Kapitel

• FrontmatterMüller-Putz, Gernot; Kostoglou, Kyriaki; Oberndorfer, Markus E.; Wriessnegger, Selina C.; 10.3217/978-3-99161-014-4-000
• WORD PREDICTION DURING NATURALISTIC SPEECH PERCEPTIONLekhnitskaya, Polina A.; 10.3217/978-3-99161-014-4-001The mechanisms of word prediction have not been studied in the natural speech perception paradigm, which formed the aim of the study: to explore the connection between the function of the EEG responses and the omitted words during naturalistic speech perception, confidence score of trained language model. 14 neurotypical subjects (mean age - 23,5 years; 5 males) participated in the research. EEG included 24 channels. It was proposed to listen to the story and comprehend it. The obtained results show differences in listening to omitted and non-omitted words in T3, T5, P3 electrodes. For modelling the connection between neural signals and naturalistic speech stimuli, mTRF was applied. One of the possible future directions of the research is to explore the communication processes in this paradigm.
• INVESTIGATING TEMPORAL VARIATIONS IN MRCPS AND THEIR INFLUENCE ON CLASSIFICATION: A 10-HOUR EEG STUDYEgger, Johanna; Kostoglou, Kyriaki; Müller-Putz, Gernot R.; 10.3217/978-3-99161-014-4-002Locked-in patients rely on stable performance of BCIs to provide them with a means of communication. To build a robust BCI, we demonstrate the need for adaptive decoding that accounts for temporal variations in electroencephalogram (EEG) dynamics. We analyzed six consecutive EEG sessions recorded between 2p.m. (afternoon). and 12a.m. (midnight) of 15 healthy participants engaged in a four-right-hand gesture task. We employed four-class classifiers trained on movement-related cortical potentials of different sessions and applied the decoders to the same session to evaluate the impact of temporal fluctuations in EEG on decoding capabilities. As a step towards adaptive decoding, we developed constantly updated classifiers by training on the most recently collected data and compared these to a stationary classifier trained once on the first session. Our findings revealed that temporal variations in EEG during movement tasks influence classification performance. In this context, we demonstrated that adaptive decoding provides a remedy to build a robust BCI usable for patients in the home- environment.
• S-JEPA: TOWARDS SEAMLESS CROSS-DATASET TRANSFER THROUGH DYNAMIC SPATIAL ATTENTIONGuetschel, Pierre; Moreau, Thomas; Tangermann, Michael; 10.3217/978-3-99161-014-4-003Motivated by the challenge of seamless cross-dataset transfer in EEG signal processing, this ar- ticle presents an exploratory study on the use of Joint Embedding Predictive Architectures (JEPAs). In recent years, self-supervised learning has emerged as a promis- ing approach for transfer learning in various domains. However, its application to EEG signals remains largely unexplored. In this article, we introduce Signal-JEPA for representing EEG recordings which includes a novel domain-specific spatial block masking strategy and three novel architectures for downstream classification. The study is conducted on a 54 subjects dataset and the downstream performance of the models is evaluated on three different BCI paradigms: motor imagery, ERP and SSVEP. Our study provides preliminary evidence for the potential of JEPAs in EEG signal encoding. Notably, our results highlight the importance of spatial filtering for accurate downstream classification and reveal an influence of the length of the pre-training examples but not of the mask size on the downstream performance.
• OPTIMIZING TIME-VARYING AUTOREGRESSIVE MODELS FOR BCI APPLICATIONSKostoglou, Kyriaki; Müller-Putz, Gernot R.; 10.3217/978-3-99161-014-4-004Monitoring the spectral characteristics of brain signals can provide insights into the underlying processes responsible for their generation. In brain- computer interface (BCI) applications, this is relatively important in decoding neural activity as it can provide a means to differentiate between various tasks or mental states. To capture spectral variations, herein, we focus on time-varying autoregressive models (TVAR). We introduce a framework designed to efficiently optimize and apply these models to multi-trial and multi-channel data, including electroencephalography (EEG) signals. Our approach was validated using EEG data from motor imagery tasks.
• RECOGNITION OF PERTURBATION EVOKED POTENTIAL BY USING MIXED- DEPTHWISE CONVOLUTIONSJalilpour, Shayan; Müller-Putz, Gernot R.; 10.3217/978-3-99161-014-4-005Prior studies have explored the capability of decoding balance perturbations using electroencephalography (EEG) in single-trial classifications. The potential for real-time detection of perturbation-evoked potentials (PEPs) could facilitate the implementation of brain-computer interfaces (BCIs) in everyday assistive systems. Achieving the detection of these potentials in a subject-independent manner is crucial for this advancement. A key step towards this objective is the development of a model capable of identifying balance loss without requiring individual calibration for each subject and enabling online analysis. Deep neural networks have recently achieved significant milestones and have been successfully applied in neural engineering. In this study, we propose a lightweight neural network to assess the viability of single-trial classification of PEPs in a subject-independent manner. Our model was tested on three balance perturbation datasets, demonstrating superior performance in subjectindependent classification compared to EEGNet, rLDA, and RBF-SVM classifiers.
• INTRODUCING THE ASME-SPELLER, AUDITORY BCI SPELLER UTILIZING STREAM SEGREGATION: A PILOT STUDYKojima, Simon; Kanoh, Shin'ichiro; 10.3217/978-3-99161-014-4-006The auditory BCI spellers are considered the only means of communication for late-stage patients with severe neurological disorders such as amyotrophic lateral sclerosis (ALS). To date, several auditory BCI spellers have been proposed. However, they require mul- tiple steps, visual support, or multi-channel audio systems. In this study, we proposed an ASME-speller, which stands for Auditory Stream segregation, Multiclass, ERP speller, that uses an auditory BCI paradigm based on auditory stream segregation to detect the target of the user’s selective attention by presenting a QWERTY keyboardlike audio stimuli. The 64-channel electroencephalogram was measured while the six subjects carried out 15-character ASME-speller paradigms. Offline simulation using dynamic stopping showed that the ASME speller achieved an average accuracy of 0.73 and an average ITR of 3.78 bits/min. The best results were achieved with an accuracy of 0.97 and an ITR of 7.61 bits/min. These results indicate that the ASME speller can be used as a new auditory BCI speller. This study provides more users with a high-accuracy and intuitive new speller option.
• A NEW AUDITORY BRAIN-COMPUTER INTERFACE BASED ON STREAM SEGREGATION UTILIZING ASSRKanoh, Shin'ichiro; Mizukami, Naoki; Kojima, Simon; 10.3217/978-3-99161-014-4-007The authors have proposed an auditory brain-computer interface (BCI) based on stream segregation which detects users’ selective attention to one the multiple segregated streams. In this system, several oddball sequences with different frequency bands were presented to users. To detect the target stream, this system needed to wait for the arrival of deviant stimuli in oddball sequences. In this study, auditory steady-state response (ASSR) was utilized to achieve a higher information transfer rate (ITR) system. Two streams consisting of sinusoidally amplitude-modulated (SAM) tones were presented to subjects, and they were requested to attend to one of the two streams. From the results of the electroencephalogram (EEG) measurement experiment, it was found that the user’s selective attention enhanced ASSR corresponding to the modulation frequency of SAM sounds in the target tone stream, and the target stream which subject paid attention to was detected at an average accuracy of 0.77. The best accuracy was 0.92. It was concluded that an auditory BCI based on stream segregation utilizing ASSR is feasible, and it provides high performance and practical BCI options to users.
• A NOVEL CHATGPT-DRIVEN COMMUNICATION AID BASED ON CODE-MODULATED VISUAL EVOKED POTENTIALS (CVEP)Cantürk, Atilla; Volosyak, Ivan; 10.3217/978-3-99161-014-4-008Brain-computer interface (BCI) systems, including applications based on visual evoked potentials (VEPs), have proven to provide reliable and accurate control. In recent years, communication has remained one of the main application areas of modern BCIs, with a lot of advancements based e.g., on the incorporation of dictionary support and text prediction. This study explores the integration of BCIs with artificial intelligence (AI), specifically focusing on the development and evaluation of an innovative spelling interface powered by the ChatGPT application programming interface (API). Aimed at enhancing communication for individuals with severe motor impairments, this interface combines the precision of code-modulated visual evoked potentials (cVEPs) with the predictive capabilities of AI to offer a more intuitive and efficient user experience. The performance of 13 healthy participants (10 females) was evaluated in an online experiment. The participants successfully completed all spelling tasks using the cVEP BCI with aid from ChatGPT, achieving a mean information transfer rate (ITR) of 33.16 bpm, a mean accuracy of 87.49%, and an average output of 8.74 output characters per minute (OCM) for unique sentence tasks. This was slower than in our previous research using an n-gram model which achieved 18.9 characters per minute.
• A STUDY OF PERFORMANCE VARIABILITY IN DEEP NEURAL NETWORKS FOR MOTOR IMAGERY CLASSIFICATION: TOWARDS A ZERO-CALIBRATION APPROACHArpaia, Pasquale; Esposito, Antonio; Galdieri, Fortuna; Natalizio, Angela; Parvis, Marco; Pollastro, Andrea; 10.3217/978-3-99161-014-4-009This study deals with the adoption of deep learning and transfer learning in motor imagery-based brain-computer interfaces to develop a robust system with a zero-calibration approach. Deep neural networks would be also sought to improve the classification accuracies of these interfaces. However, these approaches are affected by inherent variability in their performance, so that dominating uncertainty sources appears crucial. To assess the performance variability of deep neural networks, the effects of parameter initialisation and pre-processing were studied. EEGNet and Sinc-EEGNet were used for this purpose. The results highlight that network’s weight initialisation significantly affect the performance. For instance, classification accuracy can improve from 67 % ± 3 % to 73 % ± 3 % by just changing the weight initialisation. Meanwhile, EEG pre-processing does not improve the performance, thus it can be avoided to reduce the computational effort. These results pave the way for real-time application scenarios. Keyword: brain-computer interface, motor imagery, deep learning, transfer learning, uncertainty.
