A large electroencephalographic motor imagery dataset for electroencephalographic brain computer interfaces

Human-machine systems designed for daily activities must balance direct user control with autonomous system behaviour to reduce cognitive effort while maintaining responsiveness. Douglas et al. have demonstrated how shared-autonomy frameworks can enable BCIs to operate across multiple levels of user involvement, from direct neural control to high-level goal specification. Rather than relying solely on low-level EEG commands, the proposed approach integrates intent inference with adaptive robotic planning, allowing multiple users to coordinate multiple robots in functional daily tasks. A key insight from this contribution is that increasing autonomy can significantly reduce user workload while preserving task accuracy and efficiency, supporting more sustainable long-term BCI use in real-world settings. This work highlights a broader shift in BCI design toward collaborative control paradigms, where machines proactively assist users instead of acting only as passive command receivers.Single-modality EEG BCIs are often limited by noise, signal ambiguity, and performance variability across users and environments. Hybrid approaches that fuse EEG with additional sensing modalities offer a promising solution, as demonstrated by Coutray and their team. Coutray et al. have proven how integrating EEG with eye-tracking improves command disambiguation, interaction speed, and overall control reliability, particularly in immersive virtual reality environments. This hybrid design reduces reliance on high-precision EEG decoding alone and enables more natural, intuitive hands-free interaction, expanding BCI applicability beyond clinical contexts into entertainment, accessibility, and smartenvironment control. This contribution reinforces the growing consensus that multimodal BCIs are more scalable, resilient, and user-friendly than EEG-only systems. Robust BCI performance depends not only on decoding algorithms but also on the user's cognitive and neural state. Mohamed et al. have provided evidence that baseline neural oscillations, particularly precue alpha activity, influence event-related desynchronisation (ERD) strength, a core signal used in motorimagery BCIs. A concrete implication of this finding is that BCIs could dynamically adjust classification thresholds, training protocols, or feedback timing based on real-time cognitive state estimates, potentially improving accuracy, consistency, and user learning rates. These insights support the development of state-aware BCIs that adapt decoding strategies in response to moment-to-moment brain dynamics, improving reliability in naturalistic, everyday environments.Beyond EEG decoding alone, combining physiological stress markers and subjective workload measures enables more comprehensive prediction of user performance in demanding tasks. Wei et al. have demonstrated that interpretable machine-learning models can forecast task outcomes while revealing which physiological factors most strongly influence performance. This is particularly relevant for realworld BCI deployment, where fatigue, stress, and cognitive overload can degrade system reliability. A key insight is that interpretable predictive models can support adaptive intervention strategies, such as adjusting task difficulty or providing rest prompts, while maintaining transparency and user trust.Across the contributions, several overarching themes emerge:1. Multimodal integration: Combining EEG with complementary sensors enhances decoding robustness and interaction flexibility. 2. Adaptive autonomy: Systems that balance user intent with automated assistance reduce workload and improve task efficiency. 3. Machine-learning innovation: Deep and interpretable models improve accuracy while supporting transparency and trust. 4. Cognitive context awareness: Accounting for neural and psychological state enables more reliable and personalised BCI performance.Together, these themes suggest a transition from static, single-user BCIs toward adaptive, context-aware, and multi-agent systems designed for long-term everyday use.

Brain–computer interfaces (BCIs), also known as brain–machine interfaces (BMIs), translate brain activity into new outputs that replace, restore, enhance, supplement or improve natural brain outputs. BCI research and development has grown rapidly for the past two decades. It is beginning to provide useful communication and control capacities to people with severe neuromuscular disabilities; and it is expanding into new areas such as neurorehabilitation that may greatly increase its clinical impact. At the same time, significant challenges remain, particularly in regard to translating laboratory advances into clinical use.

