EEGSym: Overcoming Inter-Subject Variability in Motor Imagery Based BCIs With Deep Learning

Multi-source deep domain adaptation ensemble framework for cross-dataset motor imagery EEG transfer learning

Objective. Electroencephalography (EEG) is an important kind of bioelectric signal for measuring physiological activities of the brain, and motor imagery (MI) EEG has significant clinical application prospects. Convolutional neural network has become a mainstream algorithm for MI EEG classification, however lack of subject-specific data considerably restricts its decoding accuracy and generalization performance. To address this challenge, a novel transfer learning (TL) framework using auxiliary dataset to improve the MI EEG classification performance of target subject is proposed in this paper.Approach. We developed a multi-source deep domain adaptation ensemble framework (MSDDAEF) for cross-dataset MI EEG decoding. The proposed MSDDAEF comprises three main components: model pre-training, deep domain adaptation, and multi-source ensemble. Moreover, for each component, different designs were examined to verify the robustness of MSDDAEF.Main results. Bidirectional validation experiments were performed on two large public MI EEG datasets (openBMI and GIST). The highest average classification accuracy of MSDDAEF reaches 74.28% when openBMI serves as target dataset and GIST serves as source dataset. While the highest average classification accuracy of MSDDAEF is 69.85% when GIST serves as target dataset and openBMI serves as source dataset. In addition, the classification performance of MSDDAEF surpasses several well-established studies and state-of-the-art algorithms.Significance. The results of this study show that cross-dataset TL is feasible for left/right-hand MI EEG decoding, and further indicate that MSDDAEF is a promising solution for addressing MI EEG cross-dataset variability.

Multi-Task Heterogeneous Ensemble Learning-Based Cross-Subject EEG Classification Under Stroke Patients

Robot-assisted motor training is applied for neurorehabilitation in stroke patients, using motor imagery (MI) as a representative paradigm of brain-computer interfaces to offer real-life assistance to individuals facing movement challenges. However, the effectiveness of training with MI may vary depending on the location of the stroke lesion, which should be considered. This paper introduces a multi-task electroencephalogram-based heterogeneous ensemble learning (MEEG-HEL) specifically designed for cross-subject training. In the proposed framework, common spatial patterns were used for feature extraction, and the features according to stroke lesions are shared and selected through sequential forward floating selection. The heterogeneous ensembles were used as classifiers. Nine patients with chronic ischemic stroke participated, engaging in MI and motor execution (ME) paradigms involving finger tapping. The classification criteria for the multi-task were established in two ways, taking into account the characteristics of stroke patients. In the cross-subject session, the first involved a direction recognition task for two-handed classification, achieving a performance of 0.7419 (±0.0811) in MI and 0.7061 (±0.1270) in ME. The second task focused on motor assessment for lesion location, resulting in a performance of 0.7457 (±0.1317) in MI and 0.6791 (±0.1253) in ME. Comparing the specific-subject session, except for ME on the motor assessment task, performance on both tasks was significantly higher than the cross-subject session. Furthermore, classification performance was similar to or statistically higher in cross-subject sessions compared to baseline models. The proposed MEEG-HEL holds promise in improving the practicality of neurorehabilitation in clinical settings and facilitating the detection of lesions.

EISATC-Fusion: Inception Self-Attention Temporal Convolutional Network Fusion for Motor Imagery EEG Decoding

The motor imagery brain-computer interface (MI-BCI) based on electroencephalography (EEG) is a widely used human-machine interface paradigm. However, due to the non-stationarity and individual differences among subjects in EEG signals, the decoding accuracy is limited, affecting the application of the MI-BCI. In this paper, we propose the EISATC-Fusion model for MI EEG decoding, consisting of inception block, multi-head self-attention (MSA), temporal convolutional network (TCN), and layer fusion. Specifically, we design a DS Inception block to extract multi-scale frequency band information. And design a new cnnCosMSA module based on CNN and cos attention to solve the attention collapse and improve the interpretability of the model. The TCN module is improved by the depthwise separable convolution to reduces the parameters of the model. The layer fusion consists of feature fusion and decision fusion, fully utilizing the features output by the model and enhances the robustness of the model. We improve the two-stage training strategy for model training. Early stopping is used to prevent model overfitting, and the accuracy and loss of the validation set are used as indicators for early stopping. The proposed model achieves within-subject classification accuracies of 84.57% and 87.58% on BCI Competition IV Datasets 2a and 2b, respectively. And the model achieves cross-subject classification accuracies of 67.42% and 71.23% (by transfer learning) when training the model with two sessions and one session of Dataset 2a, respectively. The interpretability of the model is demonstrated through weight visualization method.