• DEEP LEARNING FOR MOTOR IMAGERY-BASED BCIS USING SEEG SIGNALSFu, Zhichun; Wu, Xiaolong; Gao, Xin; Zhang, Dingguo; 10.3217/978-3-99161-014-4-010Motor imagery (MI) is the most popular paradigm for brain-computer interfaces (BCIs) based on scalp electroencephalography (EEG), while this paradigm is missing for stereo-electroencephalography (sEEG)-based BCIs. Recently, the first public dataset of sEEG has become available for MI-based BCIs. However, the performance using traditional methods is still inferior. In this study, we employed some state-of- the-art methods based on deep learning to improve the classification accuracy of MI for sEEG-based BCIs. Six different deep learning models were developed, which include Shallow ConvNet, DeepNet, ResNet20, conformer, vision transformer (ViT) and ViT with pretrained parameters. Among six deep learning models, we achieved an average accuracy of 0.83 in the hand open/closed binary classification task with the conformer model. Compared to the available work, our approach demonstrated a remarkable 16% increase in accuracy.
• HIGH-PERFORMANCE NEURAL DECODING OF 14 DUTCH KEYWORDSOffenberg, Elena Charlotte; Berezutskaya, Julia; Freudenburg, Zachary V.; Ramsey, Nick; 10.3217/978-3-99161-014-4-011Brain-computer interfaces (BCIs) can help people with locked-in syndrome to communicate. While continuous speech decoding can be used in everyday communication, navigating a computer menu or interacting with external devices may be easier and more reliable using a small set of distinct command keywords. In this preliminary study, two able-bodied epilepsy patients, temporarily implanted with high-density electrocorticography (ECoG) electrodes, spoke 14 potential keywords out loud in Dutch. With optimized Support Vector Machines (SVMs), the maximal decoding accuracy reached was a median of 93.3% for 50 repetitions per word (practical chance level 9.6%). We also identified that a minimum of 30 repetitions was needed to achieve this result, and determined that the most relevant electrodes for decoding were on the ventral sensorimotor cortex, close to the central sulcus.
• PROCESSING OF INCONGRUENT INFORMATION CAN BE DECODED FROM SINGLE-TRIAL EEG: AN AR-STUDYWimmer, Michael; Pepicelli, Alex; Volmer, Ben; Elsayed, Neven; Cunningham, Andrew; Thomas, Bruce H.; Veas, Eduardo E.; Müller-Putz, Gernot R.; 10.3217/978-3-99161-014-4-012Augmented reality (AR) allows users to display additional digital information about their physical environment. We present an interactive AR study, in which participants manipulated a Rubik’s cube which served as a physical referent for presented digital infor- mation showing the current status of the cube. In 30% of the instances, the presented information did not match its status. We recorded the electroencephalographic data of 19 participants to study their responses to incongruent stimuli and assessed if they could be classified on a single-trial level. We found that the processing of incongruent data in AR elicits both N400 and P600 components. Further, we could classify them in 15 out of 19 participants with accuracies above chance. These results contribute to the design of brain-computer interfaces, as the decoding of such correlates could inform the system about the current mental context of the user.
• AUTO-ADAPTATION OF ECOG-BASED MOTOR BCI USING NEURAL RESPONSE DECODER: A CROSS-PATIENT STUDYLafaye de Micheaux, Hugo; Martel, Félix; Sauter-Starace, Fabien; Charvet, Guillaume; Lorach, Henri; Aksenova, Tetiana; 10.3217/978-3-99161-014-4-013Motor imagery brain-computer interfaces (BCIs) face challenges in practical application, notably in decoder training. Traditionally, decoders are trained in a supervised manner. This approach requires labeled data and restricts users to predefined actions during the training period. Moreover, regular decoder updates are needed. To address these issues, the auto-adaptive BCI (aBCI) infers training labels directly from brain signals using a neural response (NR) decoder, eliminating the need for supervised sessions. This study investigates the performance and replicability of the aBCI and explores labeling strategies using electrocorticography data from three spinal cord injured patients across diverse paradigms. Results demonstrate that aBCI can be used to significantly increase decoding performance above chance level in all three patients. Performance depended on patients and labeling strategy. The labeling strategy, focusing on correct neural responses (CNR), demonstrates significantly improved performance compared to correct/error neural responses (CENR) labeling strategy. Despite limitations of pseudo-online simulation, our findings underscore the aBCI's promise in advancing BCI technology.
• CORRECTING TRAJECTORY-DECODING ERRORS VIA CORTICAL SUBSTRATES OF CONTINUOUS ERRONEOUS FEEDBACK PROCESSINGPulferer, Hannah S.; Kostoglou, Kyriaki; Srisrisawang, Nitikorn; Müller-Putz, Gernot R.; 10.3217/978-3-99161-014-4-014Decades of research thoroughly established various neural correlates of processing discrete errors, i.e., events that may be classified as either correct or wrong. However, despite many successful demonstrations of brain-computer interfaces (BCIs) utilizing these discrete correlates, a range of everyday tasks (e.g., car driving) requires fine-tuned feedback control that already transgresses such coarse distinction. Following up on recent research in the field of continuous erroneous feedback processing, we propose the regression of continuous feedback-target deviations from the electroencephalogram (EEG). Within thirty prerecorded sessions of data in ten participants, employing a 2D targettracking task that offered online feedback, we thus utilized a convolutional neural network to infer ongoing feedback-target deviations and correct the feedback’s position accordingly in an offline evaluation. The presented correction approach significantly improved correlations between feedback and target kinematics - a first indication that continuous error-related cortical activity can be utilized in BCIs as well.
• PREDICTORS OF ECOG-BCI PERFORMANCES ACROSS SUBJECTS AND SESSIONS DERIVED FROM IDLE STATE CHARACTERISTICSStruber, Lucas; Martel, Félix; Karakas, Serpil; Juillard, Violaine; Bellicha, Angelina; Sauter, Fabien; Chabardès, Stéphan; Lorach, Henri; Charvet, Guillaume; Aksenova, Tetiana; 10.3217/978-3-99161-014-4-015Comprehension of performance variabilities across subjects and sessions is crucial for real life brain-computer-interfaces (BCI) applications. This study compared three subjects that underwent implantation of minimally invasive WIMAGINE ECoG recording implants. Three training strategies to discern best achievable performance, session drift, and variability were evaluated offline using datasets recorded during real-time closed-loop BCI experiments. Results revealed distinct BCI profiles across patients, consistent with qualitative observations made during online training. These performances were correlated with two indicators computed in feature space during idle periods of BCI sessions: Euclidean distance between the current session and the session of model creation in a low- dimensional UMAP embedding, and intrinsic dimension. Between sessions distances demonstrated statistically significant correlation with models’ performances, then recalibration need may be potentially anticipated from the characteristics of idle state periods. Additionally, the intrinsic dimension was significantly correlated to subjects' overall BCI capabilities. The results are consistent with pre-implantation MEG-BCI experiments, which could make it useful for patient selection.
• UNCERTAINTY QUANTIFICATION FOR CROSS-SUBJECT MOTOR IMAGERY CLASSIFICATIONManivannan, Prithviraj; de Jong, Ivo Pascal; Valdenegro-Toro, Matias; Sburlea, Andreea Ioana; 10.3217/978-3-99161-014-4-016Uncertainty Quantification aims to determine when the prediction from a Machine Learning model is likely to be wrong. Computer Vision research has explored methods for determining epistemic uncertainty (also known as model uncertainty), which should correspond with generalisation error. These methods the- oretically allow to predict misclassifications due to intersubject variability. We applied a variety of Uncertainty Quantification methods to predict misclassifications for a Motor Imagery Brain Computer Interface. Deep Ensembles performed best, both in terms of classification performance and cross-subject Uncertainty Quantification per- formance. However, we found that standard CNNs with Softmax output performed better than some of the more advanced methods.
• TOWARDS RIEMANNIAN EEG CLASSIFIERS TO DETECT AWAKE AND ANESTHETIZED STATES USING MEDIAN NERVE STIMULATIONCueva, Valérie Marissens; Rimbert, Sébastien; Cebolla Alvarez, Ana Maria; Petieau, Mathieu; Vitkova, Viktoriya; Hashemi, Iraj; Cheron, Guy; Meistelman, Claude; Guerci, Philippe; Schmartz, Denis; Bidgoli, Seyed Javad; Bougrain, Laurent; Lotte, Fabien; 10.3217/978-3-99161-014-4-017Among all the operations carried out under general anesthesia worldwide, some patients have had the terrible experience of Accidental Awareness during General Anesthesia (AAGA), an unexpected awakening during the surgical procedure. The inability to predict and prevent AAGA before its occurrence using only con- ventional measures, such as clinical signs, leads to the use of brain activity monitors. Given AAGA patients’ first reflex to move, impeded by neuromuscular-blocking agents, we propose using a new Brain Computer Interface with Median Nerve Stimulation (MNS) to detect their movement intentions, specifically in the context of gen- eral anesthesia. Indeed, MNS induces movement-related EEG patterns, improving the detection of such intentions. In this article, we compared MNS effects on the motor cortex before and during surgery under general anesthesia. Then, a Riemannian Minimum Distance to the Mean classifier achieved 97% test balanced accuracy in distinguishing awake and anesthetized states. Additionally, we observed how the classifier’s response evolves with anesthesia depth, in terms of distance to the awake class centroid. This distance appears to track the patients’ awareness level during surgery. This holds promises for developing a future one-class classifier using only awake EEG data, as anesthesia EEG data are usually unavailable for classifier training, to detect AAGA.
• NEURONAL AVALANCHES FOR EEG-BASED MOTOR IMAGERY BCIMannino, Camilla; Sorrentino, Pierpaolo; Chavez, Mario; Corsi, Marie-Constance; 10.3217/978-3-99161-014-4-018Current features used in motor imagery-based Brain- Computer Interfaces (BCI) rely on local measurements that miss the interactions among brain areas. Such interactions can manifest as bursts of activations, called neuronal avalanches. To track their spreading, we used the avalanche transition matrix (ATM), which contains the probability that an avalanche would consecutively recruit any two brain regions. Here, we proposed to use ATMs as a potential alternative feature. We compared the classification performance resulting from ATMs to a benchmark model based on Common Spatial Patterns. In both sensor-and source-spaces, our pipeline yielded an improvement of the classification performance associated with reduced inter-subject variability. A correspondence between the selected features with the elements of the ATMs that showed a significant condition effect led to higher classification performance, which speaks to the interpretability of our findings. In conclusion, working in the sensor space provides enough spatial resolution to classify. However the source space is crucial to precisely assess the involvement of individual regions.
