Long-Term Mutual Training for the CYBATHLON BCI Race With a Tetraplegic Pilot: A Case Study on Inter-Session Transfer and Intra-Session Adaptation

Improving inter-session performance via relevant session-transfer for multi-session motor imagery classification

Motor imagery (MI)-based brain-computer interfaces (BCIs) using electroencephalography (EEG) have found practical applications in external device control. However, the non-stationary nature of EEG signals remains to obstruct BCI performance across multiple sessions, even for the same user. In this study, we aim to address the impact of non-stationarity, also known as inter-session variability, on multi-session MI classification performance by introducing a novel approach, the relevant session-transfer (RST) method. Leveraging the cosine similarity as a benchmark, the RST method transfers relevant EEG data from the previous session to the current one. The effectiveness of the proposed RST method was investigated through performance comparisons with the self-calibrating method, which uses only the data from the current session, and the whole-session transfer method, which utilizes data from all prior sessions. We validated the effectiveness of these methods using two datasets: a large MI public dataset (Shu Dataset) and our own dataset of gait-related MI, which includes both healthy participants and individuals with spinal cord injuries. Our experimental results revealed that the proposed RST method leads to a 2.29 % improvement (p < 0.001) in the Shu Dataset and up to a 6.37 % improvement in our dataset when compared to the self-calibrating method. Moreover, our method surpassed the performance of the recent highest-performing method that utilized the Shu Dataset, providing further support for the efficacy of the RST method in improving multi-session MI classification performance. Consequently, our findings confirm that the proposed RST method can improve classification performance across multiple sessions in practical MI-BCIs.

Evaluating the real-world usability of BCI control systems with augmented reality: a user study protocol

Brain-computer interfaces (BCI) enable users to control devices through their brain activity. Motor imagery (MI), the neural activity resulting from an individual imagining performing a movement, is a common control paradigm. This study introduces a user-centric evaluation protocol for assessing the performance and user experience of an MI-based BCI control system utilizing augmented reality. Augmented reality is employed to enhance user interaction by displaying environment-aware actions, and guiding users on the necessary imagined movements for specific device commands. One of the major gaps in existing research is the lack of comprehensive evaluation methodologies, particularly in real-world conditions. To address this gap, our protocol combines quantitative and qualitative assessments across three phases. In the initial phase, the BCI prototype's technical robustness is validated. Subsequently, the second phase involves a performance assessment of the control system. The third phase introduces a comparative analysis between the prototype and an alternative approach, incorporating detailed user experience evaluations through questionnaires and comparisons with non-BCI control methods. Participants engage in various tasks, such as object sorting, picking and placing, and playing a board game using the BCI control system. The evaluation procedure is designed for versatility, intending applicability beyond the specific use case presented. Its adaptability enables easy customization to meet the specific user requirements of the investigated BCI control application. This user-centric evaluation protocol offers a comprehensive framework for iterative improvements to the BCI prototype, ensuring technical validation, performance assessment, and user experience evaluation in a systematic and user-focused manner.

Transfer learning promotes acquisition of individual BCI skills

Subject training is crucial for acquiring brain-computer interface (BCI) control. Typically, this requires collecting user-specific calibration data due to high inter-subject neural variability that limits the usability of generic decoders. However, calibration is cumbersome and may produce inadequate data for building decoders, especially with naïve subjects. Here, we show that a decoder trained on the data of a single expert is readily transferrable to inexperienced users via domain adaptation techniques allowing calibration-free BCI training. We introduce two real-time frameworks, (i) Generic Recentering (GR) through unsupervised adaptation and (ii) Personally Assisted Recentering (PAR) that extends GR by employing supervised recalibration of the decoder parameters. We evaluated our frameworks on 18 healthy naïve subjects over five online sessions, who operated a customary synchronous bar task with continuous feedback and a more challenging car racing game with asynchronous control and discrete feedback. We show that along with improved task-oriented BCI performance in both tasks, our frameworks promoted subjects' ability to acquire individual BCI skills, as the initial neurophysiological control features of an expert subject evolved and became subject specific. Furthermore, those features were task-specific and were learned in parallel as participants practiced the two tasks in every session. Contrary to previous findings implying that supervised methods lead to improved online BCI control, we observed that longitudinal training coupled with unsupervised domain matching (GR) achieved similar performance to supervised recalibration (PAR). Therefore, our presented frameworks facilitate calibration-free BCIs and have immediate implications for broader populations-such as patients with neurological pathologies-who might struggle to provide suitable initial calibration data.

