The Cybathlon BCI race: Successful longitudinal mutual learning with two tetraplegic users.

This work aims at corroborating the importance and efficacy of mutual learning in motor imagery (MI) brain-computer interface (BCI) by leveraging the insights obtained through our participation in the BCI race of the Cybathlon event. We hypothesized that, contrary to the popular trend of focusing mostly on the machine learning aspects of MI BCI training, a comprehensive mutual learning methodology that reinstates the three learning pillars (at the machine, subject, and application level) as equally significant could lead to a BCI-user symbiotic system able to succeed in real-world scenarios such as the Cybathlon event. Two severely impaired participants with chronic spinal cord injury (SCI), were trained following our mutual learning approach to control their avatar in a virtual BCI race game. The competition outcomes substantiate the effectiveness of this type of training. Most importantly, the present study is one among very few to provide multifaceted evidence on the efficacy of subject learning during BCI training. Learning correlates could be derived at all levels of the interface-application, BCI output, and electroencephalography (EEG) neuroimaging-with two end-users, sufficiently longitudinal evaluation, and, importantly, under real-world and even adverse conditions.

Sensory threshold neuromuscular electrical stimulation fosters motor imagery performance.

Motor imagery (MI) has been largely studied as a way to enhance motor learning and to restore motor functions. Although it is agreed that users should emphasize kinesthetic imagery during MI, recordings of MI brain patterns are not sufficiently reliable for many subjects. It has been suggested that the usage of somatosensory feedback would be more suitable than standardly used visual feedback to enhance MI brain patterns. However, somatosensory feedback should not interfere with the recorded MI brain pattern. In this study we propose a novel feedback modality to guide subjects during MI based on sensory threshold neuromuscular electrical stimulation (St-NMES). St-NMES depolarizes sensory and motor axons without eliciting any muscular contraction. We hypothesize that St-NMES does not induce detectable ERD brain patterns and fosters MI performance. Twelve novice subjects were included in a cross-over design study. We recorded their EEG, comparing St-NMES with visual feedback during MI or resting tasks. We found that St-NMES not only induced significantly larger desynchronization over sensorimotor areas (p<0.05) but also significantly enhanced MI brain connectivity patterns. Moreover, classification accuracy and stability were significantly higher with St-NMES. Importantly, St-NMES alone did not induce detectable artifacts, but rather the changes in the detected patterns were due to an increased MI performance. Our findings indicate that St-NMES is a promising feedback in order to foster MI performance and cold be used for BMI online applications.

Learning to control a BMI-driven wheelchair for people with severe tetraplegia.

Mind-controlled wheelchairs are an intriguing assistive mobility solution applicable in complete paralysis. Despite progress in brain-machine interface (BMI) technology, its translation remains elusive. The primary objective of this study is to probe the hypothesis that BMI skill acquisition by end-users is fundamental to control a non-invasive brain-actuated intelligent wheelchair in real-world settings. We demonstrate that three tetraplegic spinal-cord injury users could be trained to operate a non-invasive, self-paced thought-controlled wheelchair and execute complex navigation tasks. However, only the two users exhibiting increasing decoding performance and feature discriminancy, significant neuroplasticity changes and improved BMI command latency, achieved high navigation performance. In addition, we show that dexterous, continuous control of robots is possible through low-degree of freedom, discrete and uncertain control channels like a motor imagery BMI, by blending human and artificial intelligence through shared-control methodologies. We posit that subject learning and shared-control are the key components paving the way for translational non-invasive BMI.

Context-Aware Learning for Generative Models.

This work studies the class of algorithms for learning with side-information that emerges by extending generative models with embedded context-related variables. Using finite mixture models (FMMs) as the prototypical Bayesian network, we show that maximum-likelihood estimation (MLE) of parameters through expectation-maximization (EM) improves over the regular unsupervised case and can approach the performances of supervised learning, despite the absence of any explicit ground-truth data labeling. By direct application of the missing information principle (MIP), the algorithms' performances are proven to range between the conventional supervised and unsupervised MLE extremities proportionally to the information content of the contextual assistance provided. The acquired benefits regard higher estimation precision, smaller standard errors, faster convergence rates, and improved classification accuracy or regression fitness shown in various scenarios while also highlighting important properties and differences among the outlined situations. Applicability is showcased with three real-world unsupervised classification scenarios employing Gaussian mixture models. Importantly, we exemplify the natural extension of this methodology to any type of generative model by deriving an equivalent context-aware algorithm for variational autoencoders (VAs), thus broadening the spectrum of applicability to unsupervised deep learning with artificial neural networks. The latter is contrasted with a neural-symbolic algorithm exploiting side information.

