approach. We first diminish noise artifacts from input signals digital filtering. After that, we use mining. These are "Fractal Dimension" (FD), "Higher Order Spectra" (HOS), "Recurrence Quantification Analysis" (RQA) features, Entropy features. Onward, Genetic Algorithm (GA) employed Features Selection (FS). Lastly, employ Machine Learning (ML) technique several namely, "Naïve Bayes" (NB), "Support Vector Machine" (SVM), "Random Forest" (RF), "K-Nearest Neighbor" (KNN). classifiers combined an recognizing intended activities. applicability tested publicly available multi-subject multiclass dataset. Our method has reached highest accuracy, F1-score, sensitivity 95.48%, 97.67% 97.83% respectively.

Language: Английский

---

Implementation of artificial intelligence and machine learning-based methods in brain–computer interaction

Computers in Biology and Medicine, Journal Year: 2023, Volume and Issue: 163, P. 107135 - 107135

Published: June 8, 2023

Brain–computer interfaces are used for direct two-way communication between the human brain and computer. Brain signals contain valuable information about mental state activity of examined subject. However, due to their non-stationarity susceptibility various types interference, processing, analysis interpretation challenging. For these reasons, research in field brain–computer is focused on implementation artificial intelligence, especially five main areas: calibration, noise suppression, communication, condition estimation, motor imagery. The use algorithms based intelligence machine learning has proven be very promising application domains, ability predict learn from previous experience. Therefore, within medical technologies can contribute more accurate subjects, alleviate consequences serious diseases or improve quality life disabled patients.

Language: Английский

---

NF-EEG: A generalized CNN model for multi class EEG motor imagery classification without signal preprocessing for brain computer interfaces

Biomedical Signal Processing and Control, Journal Year: 2024, Volume and Issue: 92, P. 106081 - 106081

Published: Feb. 8, 2024

Language: Английский

---

A hybrid CNN model for classification of motor tasks obtained from hybrid BCI system

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: Jan. 8, 2025

The Hybrid-Brain Computer Interface (BCI) has shown improved performance, especially in classifying multi-class data. Two non-invasive BCI modules are combined to achieve an classification which Electroencephalogram (EEG) and functional Near Infra-red Spectroscopy (fNIRS). Classifying contralateral ipsilateral motor movements is found challenging among the other mental activity signals. current work focuses on performance of deep learning methods like – Convolutional Neural Networks (CNN) Bidirectional Long-Short term memory (Bi-LSTM) a four-class execution Right Hand, Left Arm taken from CORE dataset. model was evaluated using metrics such as Accuracy, F1 score, Precision, Recall, AUC ROC curve. CNN Hybrid models have resulted 98.3% 99% accuracy respectively.

Language: Английский

---

Multimodal Multitask Neural Network for Motor Imagery Classification With EEG and fNIRS Signals

IEEE Sensors Journal, Journal Year: 2022, Volume and Issue: 22(21), P. 20695 - 20706

Published: Sept. 19, 2022

Brain–computer interface (BCI) based on motor imagery (MI) can control external applications by decoding different brain physiological signals, such as electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS). Traditional unimodal-based MI methods cannot obtain satisfactory classification performance due to the limited representation ability in EEG or fNIRS signals. Usually, signals have complementarity with sensitivity patterns. To improve recognition rate generalization of MI, we propose a novel end-to-end multimodal multitask neural network (M2NN) model fusion M2NN method integrates spatial–temporal feature extraction module, learning (MTL) module. Specifically, MTL module includes two tasks, namely one main task for auxiliary deep metric learning. This approach was evaluated using public dataset, experimental results show that achieved accuracy improvement 8.92%, 6.97%, 8.62% higher than unimodal signal (MEEG), HbR (MHbR), single-task (MDNN), respectively. Classification accuracies multitasking MEEG, MHbR, are improved 4.8%, 4.37%, compared EEG, HbR, MDNN, The best six methods, average 29 subjects being 82.11% ± 7.25%. effectiveness verified. is superior baseline state-of-the-art (SOTA) methods.

Language: Английский

---

Multimodal neuroimaging with optically pumped magnetometers: A simultaneous MEG-EEG-fNIRS acquisition system

NeuroImage, Journal Year: 2022, Volume and Issue: 259, P. 119420 - 119420

Published: June 29, 2022

Multimodal neuroimaging plays an important role in neuroscience research. Integrated noninvasive modalities, such as magnetoencephalography (MEG), electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS), allow neural activity related physiological processes the brain to be precisely comprehensively depicted, providing effective advanced platform study function. Noncryogenic optically pumped magnetometer (OPM) MEG has high signal power due its on-scalp sensor layout enables more flexible configurations than traditional commercial superconducting MEG. Here, we integrate OPM-MEG with EEG fNIRS develop a multimodal system that can simultaneously measure electrophysiology hemodynamics. We conducted series of experiments demonstrate feasibility robustness our MEG-EEG-fNIRS acquisition system. The complementary signals collected by imaging provide opportunities for wide range potential applications neurovascular coupling, wearable neuroimaging, hyperscanning brain-computer interfaces.

Language: Английский

---

Early-stage fusion of EEG and fNIRS improves classification of motor imagery

Frontiers in Neuroscience, Journal Year: 2023, Volume and Issue: 16

Published: Jan. 9, 2023

Many research papers have reported successful implementation of hybrid brain-computer interfaces by complementarily combining EEG and fNIRS, to improve classification performance. However, modality or feature fusion fNIRS was usually designed for specific user cases, which were generally customized hard be generalized. How effectively utilize information from the two modalities still unclear.In this paper, we conducted a study investigate stage bi-modal based on fNIRS. A Y-shaped neural network proposed evaluated an open dataset, fuses bimodal in different stages.The results suggests that early-stage significantly higher performance compared middle-stage late-stage configuration (N = 57, P < 0.05). With framework, average accuracy 29 participants reaches 76.21% left-or-right hand motor imagery task leave-one-out cross-validation, using data as inputs respectively, is same level state-of-the-art BCI methods data.

Language: Английский

---

From brain to movement: Wearables-based motion intention prediction across the human nervous system

Nano Energy, Journal Year: 2023, Volume and Issue: 115, P. 108712 - 108712

Published: July 18, 2023

Fueled by the recent proliferation of energy-efficient and energy-autonomous or self-powered nanotechnology-based wearable smart systems, human motion intention prediction (MIP) plays a critical role in wide range applications, such as rehabilitation assistive robotics, to enable more natural, biologically inspired, seamless integrated assistance task execution, including for elders physically impaired patients. With increasing complexity human-machine interactions need personalized assistance, there is growing demand real-time accurate MIP systems. This review aims provide comprehensive understanding interdisciplinary field MIP, under logic its physiological foundations, discussing state-of-the-art sensing technologies, brain-computer interfaces (BCI), electromyography (EMG), sensors, alongside relevant data processing techniques decoding algorithms. We emphasize importance fostering collaboration among scholars from different domains capture intricate dependencies between set stimuli responses central nervous system activation complex muscles joints that produce motion. By offering insights into advancements future prospects field, this seeks stimulate further research innovation rapidly evolving area prediction, where technologies understand respond intentions patterns, anticipating their needs.

Language: Английский

---