Transferring a deep learning model from healthy subjects to stroke patients in a motor imagery brain-computer interface

Objective.Motor imagery (MI) brain-computer interfaces (BCIs) based on electroencephalogram (EEG) have been developed primarily for stroke rehabilitation, however, due to limited stroke data, current deep learning methods for cross-subject classification rely on healthy data. This study aims to assess the feasibility of applying MI-BCI models pre-trained using data from healthy individuals to detect MI in stroke patients.Approach.We introduce a new transfer learning approach where features from two-class MI data of healthy individuals are used to detect MI in stroke patients. We compare the results of the proposed method with those obtained from analyses within stroke data. Experiments were conducted using Deep ConvNet and state-of-the-art subject-specific machine learning MI classifiers, evaluated on OpenBMI two-class MI-EEG data from healthy subjects and two-class MI versus rest data from stroke patients.Main results.Results of our study indicate that through domain adaptation of a model pre-trained using healthy subjects' data, an average MI detection accuracy of 71.15% (±12.46%) can be achieved across 71 stroke patients. We demonstrate that the accuracy of the pre-trained model increased by 18.15% after transfer learning (p<0.001). Additionally, the proposed transfer learning method outperforms the subject-specific results achieved by Deep ConvNet and FBCSP, with significant enhancements of 7.64% (p<0.001) and 5.55% (p<0.001) in performance, respectively. Notably, the healthy-to-stroke transfer learning approach achieved similar performance to stroke-to-stroke transfer learning, with no significant difference (p>0.05). Explainable AI analyses using transfer models determined channel relevance patterns that indicate contributions from the bilateral motor, frontal, and parietal regions of the cortex towards MI detection in stroke patients.Significance.Transfer learning from healthy to stroke can enhance the clinical use of BCI algorithms by overcoming the challenge of insufficient clinical data for optimal training.

Towards user-centric BCI design: Markov chain-based user assessment for mental imagery EEG-BCIs

Objective.While electroencephalography (EEG)-based brain-computer interfaces (BCIs) have many potential clinical applications, their use is impeded by poor performance for many users. To improve BCI performance, either via enhanced signal processing or user training, it is critical to understand and describe each user's ability to perform mental control tasks and produce discernible EEG patterns. While classification accuracy has predominantly been used to assess user performance, limitations and criticisms of this approach have emerged, thus prompting the need to develop novel user assessment approaches with greater descriptive capability. Here, we propose a combination of unsupervised clustering and Markov chain models to assess and describe user skill.Approach.Using unsupervisedK-means clustering, we segmented the EEG signal space into regions representing pattern states that users could produce. A user's movement through these pattern states while performing different tasks was modeled using Markov chains. Finally, using the steady-state distributions and entropy rates of the Markov chains, we proposed two metricstaskDistinctandrelativeTaskInconsistencyto assess, respectively, a user's ability to (i) produce distinct task-specific patterns for each mental task and (ii) maintain consistent patterns during individual tasks.Main results.Analysis of data from 14 adolescents using a three-class BCI revealed significant correlations between thetaskDistinctandrelativeTaskInconsistencymetrics and classification F1 score. Moreover, analysis of the pattern states and Markov chain models yielded descriptive information regarding user performance not immediately apparent from classification accuracy.Significance.Our proposed user assessment method can be used in concert with classifier-based analysis to further understand the extent to which users produce task-specific, time-evolving EEG patterns. In turn, this information could be used to enhance user training or classifier design.

Transfer Learning in Trajectory Decoding: Sensor or Source Space?