On Error-Related Potentials During Sensorimotor-Based Brain-Computer Interface: Explorations With a Pseudo-Online Brain-Controlled Speller.

Objective: Brain-computer interface (BCI) spelling is a promising communication solution for people in paralysis. Currently, BCIs suffer from imperfect decoding accuracy which calls for methods to handle spelling mistakes. Detecting error-related potentials (ErrPs) has been early identified as a potential remedy. Nevertheless, few works have studied the elicitation of ErrPs during engagement with other BCI tasks, especially when BCI feedback is provided continuously. Methods: Here, we test the possibility of correcting errors during pseudo-online Motor Imagery (MI) BCI spelling through ErrPs, and investigate whether BCI feedback hinders their generation. Ten subjects performed a series of MI spelling tasks with and without observing BCI feedback. Results: The average pseudo-online ErrP detection accuracy was found to be significantly above the chance level in both conditions and did not significantly differ between the two (74% with, and 78% without feedback). Conclusions: Our results support the possibility to detect ErrPs during MI-BCI spelling and suggest the absence of any BCI feedback-related interference.

Peri-personal space encoding in patients with disorders of consciousness and cognitive-motor dissociation.

Behavioral assessments of consciousness based on overt command following cannot differentiate patients with disorders of consciousness (DOC) from those who demonstrate a dissociation between intent/awareness and motor capacity: cognitive motor dissociation (CMD). We argue that delineation of peri-personal space (PPS) - the multisensory-motor space immediately surrounding the body - may differentiate these patients due to its central role in mediating human-environment interactions, and putatively in scaffolding a minimal form of selfhood. In Experiment 1, we determined a normative physiological index of PPS by recording electrophysiological (EEG) responses to tactile, auditory, or audio-tactile stimulation at different distances (5 vs. 75 cm) in healthy volunteers (N = 19). Contrasts between paired (AT) and summed (A + T) responses demonstrated multisensory supra-additivity when AT stimuli were presented near, i.e., within the PPS, and highlighted somatosensory-motor sensors as electrodes of interest. In Experiment 2, we recorded EEG in patients behaviorally diagnosed as DOC or putative CMD (N = 17, 30 sessions). The PPS-measure developed in Experiment 1 was analyzed in relation with both standard clinical diagnosis (i.e., Coma Recovery Scale; CRS-R) and a measure of neural complexity associated with consciousness. Results demonstrated a significant correlation between the PPS measure and neural complexity, but not with the CRS-R, highlighting the added value of the physiological recordings. Further, multisensory processing in PPS was preserved in putative CMD but not in DOC patients. Together, the findings suggest that indexing PPS allows differentiating between groups of patients whom both show overt motor impairments (DOC and CMD) but putatively distinct levels of awareness or motor intent.

Cortico-Muscular Coherence Is Reduced Acutely Post-stroke and Increases Bilaterally During Motor Recovery: A Pilot Study.

Motor recovery following stroke is believed to necessitate alteration in functional connectivity between cortex and muscle. Cortico-muscular coherence has been proposed as a potential biomarker for post-stroke motor deficits, enabling a quantification of recovery, as well as potentially indicating the regions of cortex involved in recovery of function. We recorded simultaneous EEG and EMG during wrist extension from healthy participants and patients following ischaemic stroke, evaluating function at three time points post-stroke. EEG-EMG coherence increased over time, as wrist mobility recovered clinically, and by the final evaluation, coherence was higher in the patient group than in the healthy controls. Moreover, the cortical distribution differed between the groups, with coherence involving larger and more bilaterally scattered areas of cortex in the patients than in the healthy participants. The findings suggest that EEG-EMG coherence has the potential to serve as a biomarker for motor recovery and to provide information about the cortical regions that should be targeted in rehabilitation therapies based on real-time EEG.

Motor Attempt EEG Paradigm as a Diagnostic Tool for Disorders of Consciousness.