Unraveling motor imagery brain patterns using explainable artificial intelligence based on Shapley values

BACKGROUND AND OBJECTIVE

Motor imagery (MI) based brain-computer interfaces (BCIs) are widely used in rehabilitation due to the close relationship that exists between MI and motor execution (ME). However, the underlying brain mechanisms of MI remain not well understood. Most MI-BCIs use the sensorimotor rhythms elicited in the primary motor cortex (M1) and somatosensory cortex (S1), which consist of an event-related desynchronization followed by an event-related synchronization. Consequently, this has resulted in systems that only record signals around M1 and S1. However, MI could involve a more complex network including sensory, association, and motor areas. In this study, we hypothesize that the superior accuracies achieved by new deep learning (DL) models applied to MI decoding rely on focusing on a broader MI activation of the brain. Parallel to the success of DL, the field of explainable artificial intelligence (XAI) has seen continuous development to provide explanations for DL networks success. The goal of this study is to use XAI in combination with DL to extract information about MI brain activation patterns from non-invasive electroencephalography (EEG) signals.

METHODS

We applied an adaptation of Shapley additive explanations (SHAP) to EEGSym, a state-of-the-art DL network with exceptional transfer learning capabilities for inter-subject MI classification. We obtained the SHAP values from two public databases comprising 171 users generating left and right hand MI instances with and without real-time feedback.

RESULTS

We found that EEGSym based most of its prediction on the signal of the frontal electrodes, i.e. F7 and F8, and on the first 1500 ms of the analyzed imagination period. We also found that MI involves a broad network not only based on M1 and S1, but also on the prefrontal cortex (PFC) and the posterior parietal cortex (PPC). We further applied this knowledge to select a 8-electrode configuration that reached inter-subject accuracies of 86.5% ± 10.6% on the Physionet dataset and 88.7% ± 7.0% on the Carnegie Mellon University's dataset.

CONCLUSION

Our results demonstrate the potential of combining DL and SHAP-based XAI to unravel the brain network involved in producing MI. Furthermore, SHAP values can optimize the requirements for out-of-laboratory BCI applications involving real users.

Subject-independent meta-learning framework towards optimal training of EEG-based classifiers

Advances in deep learning have shown great promise towards the application of performing high-accuracy Electroencephalography (EEG) signal classification in a variety of tasks. However, many EEG-based datasets are often plagued by the issue of high inter-subject signal variability. Robust deep learning models are notoriously difficult to train under such scenarios, often leading to subpar or widely varying performance across subjects under the leave-one-subject-out paradigm. Recently, the model agnostic meta-learning framework was introduced as a way to increase the model's ability to generalize towards new tasks. While the original framework focused on task-based meta-learning, this research aims to show that the meta-learning methodology can be modified towards subject-based signal classification while maintaining the same task objectives and achieve state-of-the-art performance. Namely, we propose the novel implementation of a few/zero-shot subject-independent meta-learning framework towards multi-class inner speech and binary class motor imagery classification. Compared to current subject-adaptive methods which utilize large number of labels from the target, the proposed framework shows its effectiveness in training zero-calibration and few-shot models for subject-independent EEG classification. The proposed few/zero-shot subject-independent meta-learning mechanism performs well on both small and large datasets and achieves robust, generalized performance across subjects. The results obtained shows a significant improvement over the current state-of-the-art, with the binary class motor imagery achieving 88.70% and the accuracy of multi-class inner speech achieving an average of 31.15%. Codes will be made available to public upon publication.

Cross-Subject Motor Imagery Decoding by Transfer Learning of Tactile ERD

For Brain-Computer Interface (BCI) based on motor imagery (MI), the MI task is abstract and spontaneous, presenting challenges in measurement and control and resulting in a lower signal-to-noise ratio. The quality of the collected MI data significantly impacts the cross-subject calibration results. To address this challenge, we introduce a novel cross-subject calibration method based on passive tactile afferent stimulation, in which data induced by tactile stimulation is utilized to calibrate transfer learning models for cross-subject decoding. During the experiments, tactile stimulation was applied to either the left or right hand, with subjects only required to sense tactile stimulation. Data from these tactile tasks were used to train or fine-tune models and subsequently applied to decode pure MI data. We evaluated BCI performance using both the classical Common Spatial Pattern (CSP) combined with the Linear Discriminant Analysis (LDA) algorithm and a state-of-the-art deep transfer learning model. The results demonstrate that the proposed calibration method achieved decoding performance at an equivalent level to traditional MI calibration, with the added benefit of outperforming traditional MI calibration with fewer trials. The simplicity and effectiveness of the proposed cross-subject tactile calibration method make it valuable for practical applications of BCI, especially in clinical settings.

