Brain-computer Interface Paradigm Research Articles

Benchmarking brain-computer interface algorithms: Riemannian approaches vs convolutional neural networks.

To date, a comprehensive comparison of Riemannian decoding methods with deep convolutional neural networks for EEG-based Brain-Computer Interfaces remains absent from published work. We address this research gap by using MOABB, The Mother Of All BCI Benchmarks, to compare novel convolutional neural networks to state-of-the-art Riemannian approaches across a broad range of EEG datasets, including motor imagery, P300, and steady-state visual evoked potentials paradigms. 
Approach. We systematically evaluated the performance of convolutional neural networks, specifically EEGNet, shallow ConvNet, and deep ConvNet, against well-established Riemannian decoding methods using MOABB processing pipelines. This evaluation included within- session, cross-session, and cross-subject methods, to provide a practical analysis of model effectiveness and to find an overall solution that performs well across different experimental settings. 
Main results. We find no significant differences in decoding performance between convolutional neural networks and Riemannian methods for within-session, cross-session, and cross-subject analyses. 
Significance. The results show that, when using traditional Brain-Computer Interface paradigms, the choice between CNNs and Riemannian methods may not heavily impact decoding performances in many experimental settings. These findings provide researchers with flexibility in choosing decoding approaches based on factors such as ease of implementation, computational efficiency or individual preferences.

Yes or no? A study of ErrPs in the “guess what I am thinking” paradigm with stimuli of different visual content

Error-related potentials (ErrPs) have attracted attention in part because of their practical potential for building brain-computer interface (BCI) paradigms. BCIs, facilitating direct communication between the brain and machines, hold great promise for brain-AI interaction. Therefore, a comprehensive understanding of ErrPs is crucial to ensure reliable BCI outcomes. In this study, we investigated ErrPs in the context of the “guess what I am thinking” paradigm. 23 healthy participants were instructed to imagine an object from a predetermined set, while an algorithm randomly selected another object that was either the same as or different from the imagined object. We recorded and analyzed the participants’ EEG activity to capture their mental responses to the algorithm’s “predictions”. The study identified components distinguishing correct from incorrect responses. It discusses their nature and how they differ from ErrPs extensively studied in other BCI paradigms. We observed pronounced variations in the shape of ErrPs across different stimulus sets, underscoring the significant influence of visual stimulus appearance on ErrP peaks. These findings have implications for designing effective BCI systems, especially considering the less conventional BCI paradigm employed. They emphasize the necessity of accounting for stimulus factors in BCI development.

Correcting for ERP latency jitter improves gaze-independent BCI decoding

Objective. Patients suffering from heavy paralysis or Locked-in-Syndrome can regain communication using a Brain–Computer Interface (BCI). Visual event-related potential (ERP) based BCI paradigms exploit visuospatial attention (VSA) to targets laid out on a screen. However, performance drops if the user does not direct their eye gaze at the intended target, harming the utility of this class of BCIs for patients suffering from eye motor deficits. We aim to create an ERP decoder that is less dependent on eye gaze. Approach. ERP component latency jitter plays a role in covert visuospatial attention (VSA) decoding. We introduce a novel decoder which compensates for these latency effects, termed Woody Classifier-based Latency Estimation (WCBLE). We carried out a BCI experiment recording ERP data in overt and covert visuospatial attention (VSA), and introduce a novel special case of covert VSA termed split VSA, simulating the experience of patients with severely impaired eye motor control. We evaluate WCBLE on this dataset and the BNCI2014-009 dataset, within and across VSA conditions to study the dependency on eye gaze and the variation thereof during the experiment. Main results. WCBLE outperforms state-of-the-art methods in the VSA conditions of interest in gaze-independent decoding, without reducing overt VSA performance. Results from across-condition evaluation show that WCBLE is more robust to varying VSA conditions throughout a BCI operation session. Significance. Together, these results point towards a pathway to achieving gaze independence through suited ERP decoding. Our proposed gaze-independent solution enhances decoding performance in those cases where performing overt VSA is not possible.

