Spatial interpretability of time-frequency relevance optimized in motor imagery discrimination using Deep&amp;Wide networks

Collazos-Huertas, Diego Fabian; Álvarez-Meza, Andrés Marino; Castellanos-Domínguez, Germán

doi:10.1016/j.bspc.2021.102626

Cited by 6 publications

(12 citation statements)

References 35 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Furthermore, to reduce the low SNR phenomenon in subjects performing poor motor imagery tasks, more elaborate feature extraction methods are to be investigated. One more issue of consideration is to explore the effectiveness of the developed approach for artifact removal in modern architectures of deep learning [ 15 ]. Moreover, eye-tracking and cognitive psychological attention test data could be of benefit to enhance our artifact removal approach and favor both BCI performance and interpretability [ 66 , 67 ].…”

Section: Discussionmentioning

confidence: 99%

“…However, capturing neural activity from the scalp faces several restrictions: non-stationarity and nonlinearity of EEG data [ 11 ], a low-spatial resolution that may also affect the time-resolution [ 12 ], and the inter- and intra-subject variability for which the distribution of features extracted from time-variant brain patterns across subjects can be different for the same tasks [ 13 ]. All those factors result in low Signal-to-Noise Ratio (SNR) phenomena, posing a challenge in EEG analysis [ 14 , 15 , 16 ]. SNR values, in general, are affected by several factors [ 17 ].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Subject-Dependent Artifact Removal for Enhancing Motor Imagery Classifier Performance under Poor Skills

Tobón-Henao

Álvarez-Meza

Castellanos-Domínguez

2022

Sensors

View full text Add to dashboard Cite

The Electroencephalography (EEG)-based motor imagery (MI) paradigm is one of the most studied technologies for Brain-Computer Interface (BCI) development. Still, the low Signal-to-Noise Ratio (SNR) poses a challenge when constructing EEG-based BCI systems. Moreover, the non-stationary and nonlinear signal issues, the low-spatial data resolution, and the inter- and intra-subject variability hamper the extraction of discriminant features. Indeed, subjects with poor motor skills have difficulties in practicing MI tasks against low SNR scenarios. Here, we propose a subject-dependent preprocessing approach that includes the well-known Surface Laplacian Filtering and Independent Component Analysis algorithms to remove signal artifacts based on the MI performance. In addition, power- and phase-based functional connectivity measures are studied to extract relevant and interpretable patterns and identify subjects of inefficency. As a result, our proposal, Subject-dependent Artifact Removal (SD-AR), improves the MI classification performance in subjects with poor motor skills. Consequently, electrooculography and volume-conduction EEG artifacts are mitigated within a functional connectivity feature-extraction strategy, which favors the classification performance of a straightforward linear classifier.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Subject-Dependent Artifact Removal for Enhancing Motor Imagery Classifier Performance under Poor Skills

Tobón-Henao

Álvarez-Meza

Castellanos-Domínguez

2022

Sensors

View full text Add to dashboard Cite

show abstract

“…In terms of average accuracy, ESSDM-FTFG-CapsNet achieves the highest accuracy of 85.8% and kappa of 0.84 with fewer layers, which is better than the baseline models. Especially, the average accuracy is 12.1% higher than the ORDW [5] with optimal subject-specific network parameters and slightly deteriorated with the MSFT [1] model. ESSDM-FTFG-CapsNet does not change any hyper-parameters when testing all subjects.…”

Section: A Performance Comparison With Specific-subject Benchmarksmentioning

confidence: 91%

“…Table III demonstrates the classification performance (AC-Cavg, kappa, and p-values) following 10-fold cross-validation for proposed ESSDM and SWD with the state-of-the-art models on the BCI IV-2a dataset (four classes). The following baseline models were utilized for performance comparison: sparse filter band common spatial pattern (SFBCSP) [12], compact convolutional recurrent neural network (SCCRNN) [37], ORDW [5], metric-based spatial filtering transformer (MSFT) [1], convolutional neural network-long short-term memory (CNN-LSTM) [37], Global [38], and MS-AMF [3]. For a more reasonable method analysis, these four types of imagination classes are first paired for bi-classification.…”