A comprehensive dataset for home appliance control using ERP-based BCIs with the application of inter-subject transfer learning

Brain-computer interfaces (BCIs) have a potential to revolutionize human-computer interaction by enabling direct links between the brain and computer systems. Recent studies are increasingly focusing on practical applications of BCIs-e.g., home appliance control just by thoughts. One of the non-invasive BCIs using electroencephalography (EEG) capitalizes on event-related potentials (ERPs) in response to target stimuli and have shown promise in controlling home appliance. In this paper, we present a comprehensive dataset of online ERP-based BCIs for controlling various home appliances in diverse stimulus presentation environments. We collected online BCI data from a total of 84 subjects among whom 60 subjects controlled three types of appliances (TV: 30, door lock: 15, and electric light: 15) with 4 functions per appliance, 14 subjects controlled a Bluetooth speaker with 6 functions via an LCD monitor, and 10 subjects controlled air conditioner with 4 functions via augmented reality (AR). Using the dataset, we aimed to address the issue of inter-subject variability in ERPs by employing the transfer learning in two different approaches. The first approach, "within-paradigm transfer learning," aimed to generalize the model within the same paradigm of stimulus presentation. The second approach, "cross-paradigm transfer learning," involved extending the model from a 4-class LCD environment to different paradigms. The results demonstrated that transfer learning can effectively enhance the generalizability of BCIs based on ERP across different subjects and environments.

EEGProgress: A fast and lightweight progressive convolution architecture for EEG classification

Because of the intricate topological structure and connection of the human brain, extracting deep spatial features from electroencephalograph (EEG) signals is a challenging and time-consuming task. The extraction of topological spatial information plays a crucial role in EEG classification, and the architecture of the spatial convolution greatly affects the performance and complexity of convolutional neural network (CNN) based EEG classification models. In this study, a progressive convolution CNN architecture named EEGProgress is proposed, aiming to efficiently extract the topological spatial information of EEG signals from multi-scale levels (electrode, brain region, hemisphere, global) with superior speed. To achieve this, the raw EEG data is permuted using the empirical topological permutation rule, integrating the EEG data with numerous topological properties. Subsequently, the spatial features are extracted by a progressive feature extractor including prior, electrode, region, and hemisphere convolution blocks, progressively extracting the deep spatial features with reduced parameters and speed. Finally, the comparison and ablation experiments under both cross-subject and within-subject scenarios are conducted on a public dataset to verify the performance of the proposed EEGProgress and the effectiveness of the topological permutation. The results demonstrate the superior feature extraction ability of the proposed EEGProgress, with an average increase of 4.02% compared to other CNN-based EEG classification models under both cross-subject and within-subject scenarios. Furthermore, with the obtained average testing time, FLOPs, and parameters, the proposed EEGProgress outperforms other comparison models in terms of model complexity.

TSPNet: a time-spatial parallel network for classification of EEG-based multiclass upper limb motor imagery BCI

The classification of electroencephalogram (EEG) motor imagery signals has emerged as a prominent research focus within the realm of brain-computer interfaces. Nevertheless, the conventional, limited categories (typically just two or four) offered by brain-computer interfaces fail to provide an extensive array of control modes. To address this challenge, we propose the Time-Spatial Parallel Network (TSPNet) for recognizing six distinct categories of upper limb motor imagery. Within TSPNet, temporal and spatial features are extracted separately, with the time dimension feature extractor and spatial dimension feature extractor performing their respective functions. Following this, the Time-Spatial Parallel Feature Extractor is employed to decouple the connection between temporal and spatial features, thus diminishing feature redundancy. The Time-Spatial Parallel Feature Extractor deploys a gating mechanism to optimize weight distribution and parallelize time-spatial features. Additionally, we introduce a feature visualization algorithm based on signal occlusion frequency to facilitate a qualitative analysis of TSPNet. In a six-category scenario, TSPNet achieved an accuracy of 49.1% ± 0.043 on our dataset and 49.7% ± 0.029 on a public dataset. Experimental results conclusively establish that TSPNet outperforms other deep learning methods in classifying data from these two datasets. Moreover, visualization results vividly illustrate that our proposed framework can generate distinctive classifier patterns for multiple categories of upper limb motor imagery, discerned through signals of varying frequencies. These findings underscore that, in comparison to other deep learning methods, TSPNet excels in intention recognition, which bears immense significance for non-invasive brain-computer interfaces.