Analysis and recognition of a novel experimental paradigm for musical emotion brain-computer interface

Musical emotions have received increasing attention over the years. To better recognize the emotions by brain-computer interface (BCI), the random music-playing and sequential music-playing experimental paradigms are proposed and compared in this paper. Two experimental paradigms consist of three positive pieces, three neutral pieces and three negative pieces of music. Ten subjects participate in two experimental paradigms. The features of electroencephalography (EEG) signals are firstly analyzed in the time, frequency and spatial domains. To improve the effect of emotion recognition, a recognition model is proposed with the optimal channels selecting by Pearson’s correlation coefficient, and the feature fusion combining differential entropy and wavelet packet energy. According to the analysis results, the features of sequential music-playing experimental paradigm are more different among three emotions. The classification results of sequential music-playing experimental paradigm are also better, and its average results of positive, neutral and negative emotions are 78.53%, 72.81% and 77.35%, respectively. The more obvious the changes of EEG induced by the emotions, the higher the classification accuracy will be. After analyzing two experimental paradigms, a better way for music to induce the emotions can be explored. Therefore, our research offers a novel perspective on affective BCIs.

Multi-source transfer learning via optimal transport feature ranking for EEG classification

Motor imagery (MI) brain-computer interface (BCI) paradigms have been extensively used in neurological rehabilitation. However, due to the required long calibration time and non-stationary nature of electroencephalogram (EEG) signals, it is challenging to obtain a substantial sample size of EEG data. Transfer learning (TL), which leverages knowledge from existing subjects to enhance the learning performance in new subjects, has been employed to solve this problem. Previous studies often extract sample features directly from the two domains for transfer learning, leading to poor features and negative transfer. To overcome this limitation, an optimal transport feature selection method was developed in this study to select the most suitable features for migration to enhance the generalization capability of the TL. This was achieved by analyzing the diagonal value similarity of the optimal transport coupling matrix. First, the Riemannian tangent space mapping method was used to map the sample covariance matrix of EEG trials from the Riemannian manifold to tangent space. Secondly, the optimal subset of EEG features was selected via the optimal transport feature selection method. Subsequently, the separate distribution alignment method was employed to minimize dissimilarities between the source and target domains. Finally, the weighted voting mechanisms were integrated for decision fusion. The proposed multi-source transfer feature learning (MSTFL) method was validated based on two public BCI Competition IV datasets. Our results demonstrated superior classification accuracy of 85.93% and 79.48%, respectively, on the two public datasets, compared to state-of-the-art models. These findings indicate the effectiveness of our proposed MSTFL model for both feature extraction and classification of MI signals.

Feature fusion improves brain-interface paradigm based on steady state visual evoked potential blocking response

The steady-state visual evoked potential blocking response (SSVEP-BR) is produced on an electroencephalogram (EEG) when the SSVEP is interrupted or abolished and allows augmentation of brain-computer interface (BCI) paradigms. Integration of the SSVEP-BR with the SSVEP enables an increase in the number of commands without the addition of further stimuli but refinement would improve performance. The current study evaluated SSVEP-BR and a novel method to combine multiple features and enhance the performance of frequency recognition and SSVEP-BR identification is proposed. Correlation features were extracted by filter bank canonical correlation analysis (FBCCA) and synchronization features by multivariate synchronization index (MSI) before being integrated. The performance of correlation features extracted by task-related component analysis (TRCA) were also evaluated. The novel integrated method was compared with FBCCA, MSI and TRCA for performance in frequency recognition and SSVEP-BR identification. The novel integrated method achieved higher classification accuracy than FBCCA and MSI for benchmark datasets when the sliding window was used for EEG data. However, the accuracy of TRCA was not stable when the sliding window was used. The novel integrated method produced an improvement in SSVEP-BR identification over FBCCA and MSI in the blocking dataset. TRCA was not found to be effective for SSVEP-BR identification. A novel integrated method is proposed which gives higher classification accuracy and more stable performance than FBCCA, MSI and TRCA for frequency recognition and SSVEP-BR identification when the EEG data sliding window was used and shows superior performance for the BCI paradigm based on SSVEP and SSVEP-BR.

Channel reflection: Knowledge-driven data augmentation for EEG-based brain–computer interfaces

A brain–computer interface (BCI) enables direct communication between the human brain and external devices. Electroencephalography (EEG) based BCIs are currently the most popular for able-bodied users. To increase user-friendliness, usually a small amount of user-specific EEG data are used for calibration, which may not be enough to develop a pure data-driven decoding model. To cope with this typical calibration data shortage challenge in EEG-based BCIs, this paper proposes a parameter-free channel reflection (CR) data augmentation approach that incorporates prior knowledge on the channel distributions of different BCI paradigms in data augmentation. Experiments on eight public EEG datasets across four different BCI paradigms (motor imagery, steady-state visual evoked potential, P300, and seizure classifications) using different decoding algorithms demonstrated that: (1) CR is effective, i.e., it can noticeably improve the classification accuracy; (2) CR is robust, i.e., it consistently outperforms existing data augmentation approaches in the literature; and, (3) CR is flexible, i.e., it can be combined with other data augmentation approaches to further improve the performance. We suggest that data augmentation approaches like CR should be an essential step in EEG-based BCIs. Our code is available online.