Section: A Performance Comparison With Specific-subject Benchmarksmentioning

confidence: 99%

“…In the second category, shallow learning and deep learning methods have been employed to estimate the MI-EEG tasks with promising results based on EEG [2], [4]- [6], [10]. These approaches involve, spatial timefrequency features with optimized Deep&Wide (ORDW) [5], 3D-convolutional neural network (3D-CNN) [4], frequential deep belief network (FDBN) [10], spatial filters based deep convolutional neural networks [6], and multi-scale fusion attention mechanism (MS-AMF) [3], demonstrating the high discriminability of the feature for enhancement of MI-based actions. Among these approaches, the decomposition methods became crucial for MI recognition and gaining attention.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

ESSDM: An Enhanced Sparse Swarm Decomposition Method and Its Application in Multi‐class Motor Imagery–Based EEG-BCI System

Bhalerao,

Pachori

2024

Preprint

View full text Add to dashboard Cite

Electroencephalography (EEG)-based motor imagery (MI) decoding has established a novel experimental paradigm in brain-computer interface (BCI) applications that offer effective treatment for stroke paralyzed patients. However, existing MI-EEG-based BCI systems introduce deployment issues because of nonstationary EEG signals, suboptimal features, and limited multi-class scalability. To tackle these issues, we propose an enhanced sparse swarm decomposition method (ESSDM) based on selfish-herd optimization and sparse spectrum to solve the issue of choice of uniform decomposition and hyper-parameters in swarm decomposition and applied to enhance MI-EEG classification. ESSDM adopts improved swarm filtering to automatically deliver optimal frequency bands in sparse spectrums with optimized hyper-parameters to extract dominant oscillatory components (OCs) that significantly enhance MI activation-related sub-bands. In addition, new fitness criteria is designed based on the Kullback–Leibler divergence distance from spectral kurtosis of obtained modes to select hyper-parameters that optimize decomposition effect, avoid excessive iterations, and provide fast convergence with optimal modes. Further, fused time-frequency graph (FTFG) features were derived from computed time-frequency representation to find cross-channel mutual spectral information. The experimental results on the 2-class BCI III-4a and 4-class BCI IV-2a datasets reveal that the proposed FTFG feature with CapsNet classifier framework (ESSDM-FTFG-CapsNet) outperformed existing methods in specific-subject and cross-subject scenarios.

show abstract

GMMPLS: A Novel Automatic State-Based Algorithm for Continuous Decoding in BMIs

Foodeh¹,

Shalchyan²,

Daliri³

2021

IEEE Access

View full text Add to dashboard Cite

In this paper, a novel fully-automated state-based decoding method has been proposed for continuous decoding problems in brain-machine interface (BMI) systems, called Gaussian mixture of model (GMM)-assisted PLS (GMMPLS). In contrast to other state-based and hierarchical decoders, the proposed method does not demand any prior information about the desired output structure. Instead, GMMPLS uses the GMM algorithm to divide the desired output into a specific number of states (clusters). Next, a logistic regression model is trained to predict the probability membership of each time sample for each state. Finally, using the concept of the partial least square (PLS) algorithm, GMMPLS constructs a model for decoding the desired output using the input data and the achieved membership probabilities. The performance of the GMMPLS has been evaluated and compared to PLS, the nonlinear quadratic PLS (QPLS), and the bayesian PLS (BPLS) methods through a simulated dataset and two different real-world BMI datasets. The achieved results demonstrated that the GMMPLS significantly outperformed PLS, QPLS, and BPLS overall datasets. INDEX TERMSBrain-machine interfaces (BMI), Partial least square (PLS), State-based decoding, Continuous decoding, Gaussian mixture of model (GMM).

show abstract

Spatial interpretability of time-frequency relevance optimized in motor imagery discrimination using Deep&Wide networks

Cited by 6 publications

References 35 publications

Subject-Dependent Artifact Removal for Enhancing Motor Imagery Classifier Performance under Poor Skills

Subject-Dependent Artifact Removal for Enhancing Motor Imagery Classifier Performance under Poor Skills

ESSDM: An Enhanced Sparse Swarm Decomposition Method and Its Application in Multi‐class Motor Imagery–Based EEG-BCI System

GMMPLS: A Novel Automatic State-Based Algorithm for Continuous Decoding in BMIs

Contact Info

Product

Resources

About