Physical principles of brain–computer interfaces and their applications for rehabilitation, robotics and control of human brain states

Brain-computer interfaces (BCIs) can provide non-muscular communication and control for people with severe motor disabilities. Current BCIs use a variety of invasive and non-invasive methods to record brain signals and a variety of signal processing methods. Whatever the recording and processing methods used, BCI performance (e.g. the ability of a BCI to control movement of a computer cursor) is highly variable and, by the standards applied to neuromuscular control, could be described as ataxic. In an effort to understand this imperfection, this paper discusses the relevance of two principles that underlie the brain's normal motor outputs. The first principle is that motor outputs are normally produced by the combined activity of many CNS areas, from the cortex to the spinal cord. Together, these areas produce appropriate control of the spinal motoneurons that activate muscles. The second principle is that the acquisition and life-long preservation of motor skills depends on continual adaptive plasticity throughout the CNS. This plasticity optimizes the control of spinal motoneurons. In the light of these two principles, a BCI may be viewed as a system that changes the outcome of CNS activity from control of spinal motoneurons to, instead, control of the cortical (or other) area whose signals are used by the BCI to determine the user's intent. In essence, a BCI attempts to assign to cortical neurons the role normally performed by spinal motoneurons. Thus, a BCI requires that the many CNS areas involved in producing normal motor actions change their roles so as to optimize the control of cortical neurons rather than spinal motoneurons. The disconcerting variability of BCI performance may stem in large part from the challenge presented by the need for this unnatural adaptation. This difficulty might be reduced, and BCI development might thereby benefit, by adopting a 'goal-selection' rather than a 'process- control' strategy. In 'process control', a BCI manages all the intricate high-speed interactions involved in movement. In 'goal selection', by contrast, the BCI simply communicates the user's goal to software that handles the high-speed interactions needed to achieve the goal. Not only is 'goal selection' less demanding, but also, by delegating lower-level aspects of motor control to another structure (rather than requiring that the cortex do everything), it more closely resembles the distributed operation characteristic of normal motor control.

Brain-computer interfaces (BCIs) can provide communication and control to people who are totally paralyzed. BCIs can use noninvasive or invasive methods for recording the brain signals that convey the user's commands. Whereas noninvasive BCIs are already in use for simple applications, it has been widely assumed that only invasive BCIs, which use electrodes implanted in the brain, can provide multidimensional movement control of a robotic arm or a neuroprosthesis. We now show that a noninvasive BCI that uses scalp-recorded electroencephalographic activity and an adaptive algorithm can provide humans, including people with spinal cord injuries, with multidimensional point-to-point movement control that falls within the range of that reported with invasive methods in monkeys. In movement time, precision, and accuracy, the results are comparable to those with invasive BCIs. The adaptive algorithm used in this noninvasive BCI identifies and focuses on the electroencephalographic features that the person is best able to control and encourages further improvement in that control. The results suggest that people with severe motor disabilities could use brain signals to operate a robotic arm or a neuroprosthesis without needing to have electrodes implanted in their brains.

Brain-computer interface (BCI) technology bridges the direct communication between the brain and machines, unlocking new possibilities for human interaction and rehabilitation. EEG-based motor imagery (MI) plays a pivotal role in BCI, enabling the translation of thought into actionable commands for interactive and assistive technologies. However, the constrained decoding performance of brain signals poses a limitation to the broader application and development of BCI systems. In this study, we introduce a convolutional Transformer network (CTNet) designed for EEG-based MI classification. Firstly, CTNet employs a convolutional module analogous to EEGNet, dedicated to extracting local and spatial features from EEG time series. Subsequently, it incorporates a Transformer encoder module, leveraging a multi-head attention mechanism to discern the global dependencies of EEG's high-level features. Finally, a straightforward classifier module comprising fully connected layers is followed to categorize EEG signals. In subject-specific evaluations, CTNet achieved remarkable decoding accuracies of 82.52% and 88.49% on the BCI IV-2a and IV-2b datasets, respectively. Furthermore, in the challenging cross-subject assessments, CTNet achieved recognition accuracies of 58.64% on the BCI IV-2a dataset and 76.27% on the BCI IV-2b dataset. In both subject-specific and cross-subject evaluations, CTNet holds a leading position when compared to some of the state-of-the-art methods. This underscores the exceptional efficacy of our approach and its potential to set a new benchmark in EEG decoding.

This paper presents an eye movement detection system based on EEG-based brain-computer interface (BCI). The proposed system performs feature extraction of the EEG signals and uses hidden Markov model (HMM) to perform eye movement detection. The proposed system is implemented on an FPGA board to form a real time BCI control system with a laptop. The users wearing a EEG headset can control a game on the computer by moving their eyeballs toward directions of left/right and up/down. This proposed system achieves a nearly 90% detection rate without training.