A novel deep transfer learning framework integrating general and domain-specific features for EEG-based brain–computer interface

The statistical variability of electroencephalographic (EEG) signals across individuals poses a common challenge for brain–computer interfaces (BCI). Specifically, the reuse of pre-recorded data from previous subjects presents a significant obstacle in decoding new subjects. To address these challenges, this paper introduces EEGTransferNet, an end-to-end modular transfer learning framework that enhances the similarity of statistical distributions across subjects. It utilizes convolutional neural networks as backbones for feature extraction and incorporates statistical distribution alignment and domain adaptation in different network layers to learn general features and domain-specific features respectively. The robust generalization design of EEGTransferNet enables it to handle multiple BCI paradigms using a single network making it efficient and effective. Its modular structure facilitates customization of backbone and transfer modules for diverse tasks. The optimal structure of EEGTransferNet was determined through analysis and extensive experimentation. We validated it on motor imagery (MI), error-related negativity (ERN), and rapid serial visual presentation (RSVP) datasets, which exhibits competitive performance compared to state-of-the-art methods. The experimental results confirm the effectiveness and superiority of EEGTransferNet.

Neural Correlates of Motor/Tactile Imagery and Tactile Sensation in a BCI paradigm: A High-Density EEG Source Imaging Study.

Complementary to brain-computer interface (BCI) based on motor imagery (MI) task, sensory imagery (SI) task provides a way for BCI construction using brain activity from somatosensory cortex. The underlying neurophysiological correlation between SI and MI was unclear and difficult to measure through behavior recording. In this study, we investigated the underlying neurodynamic of motor/tactile imagery and tactile sensation tasks through a high-density electroencephalogram (EEG) recording, and EEG source imaging was used to systematically explore the cortical activation differences and correlations between the tasks. In the experiment, participants were instructed to perform the left and right hand tasks in MI paradigm, sensory stimulation (SS) paradigm and SI paradigm. The statistical results demonstrated that the imagined MI and SI tasks differed from each other within ipsilateral sensorimotor scouts, frontal and right temporal areas in α bands, whereas real SS and imagined SI showed a similar activation pattern. The similarity between SS and SI may provide a way to train the BCI system, while the difference between MI and SI may provide a way to integrate the discriminative information between them to enhance BCI performance. The combination of the tasks and its underlying neurodynamic would provide a new approach for BCI designation for a wider application. BCI studies concentrate on the hybrid decoding method combining MI or SI with SS, but the underlining neurophysiological correlates between them were unclear. MI and SI differed from each other within the ipsilateral sensorimotor cortex in alpha bands. This is a first study to investigate the neurophysiological relationship between MI and SI through an EEG source imaging approach from high-density EEG recording.

Adversarial filtering based evasion and backdoor attacks to EEG-based brain-computer interfaces

A brain-computer interface (BCI) enables direct communication between the brain and an external device. Electroencephalogram (EEG) is a common input signal for BCIs, due to its convenience and low cost. Most research on EEG-based BCIs focuses on the accurate decoding of EEG signals, while ignoring their security. Recent studies have shown that machine learning models in BCIs are vulnerable to adversarial attacks. This paper proposes adversarial filtering based evasion and backdoor attacks to EEG-based BCIs, which are very easy to implement. Experiments on three datasets from different BCI paradigms demonstrated the effectiveness of our proposed attack approaches. To our knowledge, this is the first study on adversarial filtering for EEG-based BCIs, raising a new security concern and calling for more attention on the security of BCIs.