• LOCALIZING NEURAL SOURCES IN THE CERVICAL SPINAL CORDOberndorfer, Markus E.; Müller-Putz, Gernot R.; 10.3217/978-3-99161-014-4-019Mapping neural activity along the spinal cord is a task that is hardly researched compared to human brain mapping. By identifying neural sources in the spinal cord and detecting unique activity patterns associated with various motor tasks or specific sensory input, it becomes possible to establish a baseline for healthy individuals. This could be utilized to classify spinal cord injuries or monitor changes in the spinal cord. This study demonstrates the effective application of an innovative approach to localizing the spinal sources of spinal cord potentials (SCPs) using the finite element method (FEM) to solve the forward problem and an abstraction of the sLORETA algorithm to identify the neural sources, which were induced by functional electrical stimulation (FES) on the forearms of healthy individuals.
• A SMALL STEP TOWARDS THE DETECTION OF MENTAL FATIGUE INDUCED BY BCI-HMD TRAININGPolyanskaya, Arina; Rosipal, Roman; Sobolová, Gabriela; Rošt’áková, Zuzana; Porubcová, Natália; 10.3217/978-3-99161-014-4-020We used a proprietary constructed brain- computer interface system with a head-mounted display for motor neurorehabilitation training of a subject after a stroke. This study analyzes quantitative EEG (qEEG) changes during resting state periods before and after the neurorehabilitation training. Eyes closed and eyes open resting state EEG collected during 13 training sessions is analyzed to determine qEEG changes indicating mental state changes like increased mental fatigue, tiredness, or sleepiness. We decomposed the EEG spectrum into oscillatory and fractal parts, allowing us to investigate changes in the oscillatory component of qEEG separately. We observed increased post-training oscillatory EEG amplitudes in slow frequency bands (delta and theta) and decreased in faster alpha to beta bands. A shift to a slower frequency of the dominant alpha frequency was also observed in the post-training resting state EEG. Compared with existing literature, these changes indicate increased mental fatigue and sleepiness.
• RECORDING THE SSSEP WITH THE CEEGRIDPetit, Jimmy; Eidel, Matthias; Rouillard, José; Kübler, Andrea; 10.3217/978-3-99161-014-4-021The feasibility of EEG systems in realworld scenarios, particularly as assistive devices for peo- ple with impairments, remains limited by practical issues of conventional cap EEG. However, the emergence of the cEEGrid, an unobtrusive around-the-ear EEG system, might offer a solution. While the cEEGrid has demonstrated success in measuring event-related potentials, essential for brain-computer interfaces (BCIs) in a variety of settings, its ability to measure steady-state somatosensory evoked potentials (SSSEPs) remains unex- plored. Here, we recorded SSSEPs from seven stimulation frequencies in six participants. To allow for a direct comparison, the signal was recorded from a conven- tional scalp EEG (Brain Products Acticap) and two cEEGrids under the same conditions. Results indicate significant SSSEP elicitation with the Acticap, whereas this was only found for one participant with the cEEGrid. Amplitudes measured with cEEGrids are generally smaller, however, their relative discreet design make them an interesting alternative. Further exploration is necessary to characterise the capabilities of the cEEGrid in a potential SSSEP-based BCI application.
• INVESTIGATING COORDINATES REPRESENTATION DURING REACHING VIA LOW-FREQUENCY EEG: A PRELIMINARY STUDYSrisrisawang, Nitikorn; Müller-Putz, Gernot R.; 10.3217/978-3-99161-014-4-022Understanding how the brain plans reaching movements is crucial in designing a brain- computer interface (BCI) system for motor control. It is still unclear which referencing frame the brain uses to plan the movement. In this study, we investigated the global representation of a referencing frame during reaching planning via a low-frequency electroencephalogram (EEG). Participants were asked to perform directional reaching inward (from the outer target towards the center point) and outward (from the center point towards the outer target) while maintaining gaze on a target such that the reaching in inward and outward conditions should be represented similarly in eye-centered coordinates but differently in shouldercentered coordinates. We could classify the direction with a peak accuracy of 40.59% but not the inward and outward conditions. The preliminary results confirmed that low-frequency EEG may be globally represented in the eye-coordinates. The classification results suggested that the difference between inward-outward conditions was negligible in low-frequency EEG and could be combined in further analysis.
• IDENTIFYING NEW FEATURES FOR BCI CONTROL: SPECTRAL CHANGES IN THE MOTOR THALAMUS REVEAL HAND REPRESENTATION DURING OVERT AND IMAGINED MOVEMENTBaker, Matthew R.; Klassen, Bryan T.; Jensen, Michael A.; Ojeda Valencia, Gabriela; Banks, Samantha A.; Miller, Kai J.; 10.3217/978-3-99161-014-4-023This study explores the potential of the ventral intermediate nucleus (VIM) of the thalamus as a subcortical signal source for brain-computer interfaces (BCIs). We analyzed spectral changes in the VIM for overt and imagined hand movements during deep brain stimulation (DBS) lead implantation surgery. During task periods, we found suppression of power in the stereotypical beta range (13-30 Hz). Only in one recording site did we find a significant increase in broadband power (65- 115 Hz) with overt hand movement, but not for imagined movement. We provide evidence that motor representation in the VIM could act as a subcortical control signal for future BCI applications.
• FEASIBILITY OF STEREO EEG BASED BRAIN COMPUTER INTERFACING IN AN ADULT AND PEDIATRIC COHORTJensen, Michael A.; Schalk, Gerwin; Ince, Nuri; Hermes, Dora; Brunner, Peter; Miller, Kai J.; 10.3217/978-3-99161-014-4-024Introduction: Stereoelectroencephalography (sEEG) is a mesoscale intracranial monitoring method which records from the brain volumetrically with depth electrodes. Implementation of sEEG in BCI has not been well-described across a diverse patient cohort. Methods: Across eighteen subjects, channels with high frequency broadband (HFB, 65-115Hz) power increases during hand, tongue, or foot movements during a motor screening task were provided real-time feedback based on these HFB power changes to control a cursor on a screen. Results: Seventeen subjects established successful con- trol of the overt motor BCI, but only nine were able to control imagery BCI with ≥ 80% accuracy. In successful imagery BCI, HFB power in the two target conditions separated into distinct subpopulations, which appear to engage unique subnetworks of the motor cortex compared to cued movement or imagery alone. Conclusion: sEEG-based motor BCI utilizing overt movement and kinesthetic imagery is robust across patient ages and cortical regions with substantial differences in learning proficiency between real or imagined movement.
• DETECTION OF MOTION TERMINATION FROM EEG DURING THE EXECUTION OF CONTINUOUS HAND MOVEMENTCrell, Markus; Müller-Putz, Gernot R.; 10.3217/978-3-99161-014-4-025Recent advances in the decoding of hand kinematics from neural data and the usage for the control of cursors also prompt the need to detect the begin and end of continuous movements. This study investigates the asynchronous detection of the termination of a con- tinuous hand movement in a handwriting task using electroencephalography data and the power of frequencies in the μ and β band. Results obtained with a shrinkage linear discriminant analysis classifier yield a correct deter- mination of the offset in 53.5% (chance level: ≈ 18%) of the trials. We show the general feasibility of the proposed method in the detection of the termination of a continuous hand movement and visualize the benefit of the information of the moment of movement termination in a simulated application.
• AN EMG-BASED BRAIN-COMPUTER INTERFACE FOR COMMUNICATION-IMPAIRED PATIENTS: A CASE STUDYRaggam, Philipp; Eder, Manuel; Popa, Alexia-Theodora; Fugger, Peter; Grosse-Wentrup, Moritz; 10.3217/978-3-99161-014-4-026Electromyography (EMG)–based braincomputer interface (BCI) systems primarily rely on electrical signals generated by muscle activity instead of the typically used brain activity measured via electroencephalography (EEG). Such EMG-BCIs are promising systems that enhance communication and control. This study introduces a simple EMG-BCI communication system developed as a football game for a communicationimpaired participant. The football in the game can be moved to a left-side or a right-side goal, representing answers to two-state queries, i.e., yes-or-no-questions. By using restricted game controls, correctly following verbal instructions, and showing movement-related brain activity preceding muscle contractions, our participant can deliberately control the directions of the ball movements and, thus, successfully use our game for communication.
• FINDING THE OPTIMAL SIX: DECODING FROM A LARGE SET OF HAND GESTURES WITH 7T FMRI FOR IMPROVED BCI CONTROLKromm, Maria; Schellander, Sophia; Branco, Mariana Pedroso; Raemaekers, Mathijs A.H.L.L.; Ramsey, Nick F.; 10.3217/978-3-99161-014-4-027Decoding movements from the human sensorimotor cortex has been of great interest for brain-computer interfaces (BCIs). To establish the possibility of increasing the degrees of freedom of a sensorimotor-driven BCI, we investigated the decod- ability of 20 hand gestures using 7-Tesla fMRI and narrowed it down to a set of six best distinguishable gestures. Six able-bodied volunteers performed ges- tures from the American Sign Language alphabet and single-digit movements. Results indicated significant classification accuracies across all 20 gestures (mean = 46%, range = 39.5% − 51.5%, chancelevel = 5%), with some differences in decodability across gestures. Subsequently, optimal sets of six gestures were identified by establishing classification performance for all possible permutations, and applying the identified set in a leave- one-subject-out cross-validation scheme. The results showed a near-optimal classification in five out of six subjects. Our findings contribute to the understanding of the generalizability of gesture decoding performance and offer insights for refining BCI control strategies to enhance communication for individuals with motor impairments.
• TRANSFERRING BCI MODELS FROM CALIBRATION TO CONTROL: OBSERVING SHIFTS IN EEG FEATURESde Jong, Ivo Pascal; van den Wittenboer, Lüke Luna; Valdenegro-Toro, Matias; Sburlea, Andreea Ioana; 10.3217/978-3-99161-014-4-028Public Motor Control-based brain- computer interface (BCI) datasets are being used to develop increasingly good classifiers. However, they usually follow discrete paradigms where participants perform Motor Imagery, Attempts or Execution at reg- ularly timed intervals. It is often unclear what changes may happen in the EEG patterns when users attempt to perform a control task with such a BCI. This may lead to generalisation errors. We demonstrate a new paradigm containing a standard calibration session and a novel BCI control session based on EMG. This allows us to observe similarities in sensorimotor rhythms, and observe the additional preparation effects introduced by the control paradigm. In the Movement Related Cortical Potentials we found large differences between the calibration and control sessions. We demonstrate a CSP-based Machine Learning model trained on the calibration data that can make surprisingly good predictions on the BCI-controlled driving data.