Deep Unsupervised Representation Learning for Feature-Informed EEG Domain Extraction

In electroencephalography (EEG) classification paradigms, data from a target subject is often difficult to obtain, leading to difficulties in training a robust deep learning network. Transfer learning and their variations are effective tools in improving such models suffering from lack of data. However, many of the proposed variations and deep models often rely on a single assumed distribution to represent the latent features which may not scale well due to inter- and intra-subject variations in signals. This leads to significant instability in individual subject decoding performances. The presence of non-trivial domain differences between different sets of training or transfer learning data causes poorer model generalization towards the target subject. However, the detection of these domain differences is often difficult to perform due to the ill-defined nature of the EEG domain features. This study proposes a novel inference model, the Joint Embedding Variational Autoencoder, that offers conditionally tighter approximation of the estimated spatiotemporal feature distribution through the use of jointly optimised variational autoencoders to achieve optimizable data dependent inputs as an additional variable for improved overall model optimisation and scaling without sacrificing model tightness. To learn the variational bound, we show that maximising the marginal log-likelihood of only the second embedding section is required to achieve conditionally tighter lower bounds. Furthermore, we show that this model provides state-of-the-art EEG data reconstruction and deep feature extraction. The extracted domains of the EEG signals across each subject displays the rationale as to why there exists disparity between subjects' adaptation efficacy.

Sparse Bayesian Learning for End-to-End EEG Decoding

Decoding brain activity from non-invasive electroencephalography (EEG) is crucial for brain-computer interfaces (BCIs) and the study of brain disorders. Notably, end-to-end EEG decoding has gained widespread popularity in recent years owing to the remarkable advances in deep learning research. However, many EEG studies suffer from limited sample sizes, making it difficult for existing deep learning models to effectively generalize to highly noisy EEG data. To address this fundamental limitation, this paper proposes a novel end-to-end EEG decoding algorithm that utilizes a low-rank weight matrix to encode both spatio-temporal filters and the classifier, all optimized under a principled sparse Bayesian learning (SBL) framework. Importantly, this SBL framework also enables us to learn hyperparameters that optimally penalize the model in a Bayesian fashion. The proposed decoding algorithm is systematically benchmarked on five motor imagery BCI EEG datasets ( N=192) and an emotion recognition EEG dataset ( N=45), in comparison with several contemporary algorithms, including end-to-end deep-learning-based EEG decoding algorithms. The classification results demonstrate that our algorithm significantly outperforms the competing algorithms while yielding neurophysiologically meaningful spatio-temporal patterns. Our algorithm therefore advances the state-of-the-art by providing a novel EEG-tailored machine learning tool for decoding brain activity.

TDLNet: Transfer Data Learning Network for Cross-Subject Classification Based on Multiclass Upper Limb Motor Imagery EEG

The limited number of brain-computer interface based on motor imagery (MI-BCI) instruction sets for different movements of single limbs makes it difficult to meet practical application requirements. Therefore, designing a single-limb, multi-category motor imagery (MI) paradigm and effectively decoding it is one of the important research directions in the future development of MI-BCI. Furthermore, one of the major challenges in MI-BCI is the difficulty of classifying brain activity across different individuals. In this article, the transfer data learning network (TDLNet) is proposed to achieve the cross-subject intention recognition for multiclass upper limb motor imagery. In TDLNet, the Transfer Data Module (TDM) is used to process cross-subject electroencephalogram (EEG) signals in groups and then fuse cross-subject channel features through two one-dimensional convolutions. The Residual Attention Mechanism Module (RAMM) assigns weights to each EEG signal channel and dynamically focuses on the EEG signal channels most relevant to a specific task. Additionally, a feature visualization algorithm based on occlusion signal frequency is proposed to qualitatively analyze the proposed TDLNet. The experimental results show that TDLNet achieves the best classification results on two datasets compared to CNN-based reference methods and transfer learning method. In the 6-class scenario, TDLNet obtained an accuracy of 65%±0.05 on the UML6 dataset and 63%±0.06 on the GRAZ dataset. The visualization results demonstrate that the proposed framework can produce distinct classifier patterns for multiple categories of upper limb motor imagery through signals of different frequencies. The ULM6 dataset is available at https://dx.doi.org/10.21227/8qw6-f578.