In this study, across-participant and across-session transfer learning was investigated to minimize the calibration time of the brain-computer interface (BCI) system in the context of continuous hand trajectory decoding. We reanalyzed data from a study with 10 able-bodied participants across three sessions. A leave-one-participant-out (LOPO) model was utilized as a starting model. Recursive exponentially weighted partial least squares regression (REW-PLS) was employed to overcome the memory limitation due to the large pool of training data. We considered four scenarios: generalized with no update (Gen), generalized with cumulative update (GenC), and individual models with cumulative (IndC) and non-cumulative (Ind) updates, with each one trained with sensor-space features or source-space features. The decoding performance in generalized models (Gen and GenC) was lower than the chance level. In individual models, the cumulative update (IndC) showed no significant improvement over the non-cumulative model (Ind). The performance showed the decoder's incapability to generalize across participants and sessions in this task. The results suggested that the best correlation could be achieved with the sensor-space individual model, despite additional anatomical information in the source-space features. The decoding pattern showed a more localized pattern around the precuneus over three sessions in Ind models.

Wearable System Based on Ultra-Thin Parylene C Tattoo Electrodes for EEG Recording

In an increasingly interconnected world, where electronic devices permeate every aspect of our lives, wearable systems aimed at monitoring physiological signals are rapidly taking over the sport and fitness domain, as well as biomedical fields such as rehabilitation and prosthetics. With the intent of providing a novel approach to the field, in this paper we discuss the development of a wearable system for the acquisition of EEG signals based on a portable, low-power custom PCB specifically designed to be used in combination with non-conventional ultra-conformable and imperceptible Parylene-C tattoo electrodes. The proposed system has been tested in a standard rest-state experiment, and its performance in terms of discrimination of two different states has been compared to that of a commercial wearable device for EEG signal acquisition (i.e., the Muse headset), showing comparable results. This first preliminary validation demonstrates the possibility of conveniently employing ultra-conformable tattoo-electrodes integrated portable systems for the unobtrusive acquisition of brain activity.

Competing at the Cybathlon championship for people with disabilities: long-term motor imagery brain-computer interface training of a cybathlete who has tetraplegia

BACKGROUND

The brain-computer interface (BCI) race at the Cybathlon championship, for people with disabilities, challenges teams (BCI researchers, developers and pilots with spinal cord injury) to control an avatar on a virtual racetrack without movement. Here we describe the training regime and results of the Ulster University BCI Team pilot who has tetraplegia and was trained to use an electroencephalography (EEG)-based BCI intermittently over 10 years, to compete in three Cybathlon events.

METHODS

A multi-class, multiple binary classifier framework was used to decode three kinesthetically imagined movements (motor imagery of left arm, right arm, and feet), and relaxed state. Three game paradigms were used for training i.e., NeuroSensi, Triad, and Cybathlon Race: BrainDriver. An evaluation of the pilot's performance is presented for two Cybathlon competition training periods-spanning 20 sessions over 5 weeks prior to the 2019 competition, and 25 sessions over 5 weeks in the run up to the 2020 competition.

RESULTS

Having participated in BCI training in 2009 and competed in Cybathlon 2016, the experienced pilot achieved high two-class accuracy on all class pairs when training began in 2019 (decoding accuracy > 90%, resulting in efficient NeuroSensi and Triad game control). The BrainDriver performance (i.e., Cybathlon race completion time) improved significantly during the training period, leading up to the competition day, ranging from 274-156 s (255 ± 24 s to 191 ± 14 s mean ± std), over 17 days (10 sessions) in 2019, and from 230-168 s (214 ± 14 s to 181 ± 4 s), over 18 days (13 sessions) in 2020. However, on both competition occasions, towards the race date, the performance deteriorated significantly.

CONCLUSIONS

The training regime and framework applied were highly effective in achieving competitive race completion times. The BCI framework did not cope with significant deviation in electroencephalography (EEG) observed in the sessions occurring shortly before and during the race day. Changes in cognitive state as a result of stress, arousal level, and fatigue, associated with the competition challenge and performance pressure, were likely contributing factors to the non-stationary effects that resulted in the BCI and pilot achieving suboptimal performance on race day. Trial registration not registered.