• RESTING-STATE BRAIN CRITICALITY AND PERFORMANCE WITH P300-BASED BCISSettgast, Tomko; Kübler, Andrea; 10.3217/978-3-99161-014-4-029Here we present correlations between criticality-related measures calculated from resting-state electroencephalography (EEG) recordings and subsequent performance with a visual P300-based brain- computer interface (BCI) in healthy participants. Results suggest a positive relationship between resting-state brain criticality and subsequent BCI performance using P300-based BCIs.
• BIDIRECTIONAL NEUROFEEDBACK: A CONTROL CONDITION COMPLEMENTARY TO SHAM?Pierrieau, Emeline; Pillette, Lea; Dussard, Claire; George, Nathalie; Jeunet-Kelway, Camille; 10.3217/978-3-99161-014-4-030Neurofeedback (NF) is increasingly used for experimental and therapeutic purposes. However, the lack of proper control about the specificity of NF effects is criticized and hinders the development of reliable and efficient NF procedures. Bidirectional NF is based on the self-regulation of the targeted brain activity in opposite directions and might be better suited than other typical control conditions (e.g., sham) for assessing the link between modulations of brain activity and behavior. The present study aimed to determine if bidirectional regulation of a specific pattern of brain activity, namely motor beta power, can be achieved within a single session. Thirty participants performed several NF trials aiming to either down- or up-regulate their motor beta power. Results showed that participants significantly modulated their motor beta power in opposite directions with bidirectional NF. This modulation was constrained in space (central electrodes) and frequency (al- pha/beta band). Overall, bidirectional NF appears as a valid method to probe brain-behavior relationships within a single session.
• DYNAMIC BRAIN NETWORKS IN MOTOR IMAGERY-BASED BCIVenot, Tristan; Desbois, Arthur; De Vico Fallani, Fabrizio; 10.3217/978-3-99161-014-4-031Using the interactions between brain regions has great potential as new features to discriminate between mental tasks for brain computer interface (BCI). Network approaches applied to electroencephalographic (EEG)-derived functional connectivity has been recently used to identify discriminating brain organizational features in offline classification scenarios. However how those network properties temporally vary during the task, is still poorly understood. A contrario, the dynamics of event related desynchronization/synchronization resulting from local power spectra is widely known and used for online motor imagery-based BCIs. Here, we explored the offline time-frequency properties of dynamic brain networks in two subjects performing three sessions of MI-BCI for the control of a robotic arm. Results were compared to standard time-frequency power spectra and discussed in light of future implementation for online sce- narios.
• WHICH FACTORS AFFECT THE ACCEPTABILITY OF BCIS FOR FUNCTIONAL REHABILITATION AFTER STROKE AMONG PATIENTS?Grevet, Elise; Izac, Margaux; Amadieu, Franck; Py, Jacques; Gasq, David; Jeunet-Kelway, Camille; 10.3217/978-3-99161-014-4-032Although motor imagery-based BCIs have been demonstrated to be relevant for improving motor recovery after stroke, they remain barely used in rehabilitation services. We hypohesise that acceptability (which is assessed in terms of perceived usefulness, ease of use and intention to use) could serve as a lever for fostering the adoption of BCIs through the improvement of their efficacy. More precisely, we suggest that improving the acceptability of BCIs could alleviate post-stroke patients’ anxiety, stimulate their motivation and engagement in the BCI process, and thereby, favour skill acquisition (here self-regulation abilities), which will ultimately have positive effects on motor recovery. We created a model of acceptability of BCIs specifically for functional rehabilitation after stroke, and designed an associated questionnaire that was used to empirically assess the weight each factor of the model had on acceptability. Hereinafter, we introduce the methods and results obtained based on the responses received from 140 patients, and compare them with data collected in the general public (N=753). In a nutshell, for both the general public and patients perceived usefulness, scientific relevance and ease of learning emerge as the most influential factors.
• BREAKING OUT OF THE FEEDBACK LOOP: TRANSFERRING MASTERY OF SELF-REGULATION DURING NEUROFEEDBACK TO OTHER CONTEXTSKober, Silvia Erika; Wood, Guilherme; 10.3217/978-3-99161-014-4-033One important question in neurofeedback (NF) research is the mastery of self-regulation and the generalizability of the NF training results. Here, we investigated whether NF users can voluntarily increase the Sensorimotor Rhythm (SMR, 12-15 Hz) activity during repeated NF training sessions while receiving visual feedback and if they can also increase SMR during subsequent transfer sessions without any feedback. We also assessed the used mental strategies during the sessions. Nine healthy adults received real feedback, nine received sham feedback. Only the real feedback group was able to linearly increase SMR within the six NF training sessions. However, they could not increase SMR during the transfer sessions. Participants reported multiple different mental strategies during NF training as well as during transfer sessions with different success rates. These results indicate that participants were not able to transfer successful mental strategies to other situations after six sessions of SMR-based NF training.
• INTEGRATING CORTEC BRAIN INTERCHANGE DEVICE AND BCI2000 WITH A CLOUD INTERFACEMivalt, Filip; van den Boom, Max A.; Lampert, Frederik; Kim, Jiwon; Duque Lopez, Andrea; Engelhardt, Will; Kim, Inyong; Chang, Su-Youne; Hermes, Dora; Brunner, Peter; Kremen, Vaclav; Ince, Nuri; Schalk, Gerwin; Worrell, Gregory A.; Miller, Kai J.; 10.3217/978-3-99161-014-4-034Emerging brain-computer interface (BCI) systems may aim to develop invasive implantable systems to restore functionality in people with paralytic disabilities and to deliver adaptive brain stimulation (ABS) to treat severe neurological disorders. A key characteristic of next-generation implantable systems will be their capability to record extended periods of local field potential (LFP) data. Timely transfer of the recorded LFPs to the clinical team is crucial for monitoring the implanted system’s reliability, safety and to dynamically enhance BCI and ABS applications in response to changing brain states. Our team is developing a comprehensive therapeutic BCI ecosystem that combines the Cortec BrainInterchange hardware with the BCI2000 software environment. We have designed an architecture that seamlessly integrates recorded neural signals with device performance metrics, delivering these insights to the care team through a cloud-based interface. In order for future centers-of-excellence to be able to deliver care with clinical BCIs, closed-loop algorithms will need be able to be dynamically updated without physically interacting with the patient for each adjustment. Our BCI ecosystem is currently being tested with canine subjects, and this manuscript describes how device function (impedance measures) and brain data (LFP signals) were measured daily for an 8 week period following implantation through the cloud interface. Cloud based data synchronization for implantable brain technologies is essential for dynamic re-calibration of reliable and safe BCI and ABS therapies in the clinical setting.
• SPATIAL AND SPECTRAL CHANGES IN CORTICAL POTENTIALS DURING PINCHING VERSUS THUMB AND INDEX FINGER FLEXIONKerezoudis, Panagiotis; Jensen, Michael A.; Huang, Harvey; Ojemann, Jeffrey G.; Klassen, Bryan T.; Ince, Nuri F.; Hermes, Dora; Miller, Kai J.; 10.3217/978-3-99161-014-4-035We analyzed the electrocorticographic signals of 3 patients implanted with subdural electrode arrays for identification of seizure foci. Patients performed screen cue-based flexion movement of the thumb or index finger, or a pinch movement of both. Broadband power changes with each movement were estimated. Topological maps for each type of movement were created using co-registered brain renderings, and the overlap in spatial extent was quantified using a resampling metric. Activation during pinching was compared with thumb flex, index flex and three composite metrics: the maximum, the geometric mean and a modified geometric mean. Significant increase in broadband power was observed in all three patients when pinch (max signed r2 0.64-0.80-0.89), thumb flexion (0.63-0.72-0.89) or index flexion (0.66-0.67- 0.83) was performed compared to rest. Spatial overlap was highest between pinching and index flexion (69- 87%, all p< .001), followed by the modified geometric mean (61-96%, all p< .001), while thumb flexion had the lowest overlap of all analyzed metrics (27-77%, significant in 2/3).
• PASSIVE OLFACTORY BRAIN-COMPUTER INTERFACE PARADIGM FOR AWARENESS LEVEL PREDICTIONRutkowski, Tomasz M.; Kasprzak, Hubert; Niewińska, Nina; Otake-Matsuura, Mihoko; Komendziński, Tomasz; 10.3217/978-3-99161-014-4-036The sense of smell, also known as olfaction, can improve the usability of brain-computer interfaces (BCIs) and support passive modalities for monitoring cognitive states. In reactive BCI, users can assign specific scents to commands for natural interaction, while a passive application can monitor cognition. However, some challenges still need to be addressed, such as the need for accurate odor delivery systems and robust algorithms for detecting and interpreting brain activity patterns. We propose combining electroencephalogram (EEG) and electrobulbogram (EBG) in an olfactory modality oddball paradigm to predict a user’s awareness level. Our pilot study indicates promising results for a new passive olfactory BCI modality combining CSP filtration and awareness level classification.
• SSVEP-BASED COVERT COMMUNICATION USING HYPERSCANNINGReintsema, Lars H.; Sweeney-Reed, Catherine M.; Dürschmid, Stefan; Hinrichs, Hermann; Reichert, Christoph; 10.3217/978-3-99161-014-4-037Communication by means of evoked brain signals is one of the main applications of brain-computer interfaces (BCIs). Commonly, in BCI applications the user’s intention is directly fed back and openly perceivable. Here we used hyperscanning to investigate a communication approach, in which two users can covertly communicate by brain signal modulation. To achieve this, we artificially generated synchronous and asynchronous oscillatory brain activity by presenting a choice of two flickering stimuli inducing steady-state visual evoked potentials (SSVEPs) and provided feedback that indicated the synchronicity of the brain signals of participant pairs. We used different approaches to determine synchronicity. When we used broadband activity, the accuracy varied considerably between participant pairs, which could be attributed to individual differences in the timing and the amplitudes of SSVEPs. However, when we involved features reflecting the stimulus frequencies, the predictions were highly reliable. Beyond demonstrating the feasibility of our approach, our findings have the potential to identify challenges in studying social interaction using hyperscanning.