Brain-Computer Interface

Brain-computer interfaces (BCIs)—systems that can record neural activity and translate them into commands for computer systems—are sufficiently advanced to allow users to volitionally guide them through simple tasks. Contemporary BCI research focuses on squeezing additional functionality out of standardized paradigms, be it achieving more bits per second, increased degrees of freedom, or increasing accuracy. While these studies have shown marginal advancements in recent years, our lack of understanding concerning the underlying neurophysiology continues to be the limiting factor in BCI development. In this chapter, we propose turning the way research is done on BCI systems on its head; instead of using our understanding of neural signals to incrementally advance the state of brain-machine interfaces, we apply a BCI system as a form of experimental control to study changes in neural activity. By using current BCI systems as a tool for neuroscientific study, we can probe the underlying neuroanatomy in novel, behaviorally controlled ways.

For a practical intracranial brain computer interface (BCI), minimizing the invasiveness of the electrode implantation is crucial. In this study, we used only one intracranial electrode to implement an online BCI for fast typing. When the subject attended the virtual button containing visual motion stimuli, prominent responses were elicited at the stereo-EEG (SEEG) electrodes within the fMRI defined middle temporal (MT) region, which were composed of motion-onset visual evoked potential (mVEP) around 200 ms post-stimulus and a power increase at the high gamma (70–100 Hz) frequency range. In the online BCI experiment with surgical epilepsy patients, by combining both mVEP and high gamma features and using smart stopping strategy, single SEEG electrode supported a speed of BCI typing up to 14 characters per minute. Our findings demonstrate the feasibility of implementing a minimally invasive intracranial BCI with only one electrode for fast typing.

<title>Abstract</title> Background: Amyotrophic Lateral Sclerosis (ALS) presents significant challenges for biomarker identification and brain-computer interface (BCI) development due to progressive motor neuron degeneration and high signal-to-noise ratios in neurophysiological recordings. Traditional EEG analysis methods struggle with patient-specific variability and environmental artifacts, limiting their clinical utility. Methods: We present an adaptive hybrid framework integrating dendritic-inspired noise filtering, quantum-inspired symbolic search with Hamming distance optimization, and meta-adaptive machine learning (MAML) for personalized adaptation. The system processes synchronized EEG and eye-tracking data from the EEGET-ALS dataset (45 ALS patients, 42 healthy controls), implementing multiple threshold optimization strategies including Bayesian optimization, evolutionary algorithms, and ensemble methods. The framework maintains a historical performance database enabling continuous learning and transfer learning capabilities across patient cohorts. Results: Our framework achieved an F1-score of 0.967 (95% CI: 0.958-0.976), precision of 0.98, and recall of 0.95 in distinguishing ALS patients from healthy controls (Figure 1o). Key neurophysiological findings include ~25% beta power reduction (13-30 Hz) in central electrodes (Figure 1a-c), 40% motor connectivity loss particularly in C3-C4 coupling (Figure 1e-g), and preserved gamma oscillatory patterns (Figure 1d). The meta-learning module improved performance from baseline F1=0.667 to F1=0.751 within 1000 search iterations (Figure 1h), with transfer learning accelerating convergence by 5-10x (Figure 1l). Conclusions: The quantum-inspired adaptive framework establishes a new paradigm for ALS biomarker discovery and BCI development. By combining biologically-inspired filtering with continuous meta-learning, the system achieves clinical-grade accuracy while adapting to patient-specific variations. This approach enables robust communication interfaces for ALS patients and provides a generalizable framework for neurological biomarker discovery.

Brain-computer interfaces (BCIs) use neural information recorded from the brain for the voluntary control of external devices. The development of BCI systems has largely focused on improving functional independence for individuals with severe motor impairments, including providing tools for communication and mobility. In this review, we describe recent advances in intracortical BCI technology and provide potential directions for further research.

Most brain-computer interfaces (BCIs) currently under development use the brain's electrical signals. Nevertheless, nonelectrical metabolic signals also have potential for use in BCI development. Two methods currently available for measuring brain metabolic activity that are of greatest immediate interest for BCI development are: functional near-infrared spectroscopy (fNIRS) and functional magnetic resonance imaging (fMRI). fNIRS has the advantages of being noninvasive and inexpensive. fMRI has the advantages of being noninvasive and providing very high spatial resolution. This chapter focuses on BCIs based on fNIRS and fMRI methods. It reviews the fundamental principles underlying their use, the factors important in their use for BCIs, the kinds of BCI applications that are most promising, and possible future directions and challenges.