A Transfer Learning Algorithm to Reduce Brain-Computer Interface Calibration Time for Long-Term Users

Current motor imagery-based brain-computer interface (BCI) systems require a long calibration time at the beginning of each session before they can be used with adequate levels of classification accuracy. In particular, this issue can be a significant burden for long term BCI users. This article proposes a novel transfer learning algorithm, called r-KLwDSA, to reduce the BCI calibration time for long-term users. The proposed r-KLwDSA algorithm aligns the user's EEG data collected in previous sessions to the few EEG trials collected in the current session, using a novel linear alignment method. Thereafter, the aligned EEG trials from the previous sessions and the few EEG trials from the current sessions are fused through a weighting mechanism before they are used for calibrating the BCI model. To validate the proposed algorithm, a large dataset containing the EEG data from 11 stroke patients, each performing 18 BCI sessions, was used. The proposed framework demonstrated a significant improvement in the classification accuracy, of over 4% compared to the session-specific algorithm, when there were as few as two trials per class available from the current session. The proposed algorithm was particularly successful in improving the BCI accuracy of the sessions that had initial session-specific accuracy below 60%, with an average improvement of around 10% in the accuracy, leading to more stroke patients having meaningful BCI rehabilitation.

Deep learning for biosignal control: insights from basic to real-time methods with recommendations

Biosignal control is an interaction modality that allows users to interact with electronic devices by decoding the biological signals emanating from the movements or thoughts of the user. This manner of interaction with devices can enhance the sense of agency for users and enable persons suffering from a paralyzing condition to interact with everyday devices that would otherwise be challenging for them to use. It can also improve control of prosthetic devices and exoskeletons by making the interaction feel more natural and intuitive. However, with the current state of the art, several issues still need to be addressed to reliably decode user intent from biosignals and provide an improved user experience over other interaction modalities. One solution is to leverage advances in Deep Learning (DL) methods to provide more reliable decoding at the expense of added computational complexity. This scoping review introduces the basic concepts of DL and assists readers in deploying DL methods to a real-time control system that should operate under real-world conditions. The scope of this review covers any electronic device, but with an emphasis on robotic devices, as this is the most active area of research in biosignal control. We review the literature pertaining to the implementation and evaluation of control systems that incorporate DL to identify the main gaps and issues in the field, and formulate suggestions on how to mitigate them. Additionally, we formulate guidelines on the best approach to designing, implementing and evaluating research prototypes that use DL in their biosignal control systems.

On the interpretation of linear Riemannian tangent space model parameters in M/EEG

Riemannian tangent space methods offer state-of-the-art performance in magnetoencephalography (MEG) and electroencephalography (EEG) based applications such as brain-computer interfaces and biomarker development. One limitation, particularly relevant for biomarker development, is limited model interpretability compared to established component-based methods. Here, we propose a method to transform the parameters of linear tangent space models into interpretable patterns. Using typical assumptions, we show that this approach identifies the true patterns of latent sources, encoding a target signal. In simulations and two real MEG and EEG datasets, we demonstrate the validity of the proposed approach and investigate its behavior when the model assumptions are violated. Our results confirm that Riemannian tangent space methods are robust to differences in the source patterns across observations. We found that this robustness property also transfers to the associated patterns.

A Review on Signal Processing Approaches to Reduce Calibration Time in EEG-Based Brain-Computer Interface

In an electroencephalogram- (EEG-) based brain–computer interface (BCI), a subject can directly communicate with an electronic device using his EEG signals in a safe and convenient way. However, the sensitivity to noise/artifact and the non-stationarity of EEG signals result in high inter-subject/session variability. Therefore, each subject usually spends long and tedious calibration time in building a subject-specific classifier. To solve this problem, we review existing signal processing approaches, including transfer learning (TL), semi-supervised learning (SSL), and a combination of TL and SSL. Cross-subject TL can transfer amounts of labeled samples from different source subjects for the target subject. Moreover, Cross-session/task/device TL can reduce the calibration time of the subject for the target session, task, or device by importing the labeled samples from the source sessions, tasks, or devices. SSL simultaneously utilizes the labeled and unlabeled samples from the target subject. The combination of TL and SSL can take advantage of each other. For each kind of signal processing approaches, we introduce their concepts and representative methods. The experimental results show that TL, SSL, and their combination can obtain good classification performance by effectively utilizing the samples available. In the end, we draw a conclusion and point to research directions in the future.