• EEG-BASED PERFORMANCE ESTIMATION DURING A REALISTIC DRONE PILOTING TASKHinss, Marcel F.; Vitale, Vincenzo Maria; Brock, Anke M.; Roy, Raphaëlle N.; 10.3217/978-3-99161-014-4-038Passive brain-computer interfaces (pBCIs) developed within the neuroergonomic field usually aim to improve safety by augmenting human-machine interaction. To accomplish said goal, many pBCIs classify mental states such as mental workload or mental fatigue. An alternative is to forego mental states and aim to predict performance. Despite its drawbacks, we argue that performance estimation is a more goal-oriented approach than mental state estimation. In a realistic experiment, 25 participants had to control an uncrewed aerial system for two hours, continuously switching between target search and navigation. EEG classification accuracies based on mental states and performance were compared. With a Tangent Space Logistic Regression, we could predict an increased likelihood of lapses in the form of missing instructions with an above-chance level accuracy of 62.09 %.
• LESS IS MORE: ADVANCING EEG-BASED ONLINE CONTINUOUS MACHINE ERROR DETECTION WITH THE LIGHTWEIGHT MAX-MIN AMPLITUDE NOISE FILTERING TECHNIQUEPan, Yanzhao; Rabe, Lea; Klug, Marius; 10.3217/978-3-99161-014-4-039To apply synchronous laboratory passive brain-computer interface (pBCI) systems to dynamic real-world scenarios, it is essential to develop asynchronous, event-independent pBCIs that can continuously interpret brain activity. Minimizing false alarms (FAs) caused by artifacts in continuous online sessions without compromising the hit rate is one of the primary challenges in EEG-based brain activity classification tasks. To address this challenge, this study introduces the Max-Min Amplitude Noise Filtering (MANF) technique, which is designed to reduce FAs in the online EEG-based machine error detection task. To achieve this, we pre-trained a classifier on labeled data and then tested the performance of the technique on a simulated continuous online classification. The MANF technique, using a predetermined noise threshold, simplifies the noise filtering process by comparing the difference between the maximal and minimal amplitude of incoming EEG data against this threshold, substantially reducing FAs while maintaining high hit rate. This technique outperforms the unfiltered condition and those using the Artifact Subspace Reconstruction technique, achieving an optimal balance between sensitivity and specificity with medium and conservative thresholds. Highlighting the "less is more" principle, the MANF technique proves highly suitable for continuous online pBCI applications. This development contributes to the ongoing efforts in creating more user-friendly and reliable pBCIs for dynamic real-world use.
• ANA-E: A NOVEL APPROACH FOR PRE-TRAINED ERROR DETECTION MODELS IN BRAIN-COMPUTER INTERFACESChristopoulos, Alexandros; Valdenegro-Toro, Matias; Sburlea, Andreea Ioana; 10.3217/978-3-99161-014-4-040Error-related potentials hold the potential to enhance self-correcting behaviors in Brain-Computer Interfaces (BCIs), pivotal for human-machine interactions. However, integrating error detection mechanisms poses challenges, notably in lengthy calibration sessions required for different BCI modules. To address this, we propose a novel approach using Self-Supervised Learning (SSL) with an autoencoder architecture, called Ana- E, to develop pre-trained error detection pipelines. We recorded EEG data from participants navigating a game scenario imposed with errors. Offline analyses within and between participants were conducted for both preprocessed EEG trials and Ana-E features with two classifiers. Within-participants analysis showed comparable performance between Ana-E features and EEG trials. While in between-participants analysis, Ana-E exhibited an 8% performance improvement (72%) over the secondbest pipeline (64%). Our study offers valuable insights into the future of pre-trained models for error detection in BCIs, providing a baseline for more complex architectures with the goal of significantly enhancing BCI usability and reducing dependency on calibration sessions, thereby improving user experience and applicability.
• ASSESSING CALIBRATION DURATIONS FOR C-VEP-BASED BCIS: INSIGHTS FROM NON-BINARY PATTERNS AND SPATIAL FREQUENCY VARIATIONSMartínez-Cagigal, Víctor; Fernández-Rodríguez, Álvaro; Santamaría-Vázquez, Eduardo; Martín-Fernández, Ana; Hornero, Roberto; 10.3217/978-3-99161-014-4-041BCIs using code-modulated visual evoked potentials (c-VEP) have become popular for their reliable, high-speed control of applications and devices. However, traditional circular shifting paradigms based on black & white stimuli can cause eyestrain for some users. We previously showed that adjusting the number of code events and spatial frequency can enhance user comfort. Despite c-VEP calibration being notably shorter than other BCIs, the optimal number of calibration cycles for effective system control remains unexplored. This study aims to investigate the impact of calibration duration on various c-VEP-based BCIs, with stimulus variations to improve user experience. We evaluated performance with different calibration cycles using five pary m-sequences encoded with shades of gray and eight spatial frequency variations of checkerboard-like stimuli. Results indicate that all conditions achieved over 90% accuracy and 80 bpm with calibration durations ranging between 6–70 seconds. These findings highlight the importance of selecting a configuration based on the functional requirements of the BCI.
• IMPROVED MOTOR IMAGERY DECODING WITH SPATIOTEMPORAL FILTERING BASED ON BETA BURST KERNELSPapadopoulos, Sotirios; Darmet, Ludovic; Szul, Maciej J.; Congedo, Marco; Bonaiuto, James J.; Mattout, Jérémie; 10.3217/978-3-99161-014-4-042The description of the event-related desynchronization and synchronization phenomena in the mu and beta frequency bands has to a significant extent shaped our understanding of motor-related brain processes. Accordingly, Brain-Computer Interface applications leveraging attempted or imagined movements usually depend on spatially- and band-limited power changes as the brain markers of interest. Yet, converging neuroscience evidence question the idea that signal power best describes the movement-related modulation of brain activity. On a single-trial level, beta band activity is characterized by short, transient and heterogeneous events termed bursts rather than sustained oscillations. In a recent study we demonstrated that a beta burst analysis of hand motor imagery binary classification tasks is often superior to beta power in terms of classification score. Here we expand upon this idea proposing a comparable to state-of-the-art algorithm. We confirm our previous results by using convolution kernels extracted from beta bursts. Moreover, we show that these kernels can effectively be used in inter-session transfer learning strategies.
• EEG-BASED STIMULUS CLASSIFICATION IN A FULL-BODY MOVEMENT, VIRTUAL REALITY PARADIGMRabe, Lea; Pan, Yanzhao; Klug, Marius; 10.3217/978-3-99161-014-4-043The use of EEG brain-computer interfaces (BCI) during movement is inherently difficult due to motion artifacts interfering with measured brain signals. Thus, most BCI research utilizes rather immobile conditions, thereby decidedly limiting its range of use cases. We aim to overcome this restriction by introducing a novel virtual reality paradigm which allows full-body movement of participants in combination with a processing pipeline specifically designed to deal with motion artifacts. Stimulus discrimination (target versus distractor) upon fixation was tested in 32 participants. Results indicate that targets elicit a higher P300 amplitude than distractors. Comparing the performance of different classifiers, shrinkage linear discriminant analysis (sLDA), support vector machine (SVM), and EEGNet, yielded equally sized, above chance classification accuracies. Overall, the results suggest the feasibility of studying and applying BCI in full-body motion paradigms given refined data preprocessing. The authors conclude with suggestions for future BCI studies in motion.
• EEG MARKERS OF ACCELERATION PERCEPTION IN VIRTUAL REALITYVan der Lee, Gaël; Cabestaing, François; Lécuyer, Anatole; Scherer, Reinhold; Si-Mohammed, Hakim; 10.3217/978-3-99161-014-4-044This study investigates neural patterns of acceleration in virtual reality (VR) using electroencephalography (EEG). Participants experienced accelerating white spheres in VR while EEG signals were recorded. Significant EEG differences were found at the fronto-central region between acceleration and slow speed, regardless of direction, and at the central region depending on the acceleration direction. Topographic responses also show differences in spacial patterns between the conditions. These findings give insights into the perception of acceleration in the brain and show potential for passive BCI applications.
• EYE-TRACKING AND SKIN CONDUCTANCE TO MONITOR TASK ENGAGEMENT DURING NEUROFEEDBACK SESSIONSFragueiro, Agustina; Debroize, Rene-Paul; Bannier, Elise; Cury, Claire; 10.3217/978-3-99161-014-4-045The neurofeedback (NF) inefficacy problem refers to the variability in NF success and has been associated with attentional and motivational factors. Sustaining attention on any task over an extended period is demanding and leads to attentional drops. By using eyetracking and skin conductance, we aimed at extracting physiological features linked to cognitive work, with the further purpose of monitoring changes in task engagement during NF sessions. Here, we present preliminary results on pupil diameter (PD) and phasic skin conductance responses (ISCR) linked to cognitive task execution. We observed that changes in both features are associated with performance and time-on-task. Thus, PD and ISCR decreased along the task while the performance increased. However, this trend is affected by manipulation of the task difficulty level. We also monitored, in the same participants, PD and ISCR during one NF session. Finally, we discussed preliminary ideas for target adaptation during NF sessions based on eye-tracking and skin conductance monitoring.
• RELIABILITY OF INDIVIDUAL TASK-RELATED FRONTAL-MIDLINE-THETA FREQUENCY FOR NEUROFEEDBACK TRAINING –EXPLORATORY INVESTIGATIONSPfeiffer, Maria; Masson, Eva; Kübler, Andrea; Rodrigues, Johannes; 10.3217/978-3-99161-014-4-046Neurofeedback (NF) is a technique where participants receive real-time feedback about their brain activity to learn how to modulate it. As a non-invasive neuromodulation tool, it proves useful in both research and clinical practice. However, approximately one third of users do not respond effectively to NF, prompting efforts to improve responder rates. A promising approach involves individualizing feedback by focusing on a narrow feedback band that encompasses only the individual's peak frequency (IPF), as opposed to a fixed broadband. In some frontal-midline-theta (FMT) - NF paradigms, the IPF is determined during a single calibration session and applied over several days. In a pilot study involving five participants undergoing seven sessions of FMT-NF, we calibrated the IPF using a virtual TMaze task and conducted two follow-up sessions. Our exploratory analysis across three task sessions failed to detect a stable IPF. This, as well as the scarce literature on FMT peak frequency stability, casts first doubts on the efficacy of this calibration technique.