Spatial Variation Generation Algorithm for Motor Imagery Data Augmentation: Increasing the Density of Sample Vicinity

The imbalanced development between deep learning-based model design and motor imagery (MI) data acquisition raises concerns about the potential overfitting issue-models can identify training data well but fail to generalize test data. In this study, a Spatial Variation Generation (SVG) algorithm for MI data augmentation is proposed to alleviate the overfitting issue. In essence, SVG generates MI data using variations of electrode placement and brain spatial pattern, ultimately elevating the density of the raw sample vicinity. The proposed SVG prevents models from memorizing the training data by replacing the raw samples with the proper vicinal distribution. Moreover, SVG generates a uniform distribution and stabilizes the training process of models. In comparison studies involving five deep learning-based models across eight datasets, the proposed SVG algorithm exhibited a notable improvement of 0.021 in the area under the receiver operating characteristic curve (AUC). The improvement achieved by SVG outperforms other data augmentation algorithms. Further results from the ablation study verify the effectiveness of each component of SVG. Finally, the studies in the control group with varying numbers of samples show that the SVG algorithm consistently improves the AUC, with improvements ranging from approximately 0.02 to 0.15.

Motor Imagery Multi-Tasks Classification for BCIs Using the NVIDIA Jetson TX2 Board and the EEGNet Network

Nowadays, Brain-Computer Interfaces (BCIs) still captivate large interest because of multiple advantages offered in numerous domains, explicitly assisting people with motor disabilities in communicating with the surrounding environment. However, challenges of portability, instantaneous processing time, and accurate data processing remain for numerous BCI system setups. This work implements an embedded multi-tasks classifier based on motor imagery using the EEGNet network integrated into the NVIDIA Jetson TX2 card. Therefore, two strategies are developed to select the most discriminant channels. The former uses the accuracy based-classifier criterion, while the latter evaluates electrode mutual information to form discriminant channel subsets. Next, the EEGNet network is implemented to classify discriminant channel signals. Additionally, a cyclic learning algorithm is implemented at the software level to accelerate the model learning convergence and fully profit from the NJT2 hardware resources. Finally, motor imagery Electroencephalogram (EEG) signals provided by HaLT's public benchmark were used, in addition to the k-fold cross-validation method. Average accuracies of 83.7% and 81.3% were achieved by classifying EEG signals per subject and motor imagery task, respectively. Each task was processed with an average latency of 48.7 ms. This framework offers an alternative for online EEG-BCI systems' requirements, dealing with short processing times and reliable classification accuracy.

MEDUSA©: A novel Python-based software ecosystem to accelerate brain-computer interface and cognitive neuroscience research

BACKGROUND AND OBJECTIVE

Neurotechnologies have great potential to transform our society in ways that are yet to be uncovered. The rate of development in this field has increased significantly in recent years, but there are still barriers that need to be overcome before bringing neurotechnologies to the general public. One of these barriers is the difficulty of performing experiments that require complex software, such as brain-computer interfaces (BCI) or cognitive neuroscience experiments. Current platforms have limitations in terms of functionality and flexibility to meet the needs of researchers, who often need to implement new experimentation settings. This work was aimed to propose a novel software ecosystem, called MEDUSA©, to overcome these limitations.

METHODS

We followed strict development practices to optimize MEDUSA© for research in BCI and cognitive neuroscience, making special emphasis in the modularity, flexibility and scalability of our solution. Moreover, it was implemented in Python, an open-source programming language that reduces the development cost by taking advantage from its high-level syntax and large number of community packages.

RESULTS

MEDUSA© provides a complete suite of signal processing functions, including several deep learning architectures or connectivity analysis, and ready-to-use BCI and neuroscience experiments, making it one of the most complete solutions nowadays. We also put special effort in providing tools to facilitate the development of custom experiments, which can be easily shared with the community through an app market available in our website to promote reproducibility.

CONCLUSIONS

MEDUSA© is a novel software ecosystem for modern BCI and neurotechnology experimentation that provides state-of-the-art tools and encourages the participation of the community to make a difference for the progress of these fields. Visit the official website at https://www.medusabci.com/ to know more about this project.