Characterization and classification of EEG signals evoked by different CAD models

AbstractThe past two decades have witnessed dramatic advancement in computer‐aided design (CAD). However, development of human–computer interfaces (HCI) for CAD have not kept up with these advances. Windows, Icons, Menus, Pointer (WIMP) is still the mainly used interface for CAD applications which limits the naturalness and intuitiveness of the CAD modeling process. As a novel interface, Brain–computer interfaces (BCIs) have great potential in the application of CAD modeling. Utilizing BCIs, the user can create CAD models just by thinking about it in principle, because BCIs provide an end‐to‐end interaction channel between users and CAD models. However, current related studies are mainly limited to the existing BCIs paradigms, while ignoring the relationship between electroencephalogram (EEG) signals and CAD models, which largely increases the cognitive load on the users. In this study, we aimed to explore the potential of using BCI to create CAD models directly independent of the classical BCIs paradigms. For this purpose, EEG signals evoked by six basic CAD models (i.e., point, square, trapezoid, line, triangle, and circle) were collected from 28 participants. After preprocessing and sub‐trial principal components analysis (st‐PCA) of recorded data, the peak, mean and time‐frequency energy features were extracted from EEG signals. By applying the one‐way repeated measures analysis of variance, we demonstrated that there were significant differences among these EEG features evoked by different CAD models. These features from EEG electrode channels ranked by mutual information were then used to train a discriminant classifier of genetic algorithm‐based support vector machine. The empirical result showed that this classifier can discriminate the CAD models with an average accuracy of about 72%, which turns out that EEG based model generation is feasible, and provides the technical and theoretical basis for building a novel BCI for CAD modeling.

Noise-Factorized Disentangled Representation Learning for Generalizable Motor Imagery EEG Classification.

Motor Imagery (MI) Electroencephalography (EEG) is one of the most common Brain-Computer Interface (BCI) paradigms that has been widely used in neural rehabilitation and gaming. Although considerable research efforts have been dedicated to developing MI EEG classification algorithms, they are mostly limited in handling scenarios where the training and testing data are not from the same subject or session. Such poor generalization capability significantly limits the realization of BCI in real-world applications. In this paper, we proposed a novel framework to disentangle the representation of raw EEG data into three components, subject/session-specific, MI-task-specific, and random noises, so that the subject/session-specific feature extends the generalization capability of the system. This is realized by a joint discriminative and generative framework, supported by a series of fundamental training losses and training strategies. We evaluated our framework on three public MI EEG datasets, and detailed experimental results show that our method can achieve superior performance by a large margin compared to current state-of-the-art benchmark algorithms.

Brain-computer interface paradigms and neural coding.

Brain signal patterns generated in the central nervous system of brain-computer interface (BCI) users are closely related to BCI paradigms and neural coding. In BCI systems, BCI paradigms and neural coding are critical elements for BCI research. However, so far there have been few references that clearly and systematically elaborated on the definition and design principles of the BCI paradigm as well as the definition and modeling principles of BCI neural coding. Therefore, these contents are expounded and the existing main BCI paradigms and neural coding are introduced in the review. Finally, the challenges and future research directions of BCI paradigm and neural coding were discussed, including user-centered design and evaluation for BCI paradigms and neural coding, revolutionizing the traditional BCI paradigms, breaking through the existing techniques for collecting brain signals and combining BCI technology with advanced AI technology to improve brain signal decoding performance. It is expected that the review will inspire innovative research and development of the BCI paradigm and neural coding.

Aggregating intrinsic information to enhance BCI performance through federated learning

Insufficient data is a long-standing challenge for Brain–Computer Interface (BCI) to build a high-performance deep learning model. Though numerous research groups and institutes collect a multitude of EEG datasets for the same BCI task, sharing EEG data from multiple sites is still challenging due to the heterogeneity of devices. The significance of this challenge cannot be overstated, given the critical role of data diversity in fostering model robustness. However, existing works rarely discuss this issue, predominantly centering their attention on model training within a single dataset, often in the context of inter-subject or inter-session settings. In this work, we propose a hierarchical personalized Federated Learning EEG decoding (FLEEG) framework to surmount this challenge. This innovative framework heralds a new learning paradigm for BCI, enabling datasets with disparate data formats to collaborate in the model training process. Each client is assigned a specific dataset and trains a hierarchical personalized model to manage diverse data formats and facilitate information exchange. Meanwhile, the server coordinates the training procedure to harness knowledge gleaned from all datasets, thus elevating overall performance. The framework has been evaluated in Motor Imagery (MI) classification with nine EEG datasets collected by different devices but implementing the same MI task. Results demonstrate that the proposed framework can boost classification performance up to 8.4% by enabling knowledge sharing between multiple datasets, especially for smaller datasets. Visualization results also indicate that the proposed framework can empower the local models to put a stable focus on task-related areas, yielding better performance. To the best of our knowledge, this is the first end-to-end solution to address this important challenge.

Integrating Simultaneous Motor Imagery and Spatial Attention for EEG-BCI Control.