• MOTOR IMAGERY VIVIDNESS AND NATURALISTIC INNER SPEECH HABITS IN IMAGINED SPEECH CLASSIFICATIONHons, Manuel; Kober, Silvia Erika; Wriessnegger, Selina Christin; Wood, Guilherme; 10.3217/978-3-99161-014-4-047Research on BCI-illiteracy in the imagined speech domain has been scarce. In the current study, we therefore investigate the relationships between both motor imagery vividness as well as inner speech habits, and classification accuracy based on the neural activity evoked by speech imagination. For this purpose, we classified electroencephalography-derived brain activity with respect to four imaginatively spoken phonemes: /a/, /i/, /b/ and /k/. We found that individuals who engaged more frequently in dialogic inner speech exhibited significantly higher classification accuracies, while motor imagery vividness showed no effects. Neurophysiological findings indicate that a higher expression of dialogic inner speech is associated with a suppression of redundant or counteractive neural information. These findings extend our understanding of the substrates of classification performance, respectively, BCI-illiteracy in speech imagery-based systems.
• REVIRE: A VIRTUAL REALITY PLATFORM FOR BCI-BASED MOTOR REHABILITATIONMihić Zidar, Lucija; Raggam, Philipp; Mohammadian, Farhad; Barłoga, Aneta; Grosse-Wentrup, Moritz; 10.3217/978-3-99161-014-4-048We introduce REVIRE (REhabilitation in VIrtual REality), an immersive virtual reality platform for post-stroke upper limb rehabilitation with integrated EEG recording. REVIRE immerses users in a 3D virtual environment where they can practice motor tasks that reflect everyday activities while providing comprehensive performance data with synchronized hand trajectories and EEG signals. Our proof-of-concept study tested the application on four healthy individuals across multiple training sessions. We observed significant effects of training on performance, evidenced by reduced task completion times. Changes in performance coincided with a decrease in EEG sensorimotor activity, consistent with existing motor learning research. In addition, the low incidence of cybersickness reported by participants indicates a comfortable and user-friendly experience, making our setup suitable for patient use. Our preliminary findings demonstrate the suitability of our virtual reality platform for BCI-based motor rehabilitation for clinical environments and beyond.
• WHICH IMAGINED SENSATIONS MOSTLY IMPACT ELECTROPHYSIOLOGICAL ACTIVITY?Savalle, Emile; Le Jeune, François; Driessens, Léa; Macé, Marc J-M.; Pillette, Léa; 10.3217/978-3-99161-014-4-049Motor imagery brain-computer interfaces (MI-BCI) user training aims at teaching people to control their sensorimotor cortex activity using feedback on the latter, often acquired using electroencephalography (EEG). During training, people are mostly asked to focus their imagery on the sensations associated with a movement, though very little is known on the sensations that mostly favor sensorimotor cortex activity. Our goal was to assess the influence of imagining different sensations on EEG data. Thirty participants performed MI tasks involving the following sensations: (i) interoceptive, arising from the muscles, tendons, and joints, (ii) exteroceptive, arising from the skin, such as thermal sensations, or (iii) both interoceptive and exteroceptive. The results indicate that imagining exteroceptive sensations generates a greater neurophysiological response than imagining interoceptive sensations or both. Imagining external sensations should thus not be neglected in the instructions provided during MI-BCI user training. Our results also confirm the negative influence of mental workload and use of visual imagery on the resulting neurophysiological activity.
• ONLINE DETECTION OF EPILEPTIC SPIKES FOR USE IN EPILEPSY MONITORINGMohammadpour, Mostafa; Kapeller, Christoph; Kamada, Kyosuke; Scharinger, Josef; Schwarzgruber, Michael; Korostenskaja, Milena; Guger, Christoph; 10.3217/978-3-99161-014-4-050Epileptic spikes, indicative of the seizure onset zone (SOZ), provide meaningful insight for neurosurgeons looking to find seizure locations, particularly during intraoperative procedures. Many algorithms have been proposed to detect epileptic spikes, primarily based on offline data analysis. However, none of these algorithms have been successfully adapted for online applications. In this study, we introduce a novel method for online detecting epileptic spike patterns in electrocorticography (ECoG) data. This algorithm dynamically models statistical distributions of signal envelopes, which could discriminate between signals containing epileptic spikes and those showing background activity. The effectiveness of the proposed algorithm is evaluated using resting-state data from two patients. The results reveal a sensitivity of 73% and a specificity of 95% for detecting epileptic spikes online, with an overall accuracy of 93% and an f1 score of 52%. Overall, these results validate the potential of online detection as a valuable method for epilepsy monitoring and diagnosis.
• COMPARISON OF CNN-BASED EEG CLASSIFICATION IN SENSOR AND SOURCE SPACEMaurer, Magdalena; Baumgarten, Daniel; Vorwerk, Johannes; 10.3217/978-3-99161-014-4-051Electroencephalography (EEG) is a popular tool in brain-computer interfacing (BCI), due to its unique time resolution and simplicity of application. For the design of BCIs, rapid and accurate classification algorithms are needed to classify the brain state correctly in real-time. Recent technological advancements facilitate the use of novel methods for signal processing and analysis such as real-time source estimation and classification via deep learning approaches. In this work a previously established convolutional neural network (CNN) architecture, the EEGNet, was applied to a publicly available motor imagery EEG dataset for classification of sensor measurements and source estimates that were computed with three different inverse approaches. Both for sensor signals and source estimates similar classification accuracies as in the literature could be achieved. However, no significant difference in performance between sensor and source space analysis was observed.
• IMPACT OF MENTAL FATIGUE ON REGAINING MOTOR FUNCTIONALITY: A PRELIMINARY EEG STUDY ON STROKE SURVIVORSKaniska, Samanta; KongFatt, Wong-Lin; Girijesh, Prasad; Saugat, Bhattacharyya; 10.3217/978-3-99161-014-4-052In the past few decades, research has demonstrated that brain-computer interface (BCI) based neurorehabilitation for stroke survivors can enhance the re-learning of lost motor functionality better than traditional physiotherapy involving professional physiother- apists. Though BCI-aided systems have several advantages over traditional rehabilitation methods, one of the major shortcomings of such intervention is its inability to recognize the relevant motor activity of the brain when the user gets mentally fatigued, which eventually causes the deterioration of the BCI performance. In this paper, a preliminary EEG study on stroke survivors has been reported on how mental fatigue can potentially hinder the enhancement of motor re-learning and elongate the rehab process. From the study, it has been inferred that objective measurement of mental fatigue is essential to pre- vent any subjective bias, and the rehabilitation paradigm should be adaptive to the participants’ mental status to optimize the rehab outcomes.
• MAPPING NEUROMUSCULAR REPRESENTATION OF GRASPING MOVEMENTS USING ULTRA-HIGH-DENSITY EEG AND EMGSchreiner, Leonhard; Schomaker, Pauline; Sieghartsleitner, Sebastian; Schwarzgruber, Michael; Pretl, Harald; Sburlea, Andreea I.; Guger, Christoph; 10.3217/978-3-99161-014-4-053Understanding the intricate coordination between the brain and muscles during movement tasks is crucial for advancing our knowledge of motor control and enhancing Brain-Computer Interface (BCI) devices. This study investigates the mechanisms underlying grasping movements using diverse objects and grasping techniques. Employing a novel ultra-high-density (uHD) EEG/EMG system, the study examines neural and muscular activity with high spatial resolution. Results of three healthy subjects highlight event-related desynchronization/synchronization (ERD/S) patterns and classification accuracies for EEG and EMG signals during grasping tasks. Temporal analysis reveals a strong relationship between EMG/EEG activation and classification outcomes, supported by kinematic data as evidence of motion. S02 achieved the highest average EEG and EMG classification accuracies at 69.4% and 97.8%, respectively, while S01 had the lowest at 64% and 85.4%. The observed dependencies between accuracies imply an interconnected and synergistic relationship between EEG and EMG modalities, which holds promise for enhancing overall performance in future BCIs.
• DECODING MORAL JUDGEMENT FROM TEXT: A PILOT STUDYGherman, Diana E.; Zander, Thorsten O.; 10.3217/978-3-99161-014-4-054Moral judgement is a complex human reaction that engages cognitive and emotional dimensions. While some of the morality neural correlates are known, it is currently unclear if we can detect moral violation at a single-trial level. In a pilot study, here we explore the feasibility of moral judgement decoding from text stimuli with passive brain-computer interfaces. For effective moral judgement elicitation, we use video-audio affective priming prior to text stimuli presentation and attribute the text to moral agents. Our results show that further efforts are necessary to achieve reliable classification between moral congruency vs. incongruency states. We obtain good accuracy results for neutral vs. morallycharged trials. With this research, we try to pave the way towards neuroadaptive human-computer interaction and more human-compatible large language models (LLMs).
• REAL-TIME NEUROFEEDBACK ON INTER-BRAIN SYNCHRONY: CURRENT STATES AND PERSPECTIVESWon, Kyungho; Pillette, Léa; Macé, Marc J-M.; Lécuyer, Anatole; 10.3217/978-3-99161-014-4-055During neurofeedback (NFB) user training, participants learn to control the feedback associated with specific components of their brain activity, also called neuromarkers, to improve the cognitive abilities related to these neuromarkers, such as attention and mental workload. The recent development of methods to record the activity of several people’s brains simultaneously opens up the study of neuromarkers related to social interactions, computed from inter-brain synchrony (IBS). Here, we review the previous articles that trained participants to control electroencephalographic neuromarkers computed from inter-brain metrics. The topic remains relatively unexplored as we only identified seven articles in the literature. We specifically studied the characteristics of the user’s training, i.e., instruction, task and feedback, and the neuromarkers used to provide feedback. The reported results are promising as four studies including subjective measures of interaction report higher interaction and relationship scores with higher IBS during NFB training. Finally, we draw guidelines, identify open challenges, and suggests recommendations for future studies on this topic.
• TO REPEAT OR NOT TO REPEAT? ERP-BASED ASSESSMENT OF THE LEVEL OF CONSCIOUSNESS - A CASE STUDYHalder, Sebastian; Matran-Fernandez, Ana; Nawaz, Rab; Lopes da Silva, Marina; Bertoni, Tommaso; Noel, Jean-Paul; Jöhr, Jane; Serino, Andrea; Diserens, Karin; Scherer, Reinhold; Perdikis, Serafeim; 10.3217/978-3-99161-014-4-056Determination of the wakefulness and consciousness state in patients with disorders of conscious- ness (DOC) is vital for clinical decision-making. Typically, behavioral indicators and motor responses are employed. Recent advancements in neuroimaging have en- abled motor independent assessment of DOC patients. We present a single-case analysis of a 24-year-old female, selected from a sample of n=77 patients, diagnosed with a DOC. We investigated the single-trial classification of stimuli within the peri-personal space (PPS) using eventrelated potential (ERP) features. Data from two sessions, conducted ten days apart, were analysed. We observed significant differences in classification accuracies between sessions (high in session one, low in session two), which did not correspond to the patient’s recovery from UWS to MCS. ERP analyses confirmed the difference between sessions, supporting the observed changes in classification accuracies. Our study underscores the importance of longitudinal assessments to accurately diagnose DOC patients. In future research we aim to expand our analyses to the full dataset.