Brain-computer interface (BCI) technologies, such as deep brain stimulation and transcranial magnetic stimulation, have clear medical benefits and are largely familiar to the medical community. Meanwhile, companies are developing BCIs for personal use, to promote cognitive performance and assist with everyday tasks. On a recent episode of the ANA Investigates podcast, Dr Karen Rommelfanger, professor at Emory University and founder of the Institute of Neuroethics, discusses the neuroethics of medical and commercial applications of BCI. BCI encompasses both sensing technologies (which monitor brain activity) and stimulating devices (which alter brain pathways). Dr Rommelfanger begins the podcast by distinguishing between the two. She emphasizes that many commercially available devices tend to be wearable sensing technologies as an effort to be more easily accessible, affordable, and less invasive than clinical technology.1 At the present moment, she says we should not fall prey to popular media and sci-fi tropes about so-called “mind reading” or “mind control.” However, Dr Rommelfanger says that there are legitimate concerns around privacy with existing consumer-ready devices, both around how companies use the data they collect and whether that data are adequately secured as there are gaps in regulations, particularly around potential inferences that might be made about individuals' current or future brain function. In the podcast, Dr Rommelfanger discusses how devices especially in the clinical research realm that can sense, but also stimulate are the most advanced of BCIs; for example, brain mapping to enable communication and speech in those who have lost the ability to do so and affecting aspects of personality and emotional range with implantable electrodes for intractable depression. She makes the point later in the podcast that neuroethics can guide the development in these clinical and consumer devices as well, such as considering the inclusiveness of different ethnicities and hair types when developing the standard electroencephalogram (EEG) leads.2 She also addresses the neuroethics related to brain organoids, tiny 3-dimensional aggregates of brain cells currently used in research and development of novel treatments, some of which have been observed to produce spontaneous electrical activity that resembles the activity of brains of premature human embryos. She outlines how questions about the moral status of brain organoids may influence how organoids are handled and how tissue donors are consented for research. Dr Rommelfanger concludes the podcast by discussing the clinicians' role in helping patients understand the utility and limits of commercially available BCI technologies and advocating for greater clinician involvement in BCI development.3 She suggests that clinicians familiarize themselves with information from the Institute of Neuroethics Think and Do Tank (see Institute of Neuroethics Guidance), International Neuroethics Society, and the Dana Foundation to educate patients and fellow clinicians attempting to navigate this field in applied and practical neuroethics and related policy.4, 5 On a societal level, she points to a role for physicians in promoting access and medical reimbursement of BCI technologies. She also emphasizes the need to promote equity in the development of BCI, both to improve user experiences across diverse groups and to prevent misuse of technologies. R.D.S. conducted the primary podcast interview and wrote the first draft of the manuscript. R.D.S., A.J.C., K.S.R., and A.L.G. participated in the podcast interview and edited the manuscript. R.D.S., A.J.C., A.L.G. have nothing to disclose. K.S.R. offers consulting for policy entities, foundations, and companies related to neurotechnology and ethics strategy.

EEG event-related potentials, and the P300 signal in particular, are promising modalities for brain-computer interfaces (BCI). But the nonstationarity of EEG signals and their differences across individuals have made it difficult to implement classifiers that can determine user intent without having to be retrained or calibrated for each new user and sometimes even each session. This is a major impediment to the development of consumer BCI. Recently, the EEG BCI literature has begun to apply convolutional neural networks (CNNs) for classification, but experiments have largely been limited to training and testing on single subjects. In this paper, we report a study in which EEG data were recorded from 66 subjects in a visual oddball task in virtual reality. Using wide residual networks (WideResNets), we obtain state-of-the-art performance on a test set composed of data from all 66 subjects together. Additionally , a minimal preprocessing stream to convert EEG data into square images for CNN input while adding regularization is presented and shown to be viable. This study also provides some guidance on network architecture parameters based on experiments with different models. Our results show that it may possible with enough data to train a classifier for EEG-based BCIs that can generalize across individuals without the need for individual training or calibration.