EEG-based brain-computer interfaces (BCI) are non-invasive approaches for replacing or restoring motor functions in impaired patients, and direct brain-to-device communication in the general population. Motor imagery (MI) is one of the most used BCI paradigms, but its performance varies across individuals and certain users require substantial training to develop control. In this study, we propose to integrate a MI paradigm simultaneously with a recently proposed Overt Spatial Attention (OSA) paradigm, to accomplish BCI control. We evaluated a cohort of 25 human subjects' ability to control a virtual cursor in one- and two-dimensions over 5 BCI sessions. The subjects used 5 different BCI paradigms: MI alone, OSA alone, MI, and OSA simultaneously towards the same target (MI+OSA), and MI for one axis while OSA controls the other (MI/OSA and OSA/MI). Our results show that MI+OSA reached the highest average online performance in 2D tasks at 49% Percent Valid Correct (PVC), and statistically outperforms both MI alone (42%) and OSA alone (45%). MI+OSA had a similar performance to each subject's best individual method between MI alone and OSA alone (50%) and 9 subjects reached their highest average BCI performance using MI+OSA. Integrating MI and OSA leads to improved performance over both individual methods at the group level and is the best BCI paradigm option for some subjects. This work proposes a new BCI control paradigm that integrates two existing paradigms and demonstrates its value by showing that it can improve users' BCI performance.

A dynamic window method based on reinforcement learning for SSVEP recognition.

Steady-state visual evoked potential (SSVEP) is one of the most used brain-computer interface (BCI) paradigms. Conventional methods analyze SSVEPs at a fixed window length. Compared with these methods, dynamic window methods can achieve a higher information transfer rate (ITR) by selecting an appropriate window length. These methods dynamically evaluate the credibility of the result by linear discriminant analysis (LDA) or Bayesian estimation and extend the window length until credible results are obtained. However, the hypotheses introduced by LDA and Bayesian estimation may not align with the collected real-world SSVEPs, which leads to an inappropriate window length. To address the issue, we propose a novel dynamic window method based on reinforcement learning (RL). The proposed method optimizes the decision of whether to extend the window length based on the impact of decisions on the ITR, without additional hypotheses. The decision model can automatically learn a strategy that maximizes the ITR through trial and error. In addition, compared with traditional methods that manually extract features, the proposed method uses neural networks to automatically extract features for the dynamic selection of window length. Therefore, the proposed method can more accurately decide whether to extend the window length and select an appropriate window length. To verify the performance, we compared the novel method with other dynamic window methods on two public SSVEP datasets. The experimental results demonstrate that the novel method achieves the highest performance by using RL.

Hybrid Brain-Computer Interface Controlled Soft Robotic Glove for Stroke Rehabilitation.

Soft robotic glove controlled by a brain-computer interface (BCI) have demonstrated effectiveness in hand rehabilitation for stroke patients. Current systems mostly rely on static visual representations for patients to perform motor imagination (MI) tasks, resulting in lower BCI performance. Therefore, this study innovatively used MI and high-frequency steady-state visual evoked potential (SSVEP) to construct a friendly and natural hybrid BCI paradigm. Specifically, the stimulation interface sequentially presented decomposed action pictures of the left and right hands gripping a ball, with the pictures flashing at specific stimulation frequencies (left: 34 Hz, right: 35 Hz). Integrating soft robotic glove as feedback, we established a comprehensive "peripheral - central - peripheral" hand rehabilitation system to facilitate the hand rehabilitation of patients. Filter bank common spatial pattern (FBCSP) and filter bank canonical correlation analysis (FBCCA) algorithms were used to identify MI and SSVEP signals, respectively. Additionally, to fuse the features of these two signals, we proposed a novel fusion algorithm for improving the recognition accuracy of the system. The feasibility of the proposed system was validated through online experiments involving 12 healthy subjects and 9 stroke patients, achieving accuracy rates of 95.83 ± 6.83% and 63.33 ± 10.38%, respectively. The accuracy of MI and SSVEP in 12 healthy subjects reached 81.67 ± 15.63% and 95.14 ± 7.47%, both lower than the accuracy after fusion, these results confirmed the effectiveness of the proposed algorithm. The accuracy rate was more than 50% in both healthy subjects and patients, confirming the effectiveness of the proposed system.

Evaluation of an English language phoneme-based imagined speech brain computer interface with low-cost electroencephalography.