• EXPLORING NEW TERRITORY: CALIBRATION-FREE DECODING FOR C-VEP BCIThielen, Jordy; Sosulski, Jan; Tangermann, Michael; 10.3217/978-3-99161-014-4-057This study explores two zero-training methods aimed at enhancing the usability of brain- computer interfaces (BCIs) by eliminating the need for a calibration session. We introduce a novel method rooted in the event-related potential (ERP) domain, unsupervised mean maximization (UMM), to the fast codemodulated visual evoked potential (c-VEP) stimulus protocol. We compare UMM to the state-of-the-art c-VEP zero-training method that uses canonical correlation analysis (CCA). The comparison includes instantaneous classification and classification with cumulative learning from previously classified trials for both CCA and UMM. Our study shows the effectiveness of both methods in navigating the complexities of a c-VEP dataset, highlighting their differences and distinct strengths. This research not only provides insights into the practical implementation of calibration-free BCI methods but also paves the way for further exploration and refinement. Ultimately, the fusion of CCA and UMM holds promise for enhancing the accessibility and usability of BCI systems across various application domains and a multitude of stimulus protocols.
• MACHINE LEARNING-BASED IDENTIFICATION OF TES-TREATMENT NEUROCORRELATESArpaia, Pasquale; Ammendola, Lidia; Cropano, Maria; De Luca, Matteo; Della Calce, Anna; Gargiulo, Ludovica; Lus, Giacomo; Maffei, Luigi; Malangone, Daniela; Moccaldi, Nicola; Raimo, Simona; Signoriello, Elisabetta; De Blasiis, Paolo; 10.3217/978-3-99161-014-4-058This study presents a Machine Learning based identification of electroencephalographic (EEG) features related to transcranical Electrical Stimulation (tES) in Multiple Sclerosis (MS) patients. The contri bution is a first step toward an automated system capable of adjusting electrical stimulation according to the EEG feedback (EEG-based adaptive tES). Five MS patients underwent both tES or sham treatments and a Theory of Mind (ToM) training, and the EEG signal before and af ter treatments was acquired both in Eyes-Open (EO) and in Eyes-Closed (EC) condition. tES was administered by fixed cathode electrodes on the right deltoid muscle. Power differences between post and pre tES treatment in six bands of interest were explored. Support Vector Ma chine classifier achieved 92.5 % and 100.0 % accuracy in classifying a subject treated with tES, by exploiting power differences within high beta in T3 and gamma in T3 and P3 in EO condition and power differences within gamma in T3, Pz, Cz in EC condition, respectively. In particular, absolute power in gamma band was reduced after the treatment. The result is clinically significant due to the tendency of MS patients to have high values in this band, caused by the compensation determined by the neu rons as a result of the demyelination process.
• TOWARDS AUDITORY ATTENTION DECODING WITH NOISE-TAGGING: A PILOT STUDYScheppink, Hanneke A.; Ahmadi, Sara; Desain, Peter; Tangermann, Michael; Thielen, Jordy; 10.3217/978-3-99161-014-4-059Auditory attention decoding (AAD) aims to extract from brain activity the attended speaker amidst candidate speakers, offering promising applications for neuro-steered hearing devices and brain-computer interfacing. This pilot study makes a first step towards AAD using the noise-tagging stimulus protocol, which evokes reliable code-modulated evoked potentials, but is minimally explored in the auditory modality. Partici- pants were sequentially presented with two Dutch speech stimuli that were amplitude-modulated with a unique binary pseudo-random noise-code, effectively tagging these with additional decodable information. We compared the decoding of unmodulated audio against audio modulated with various modulation depths, and a conventional AAD method against a standard method to decode noise-codes. Our pilot study revealed higher performances for the conventional method with 70 to 100 percent modulation depths compared to unmodulated audio. The noise-code decoder did not further improve these re- sults. These fundamental insights highlight the potential of integrating noise-codes in speech to enhance auditory speaker detection when multiple speakers are presented simultaneously.
• TOWARDS GAZE-INDEPENDENT C-VEP BCI: A PILOT STUDYNarayanan, Shekhar; Ahmadi, Sara; Desain, Peter; Thielen, Jordy; 10.3217/978-3-99161-014-4-060A limitation of brain-computer interface (BCI) spellers is that they require the user to be able to move the eyes to fixate on targets. This poses an issue for users who cannot voluntarily control their eye movements, for instance, people living with late-stage amy- otrophic lateral sclerosis (ALS). This pilot study makes the first step towards a gaze-independent speller based on the code-modulated visual evoked potential (c-VEP). Participants were presented with two bi-laterally located stimuli, one of which was flashing, and were tasked to attend to one of these stimuli either by directly looking at the stimuli (overt condition) or by using spatial attention, eliminating the need for eye movement (covert condition). The attended stimuli were decoded from elec- troencephalography (EEG) and classification accuracies of 88 % and 100 % were obtained for the covert and overt conditions, respectively. These fundamental insights show the promising feasibility of utilizing the cVEP protocol for gaze-independent BCIs that use covert spatial attention when both stimuli flash simultaneously.
• APPROXIMATE UMAP ALLOWS FOR HIGH-RATE ONLINE VISUALIZATION OF HIGH-DIMENSIONAL DATA STREAMSWassenaar, Peter; Guetschel, Pierre; Tangermann, Michael; 10.3217/978-3-99161-014-4-061In the BCI field, introspection and interpretation of brain signals are desired for providing feedback or to guide rapid paradigm prototyping but are challenging due to the high noise level and dimensionality of the signals. Deep neural networks are often introspected by transforming their learned feature representations into 2- or 3-dimensional subspace visualizations using projection algorithms like Uniform Manifold Approximation and Projection (UMAP) [1]. Unfortunately, these methods are computationally expensive, making the projection of data streams in real-time a non-trivial task. In this study, we introduce a novel variant of UMAP, called approximate UMAP (aUMAP). It aims at gener- ating rapid projections for real-time introspection. To study its suitability for real-time projecting, we benchmark the methods against standard UMAP and its neural network counterpart parametric UMAP [2]. Our results show that approximate UMAP delivers projections that replicate the projection space of standard UMAP while decreasing projection speed by an order of magnitude and maintaining the same training time.
• ANALYSIS OF THE EEG RESTING-STATE SIGNALS FOR BCIMattei, Enrico; Lozzi, Daniele; Di Matteo, Alessandro; Manes, Costanzo; Mignosi, Filippo; Polsinelli, Matteo; Placidi, Giuseppe; 10.3217/978-3-99161-014-4-062In the Brain-computer interface (BCI), the recognition of movements is useful for controlling exter nal devices, such as robotic arms, helping people with disabilities or performing remote operations in unsafe places. In this work, we present a new method to build an online BCI for motor execution classification that takes into account not only the movements but also the resting period beging essential to recognize when an individual is not engaged in any activity. An artificial intelligence model, EEGnet, was first trained on three classes of left and right-hand movements, and resting with 0.43 of ac curacy. The same type of network was trained on two classes by combining the three classes above, thus having left-right, rest-left, and rest-right, with 0.73, 0.67, 0.63 of accuracy, respectively. Therefore, the 2-classes EEGnet were combined in a network tree that is able to correctly classify not only left- and right-hand movements but also resting signals to improve the accuracy to 0.55 of these three classes.
• AN ALTERNATIVE TRAINING PROTOCOL FOR A MOTOR IMAGERY BMI BASED ON A COLLABORATIVE APPROACHPalatella, Alessio; Forin, Paolo; Tortora, Stefano; Menegatti, Emanuele; Tonin, Luca; 10.3217/978-3-99161-014-4-063Becoming proficient in the use of brainmachine interfaces (BMIs) represents a challenging task for the subjects, requiring long and intensive training. In this paper, we propose and explore the use of a collaborative BMI (cBMI) as an innovative training protocol that allows two subjects to learn together by cooperating in the control of a real robotic arm. Preliminary results on three pairs of subjects spanning five days of training highlight the promises of the proposed approach in reducing the training time and possibly mitigating the frustration in naive users.
• THE CHALLENGE OF DRIVING BCI WITH EMOTIONAL SIGNALS COLLECTED BY EEGLozzi, Daniele; Mattei, Enrico; Ciuffini, Roberta; Di Matteo, Alessandro; Marrelli, Alfonso; Ornello, Raffaele; Polsinelli, Matteo; Rosignoli, Chiara; Sacco, Simona; Placidi, Giuseppe; 10.3217/978-3-99161-014-4-064The paper describes the important chal lenges of driving a BCI through EEG signals of emotions. In particular, the complex emotional processing activated by the human brain and the necessity of generating elic itation protocols to synchronize the acquisition of EEG signals from emotions are presented. Besides, the limita tions of EEG in dealing with signals from emotions are also discussed. Then, the specific neuropsychological is sues related to the use of protocols for eliciting emotions are described. Due to the huge difficulty in managing the uncertainty deriving from the above issues, the sur prising results obtained by recently proposed automatic strategies for emotion classification and recognition, also raising doubts about the correctness of the results, are re ported and discussed. Finally, suggestions are presented regarding some procedures for uncertainty reduction and for the future complete development of EEG-based emotional BCIs.
• PROJECT NAFAS: ANNOUNCEMENT AND BRIEF OVERVIEWKrol, Laurens R.; Zander, Thorsten O.; 10.3217/978-3-99161-014-4-065Funded by the German federal agency Agentur für Innovation in der Cybersicherheit - “Innovation for Cybersecurity” (Cyberagentur) with a record €30 million, we announce Zander Labs’ Project NAFAS, which aims to integrate Brain-Computer Interface (BCI) technology with Artificial Intelligence (AI). By first addressing the traditional constraints of EEG-based neurotechnology and developing mobile, secure hardware capable of decoding multiple mental states in real time, this project paves the way for a new era of Neuroadaptive Human-Computer Interaction (HCI)— and, ultimately, Neuroadaptive AI. Beyond the project’s scientific aims which we briefly introduce, Project NAFAS itself represents confidence in the ability of the scientific community to solve the critical challenge of transitioning BCIs from theoretical constructs to practical real-world applications, and in the positive impact the resulting BCI technology can have in our daily lives.