To investigate whether a motor attempt EEG paradigm coupled with functional electrical stimulation can detect command following and, therefore, signs of conscious awareness in patients with disorders of consciousness, we recorded nine patients admitted to acute rehabilitation after a brain lesion. We extracted peak classification accuracy and peak session discriminant power (PSDP) and we assessed their correlation to the established coma recovery scale revised (CRS-R) and the agreement with diagnosis based on the novel motor behavior tool (MBT). Only PSDP correlated significantly with CRS-R and it also outperformed peak accuracy regarding the MBT. We conclude that PSDP might be more suitable than accuracy to complement CRS-R and MBT in evaluating ambiguous cases and in detecting cognitive motor dissociation.

Quantifying electrode reliability during brain-computer interface operation.

One of the problems of noninvasive brain-computer interface (BCI) applications is the occurrence of anomalous (unexpected) signals that might degrade BCI performance. This situation might slip the operator's attention since raw signals are not usually continuously visualized and monitored during BCI-actuated device operation. Anomalous data can for instance be the result of electrode misplacement, degrading impedance or loss of connectivity. Since this problem can develop at run time, there is a need of a systematic approach to evaluate electrode reliability during online BCI operation. In this paper, we propose two metrics detecting how much each channel is deviating from its expected behavior. This quantifies electrode reliability at run time which could be embedded into BCI data processing to increase performance. We assess the effectiveness of these metrics in quantifying signal degradation by conducting three experiments: Electrode swap, electrode manipulation, and offline artificially degradation of P300 signals.

Subject-oriented training for motor imagery brain-computer interfaces.

Successful operation of motor imagery (MI)-based brain-computer interfaces (BCI) requires mutual adaptation between the human subject and the BCI. Traditional training methods, as well as more recent ones based on co-adaptation, have mainly focused on the machine-learning aspects of BCI training. This work presents a novel co-adaptive training protocol shifting the focus on subject-related performances and the optimal accommodation of the interactions between the two learning agents of the BCI loop. Preliminary results with 8 able-bodied individuals demonstrate that the proposed method has been able to bring 3 naive users into control of a MI BCI within a few runs and to improve the BCI performances of 3 experienced BCI users by an average of 0.36 bits/sec.

A hybrid BCI for enhanced control of a telepresence robot.

Motor-disabled end users have successfully driven a telepresence robot in a complex environment using a Brain-Computer Interface (BCI). However, to facilitate the interaction aspect that underpins the notion of telepresence, users must be able to voluntarily and reliably stop the robot at any moment, not just drive from point to point. In this work, we propose to exploit the user's residual muscular activity to provide a fast and reliable control channel, which can start/stop the telepresence robot at any moment. Our preliminary results show that not only does this hybrid approach increase the accuracy, but it also helps to reduce the workload and was the preferred control paradigm of all the participants.

Transferring brain-computer interfaces beyond the laboratory: successful application control for motor-disabled users.

OBJECTIVES: Brain-computer interfaces (BCIs) are no longer only used by healthy participants under controlled conditions in laboratory environments, but also by patients and end-users, controlling applications in their homes or clinics, without the BCI experts around. But are the technology and the field mature enough for this? Especially the successful operation of applications - like text entry systems or assistive mobility devices such as tele-presence robots - requires a good level of BCI control. How much training is needed to achieve such a level? Is it possible to train naïve end-users in 10 days to successfully control such applications? MATERIALS AND METHODS: In this work, we report our experiences of training 24 motor-disabled participants at rehabilitation clinics or at the end-users' homes, without BCI experts present. We also share the lessons that we have learned through transferring BCI technologies from the lab to the user's home or clinics. RESULTS: The most important outcome is that 50% of the participants achieved good BCI performance and could successfully control the applications (tele-presence robot and text-entry system). In the case of the tele-presence robot the participants achieved an average performance ratio of 0.87 (max. 0.97) and for the text entry application a mean of 0.93 (max. 1.0). The lessons learned and the gathered user feedback range from pure BCI problems (technical and handling), to common communication issues among the different people involved, and issues encountered while controlling the applications. CONCLUSION: The points raised in this paper are very widely applicable and we anticipate that they might be faced similarly by other groups, if they move on to bringing the BCI technology to the end-user, to home environments and towards application prototype control.