Paralyzed and physically impaired patients face communication difficulties, even when they are mentally coherent and aware. Electroencephalographic (EEG) brain-computer interfaces (BCIs) offer a potential communication method for these people without invasive surgery or physical device controls. Although virtual keyboard protocols are well documented in EEG BCI paradigms, these implementations are visually taxing and fatiguing. All English words combine 44 unique phonemes, each corresponding to a unique EEG pattern. In this study, a complete phoneme-based imagined speech EEG BCI was developed and tested on 16 subjects. Using open-source hardware and software, machine learning models, such as k-nearest neighbor (KNN), reliably achieved a mean accuracy of 97 ± 0.001%, a mean F1 of 0.55 ± 0.01, and a mean AUC-ROC of 0.68 ± 0.002 in a modified one-versus-rest configuration, resulting in an information transfer rate of 304.15 bits per minute. In line with prior literature, the distinguishing feature between phonemes was the gamma power on channels F3 and F7. However, adjustments to feature selection, trial window length, and classifier algorithms may improve performance. In summary, these are iterative changes to a viable method directly deployable in current, commercially available systems and software. The development of an intuitive phoneme-based EEG BCI with open-source hardware and software demonstrates the potential ease with which the technology could be deployed in real-world applications.

Rhythmic temporal prediction enhances neural representations of movement intention for brain–computer interface

Objective. Detecting movement intention is a typical use of brain–computer interfaces (BCI). However, as an endogenous electroencephalography (EEG) feature, the neural representation of movement is insufficient for improving motor-based BCI. This study aimed to develop a new movement augmentation BCI encoding paradigm by incorporating the cognitive function of rhythmic temporal prediction, and test the feasibility of this new paradigm in optimizing detections of movement intention. Methods. A visual-motion synchronization task was designed with two movement intentions (left vs. right) and three rhythmic temporal prediction conditions (1000 ms vs. 1500 ms vs. no temporal prediction). Behavioural and EEG data of 24 healthy participants were recorded. Event-related potentials (ERPs), event-related spectral perturbation induced by left- and right-finger movements, the common spatial pattern (CSP) and support vector machine, Riemann tangent space algorithm and logistic regression were used and compared across the three temporal prediction conditions, aiming to test the impact of temporal prediction on movement detection. Results. Behavioural results showed significantly smaller deviation time for 1000 ms and 1500 ms conditions. ERP analyses revealed 1000 ms and 1500 ms conditions led to rhythmic oscillations with a time lag in contralateral and ipsilateral areas of movement. Compared with no temporal prediction, 1000 ms condition exhibited greater beta event-related desynchronization (ERD) lateralization in motor area (P< 0.001) and larger beta ERD in frontal area (P< 0.001). 1000 ms condition achieved an averaged left–right decoding accuracy of 89.71% using CSP and 97.30% using Riemann tangent space, both significantly higher than no temporal prediction. Moreover, movement and temporal information can be decoded simultaneously, achieving 88.51% four-classification accuracy. Significance. The results not only confirm the effectiveness of rhythmic temporal prediction in enhancing detection ability of motor-based BCI, but also highlight the dual encodings of movement and temporal information within a single BCI paradigm, which is promising to expand the range of intentions that can be decoded by the BCI.

A Combined Virtual Electrode-Based ESA and CNN Method for MI-EEG Signal Feature Extraction and Classification.

A Brain-Computer Interface (BCI) is a medium for communication between the human brain and computers, which does not rely on other human neural tissues, but only decodes Electroencephalography (EEG) signals and converts them into commands to control external devices. Motor Imagery (MI) is an important BCI paradigm that generates a spontaneous EEG signal without external stimulation by imagining limb movements to strengthen the brain's compensatory function, and it has a promising future in the field of computer-aided diagnosis and rehabilitation technology for brain diseases. However, there are a series of technical difficulties in the research of motor imagery-based brain-computer interface (MI-BCI) systems, such as: large individual differences in subjects and poor performance of the cross-subject classification model; a low signal-to-noise ratio of EEG signals and poor classification accuracy; and the poor online performance of the MI-BCI system. To address the above problems, this paper proposed a combined virtual electrode-based EEG Source Analysis (ESA) and Convolutional Neural Network (CNN) method for MI-EEG signal feature extraction and classification. The outcomes reveal that the online MI-BCI system developed based on this method can improve the decoding ability of multi-task MI-EEG after training, it can learn generalized features from multiple subjects in cross-subject experiments and has some adaptability to the individual differences of new subjects, and it can decode the EEG intent online and realize the brain control function of the intelligent cart, which provides a new idea for the research of an online MI-BCI system.