Electroencephalography-Based Motor Imagery Classification Using Multi-Scale Feature Fusion and Adaptive Lasso DOI Creative Commons
Shimiao Chen, Nan Li, Xiangzeng Kong

et al.

Big Data and Cognitive Computing, Journal Year: 2024, Volume and Issue: 8(12), P. 169 - 169

Published: Nov. 25, 2024

Brain–computer interfaces, where motor imagery electroencephalography (EEG) signals are transformed into control commands, offer a promising solution for enhancing the standard of living disabled individuals. However, performance EEG classification has been limited in most studies due to lack attention complementary information inherent at different temporal scales. Additionally, significant inter-subject variability sensitivity biological motion poses another critical challenge achieving accurate subject-dependent manner. To address these challenges, we propose novel machine learning framework combining multi-scale feature fusion, which captures global and local spatial from different-sized segmentations, adaptive Lasso-based selection, mechanism adaptively retaining informative features discarding irrelevant ones. Experimental results on multiple public benchmark datasets revealed substantial improvements classification, rates 81.36%, 75.90%, 68.30% BCIC-IV-2a, SMR-BCI, OpenBMI datasets, respectively. These not only surpassed existing methodologies but also underscored effectiveness our approach overcoming specific challenges classification. Ablation further confirmed efficacy both analysis selection mechanisms. This marks advancement decoding signals, positioning it practical applications real-world BCIs.

Language: Английский

Complex Motion Behavior and Synchronization Analysis of Heterogeneous Neural Network DOI
Wu Xiao, Fuhong Min, Haodong Li

et al.

IEEE Transactions on Circuits and Systems I Regular Papers, Journal Year: 2024, Volume and Issue: 71(12), P. 5618 - 5627

Published: April 22, 2024

The study on the dynamical behaviors of coupled heterogeneous neural network, including bifurcation orbits, synchronization, especially unstable firing behaviors, may have great significance for diagnosis and guarding against brain diseases. To investigate this matter in depth, discrete implicit mapping method can be employed assessing which is with Hindmarsh-Rose FitzHugh-Nagumo neuron models paper. trees periodic motions, exhibiting intricate dynamic are precisely demonstrated by maniputing coupling strength. transitions from period-1 to period-8, period-3 period-12, period-4 period-16 period-5 period-10 will achieved through saddle bifurcations period doubling bifurcations. corresponding stable patterns observed nodes phase diagrams, time-histories deviations membrane potential diagrams. Meanwhile, patterns, using particular method, also obtained, cannot calculated numerical due its accumulative errors. Moreover, synchronous asynchronous depending strength successively revealed described. Lastly, experiment network validated field-programmable gate array (FPGA) circuit. Such an investigation positively contribute development progress medicine life science engineering.

Language: Английский

Citations

12
s-TBN: A new neural decoding model to identify stimulus categories from brain activity patterns DOI Creative Commons
Chunyu Liu, Bokai Cao, Jiacai Zhang

et al.

IEEE Transactions on Neural Systems and Rehabilitation Engineering, Journal Year: 2024, Volume and Issue: 32, P. 1934 - 1943

Published: Jan. 1, 2024

Neural decoding is still a challenging and hot topic in neurocomputing science. Recently, many studies have shown that brain network patterns containing rich spatiotemporal structural information represent the brain's activation under external stimuli. In traditional method, features are directly obtained using standard machine learning method provide to classifier, subsequently However, this cannot effectively extract multidimensional hidden network. Furthermore, on tensors show tensor decomposition model can fully mine unique characteristics of structure data with structure. This research proposed stimulus-constrained Tensor Brain Network (s-TBN) involves stimulus category-constraint information. The was verified real neuroimaging via magnetoencephalograph functional mangetic resonance imaging). Experimental results s-TBN achieve accuracy matrices greater than 11.06% 18.46% matrix compared other methods two modal datasets. These prove superiority extracting discriminative STN model, especially for object stimuli semantic

Language: Английский

Citations

0
An Adaptively Weighted Averaging Method for Regional Time Series Extraction of fMRI-Based Brain Decoding DOI
Jianfei Zhu, Baichun Wei,

Jiaru Tian

et al.

IEEE Journal of Biomedical and Health Informatics, Journal Year: 2024, Volume and Issue: 28(10), P. 5984 - 5995

Published: July 11, 2024

Brain decoding that classifies cognitive states using the functional fluctuations of brain can provide insightful information for understanding mechanisms functions. Among common procedures with magnetic resonance imaging (fMRI), extracting time series each region after parcellation traditionally averages across voxels within a region. This neglects spatial among and requirement downstream tasks. In this study, we propose to use fully connected neural network is jointly trained decoder perform an adaptively weighted average We extensive evaluations by state decoding, manifold learning, interpretability analysis on Human Connectome Project (HCP) dataset. The performance comparison presents accuracy increase up 5% stable improvement under different window sizes, resampling training data sizes. results learning show our method considerable separability basically excludes subject-specific information. shows identify reasonable regions corresponding state. Our study would aid basic pipeline fMRI processing.

Language: Английский

Citations

0
Electroencephalography-Based Motor Imagery Classification Using Multi-Scale Feature Fusion and Adaptive Lasso DOI Creative Commons
Shimiao Chen, Nan Li, Xiangzeng Kong

et al.

Big Data and Cognitive Computing, Journal Year: 2024, Volume and Issue: 8(12), P. 169 - 169

Published: Nov. 25, 2024

Brain–computer interfaces, where motor imagery electroencephalography (EEG) signals are transformed into control commands, offer a promising solution for enhancing the standard of living disabled individuals. However, performance EEG classification has been limited in most studies due to lack attention complementary information inherent at different temporal scales. Additionally, significant inter-subject variability sensitivity biological motion poses another critical challenge achieving accurate subject-dependent manner. To address these challenges, we propose novel machine learning framework combining multi-scale feature fusion, which captures global and local spatial from different-sized segmentations, adaptive Lasso-based selection, mechanism adaptively retaining informative features discarding irrelevant ones. Experimental results on multiple public benchmark datasets revealed substantial improvements classification, rates 81.36%, 75.90%, 68.30% BCIC-IV-2a, SMR-BCI, OpenBMI datasets, respectively. These not only surpassed existing methodologies but also underscored effectiveness our approach overcoming specific challenges classification. Ablation further confirmed efficacy both analysis selection mechanisms. This marks advancement decoding signals, positioning it practical applications real-world BCIs.

Language: Английский

Citations

0