• NEUROFEEDBACK PERFORMANCE UNDER CHALLENGING CONDITIONS: THE THETA-AGENCY INTERPLAYDussard, Claire; Pillette, Léa; Dumas, Cassandra; Pierrieau, Emeline; Hugueville, Laurent; Lau, Brian; Jeunet, Camille; George, Nathalie; 10.3217/978-3-99161-014-4-066Neurofeedback (NF) consists in training the selfregulation of some target neural activity. Yet, the neural underpinnings of NF performance remains largely unknown. Here, we investigated Motor Imagery (MI) based NF with EEG, training subjects to regulate motor- related activity in the large β (8-30 Hz) band. We examined the electrophysiological correlates of NF performance across the whole scalp and the frequency spectrum. In addition to the rewarded β activity, frontocentral θ activity predicted NF performance. The association was modulated by the participants' sense of agency over the feedback with stronger effects in participants with lower agency. Fostering agency in NF protocols may reduce cognitive effort and reliance on additional rythms beyond β. Considering these effects could be important for optimizing NF performance.
• PERIPHERAL NERVE STIMULATION AND AUDITORY SIMULATION CLOSED LOOP SYSTEM FOR SENSORY DECISION MAKING IN TRANSHUMERAL AMPUTEESSoghoyan, Gurgen; Biktimirov, Artur R.; Piliugin, Nikita S.; Sintsov, Mikhail Y.; Lebedev, Mikhail A.; 10.3217/978-3-99161-014-4-067Peripheral nerve stimulation (PNS) is a key method for restoring sensory feedback in upper-limb prostheses, yet the necessity of invasive feedback for sensory decision- making remains uncertain. In this study, two transhumeral amputees underwent sensory restoration in their phantom limbs via PNS. They performed an active exploration task using a tablet and closed-loop feedback system to assess their sensory decision-making abilities. In the task patient needed to differentiate among three hidden objects using PNS-based tactile feedback or auditory feedback. Interestingly, one patient successfully completed the task only in PNS trials, while the other demonstrated improved speed and accuracy with auditory stimulation. These findings suggest varying responses to different feedback modalities in different subjects. They indicate the potential significance of personalized approaches in designing sensory feedback systems for prosthetic users.
• VALIDATING NEUROPHYSIOLOGICAL PREDICTORS OF BCI PERFORMANCE ON A LARGE OPEN SOURCE DATASETTrocellier, David; N'Kaoua, Bernard; Lotte, Fabien; 10.3217/978-3-99161-014-4-068Brain-computer interfaces (BCI) are systems that process brain activity to decode specific commands from it such as motor imagery patterns generated when users imagine movements. Despite the growing interest in BCI, they present significant challenges, notably in decoding distinct neural patterns, due to considerable variability across and within users. The literature showed that various predictors were correlated with subject’s BCI performance. Among these indicators, neurophysiological predictors appeared to be the most effective, although studies generally involved small samples and results were not always replicated, thus questioning their reliability. In our study, we used a large dataset with 85 subjects to analyse the relationship between different predictors identified in the literature and BCI performance. Our findings reveal that only four of the six predictors tested could be replicated on this dataset. These results underscore the necessity of validating literature findings to ensure the reliability and applicability of such predictors.
• THE GOOD, THE BAD, AND THE UGLY OF IEEG SIGNALS: IDENTIFYING ARTIFACTUAL CHANNELS USING CONVOLUTIONAL NEURAL NETWORKSFreudenburg, Zachary V.; Zhong, Walker; Branco, Mariana P.; Ramsey, Nick; 10.3217/978-3-99161-014-4-069Intracranial electroencepghalography (iEEG) signals have established themselves as a key tool for studying human brain function due to its distinct combination of high spatial and temporal precision. The use of both cortical surface and stereo-EEG in effective epilepsy treatment has allowed researchers to study electrophysiology throughout the brain in relatively large numbers of subjects. This provides an opportunity to overcome, the sparse and varied nature of the brain tissue sampling inherent to the clinical use of iEEG by aggregating data across many subjects. Essential to the success of large-scale data aggregation is the efficient and robust identification of recording channels that are dominated by ‘noise’ or artifacts introduced by the recording environment or hardware failure. Here we test the effectiveness of training a convolutional neural network (CNN) for this purpose across multiple types of iEEG recordings. We conclude that a small CNN trained on hand labeled data from a small set of subjects can be applied to identify artifactual channels.
• INTRODUCING THE USE OF THERMAL NEUROFEEDBACKLe Jeune, François; Savalle, Emile; Lécuyer, Anatole; Macé, Marc J.-M.; Maurel, Pierre; Pillette, Léa; 10.3217/978-3-99161-014-4-070Motor imagery-based brain-computer interfaces (MI-BCIs) enable users to control digital devices by performing motor imagery tasks while their brain activity is recorded, typically using electroencephalography. Performing MI is challenging, especially for novices. To tackle this challenge, neurofeedback (NFB) training is frequently used and usually relies on visual feedback to help users learn to modulate the activity of their sensorimotor cortex when performing MI tasks. Improving the feedback provided during these training is essential. This study investigates the feasibility and effectiveness of using thermal feedback for MI-based NFB compared to visual feedback. Thirteen people participated to a NFB training session with visual-only, thermalonly, and combined visuo-thermal feedback. Both visualonly and combined visuo-thermal feedback elicited sig- nificantly greater desynchronization over the sensorimotor cortex compared to thermal-only feedback. No significant difference between visual-only and combined visuo-thermal feedback was found, thermal feedback thus not impairing visual feedback. This study outlines the need for further exploration of alternative feedback modalities in BCI research.
• DOUBLE-BLIND AND SHAM-CONTROLLED AUGMENTED REALITY EEG-NEUROFEEDBACK STUDYBerger, Lisa M.; Wood, Guilherme; Kober, Silvia E.; 10.3217/978-3-99161-014-4-071Traditional Neurofeedback (NF) designs are rather dull and only little engaging, which can negatively influence training performance. NF profits from interesting paradigms implementable through tools such as Virtual and Augmented Reality (AR). AR, however, is still very new in the field of NF and BCI but seems promising in hindsight of less Cybersickness and easier and cheaper usage for tele-rehabilitation, as modern smartphones support AR implementations. However, there are still no sham-controlled AR-based NF studies with larger samples. We propose a onesession sham-controlled and double-blinded NF study comparing AR with 2D feedback. The NF training consisted of sensorimotor rhythm (SMR) up-regulation and we tested 89 healthy participants. Results showed a numerically but non-significant increase in SMR across the NF runs in all four groups. Sham and real feedback groups did not differ in their performance. The study could show that AR is equally viable to 2D feedback and participants were not able to increase SMR within one NF training session.
• INTER-TASK TRANSFER LEARNING BETWEEN UPPER-LIMB MOTOR EXECUTION AND MOTOR IMAGERYPérez-Velasco, Sergio; Marcos-Martínez, Diego; Santamaría-Vázquez, Eduardo; Martínez-Cagigal, Víctor; Pascual-Roa, Beatriz; Hornero, Roberto; 10.3217/978-3-99161-014-4-072This study addresses a key challenge in motor imagery (MI)-based brain-computer interfaces (BCIs): improving the decoding accuracy of electroencephalography (EEG) signals. We investigate the intertask transfer learning potential between motor execution (ME) and MI to enhance the calibration phase of MI- BCIs. Utilizing the EEGSym deep learning network, we demonstrate that ME data can effectively train models for MI classification. Additionally, our analysis identifies a significant positive correlation between performances on ME and MI tasks. These findings support the feasibility of a ME-based calibration approach for MI tasks in BCI systems, leveraging the neural and functional similarities between ME and MI. This approach maintains BCI performance and potentially makes it easier to accommodate new users to the MI task while recording ME data during calibration, which could serve as an indicator of the expected MI accuracy. Furthermore, our results suggest that we can exploit the synergies between ME and MI without significantly reducing decoding accuracy of the user’s intentions.
• PURSUING THE IMPLEMENTATION OF A NEUROTUTOR: AN EEG-BASED CLASSIFICATION OF READING TYPESRomero-Morales, Héctor; Muñoz-Montes de Oca, Jenny Noemí; Torres-García, Alejandro Antonio; Villaseñor-Pineda, Luis; 10.3217/978-3-99161-014-4-073Electroencephalogram (EEG)-based brain-computer interfaces (BCI) emerged as systems to aid impaired people in daily life. Nowadays, the number of applications and target users of BCI has increased, including those for education purposes. An example of these applications, called neurotutor, was posited in 2015 for improving students’ learning process. As a first step towards developing a neurotutor, we analyzed the EEG responses related to two types of reading. Specifically, this work assessed whether a machine learning algorithm can distinguish accurately between both classes from features obtained from the signals using one of three wavelet-based techniques. Also, the impact of epoch length on classifier performance was assessed. The method performance was analyzed under two scenarios (intra-subject and inter-subject), outperforming previous work. The best average accuracies were 94.40 ± 5.10% and 54.40 ± 6.7% for intrasubject and inter-subject classification, respectively. Although the progress obtained for the intra-subject scenario is promising, several steps must be done to effectively implement a neurotutor, especially in inter-subject scenarios.
• ASSESSMENT OF SEVERAL EEG ACTIVE PARADIGMS IN LOCKED-IN SYNDROMESéguin, Perrine Rose; Maby, Emmanuel; Bouet, Romain; Gattaz, Lucie; Querry, Ambre; Rizzo, Lorianna; Farnè, Alessandro; Mattout, Jérémie; 10.3217/978-3-99161-014-4-074At first glance, Brain-computer interfaces (BCIs) appear to offer promising solutions for people who have global paralysis and are unable to operate conventional communication devices. However BCI efficacy remains low. To better assess the possible clinical reasons for this lack of efficacy, we conducted a study comparing the performance of patients in three paradigms: motor attempt, sustained auditory attention and spatial selective auditory attention. We included 14 persons with locked-in syndrome (LIS), one person in complete LIS and 27 healthy subjects. Preliminary results show that for the patient in complete LIS and a significant proportion of LIS patients, we could not detect their voluntary modulation of brain signals. Surprisingly, this absence of attentional biomarkers seem more prevalent in brainstem injury than in ALS. We discuss the possible impact of global paralysis on brain signals that are used to control BCIs.
• AN ONLINE SPIKE DETECTION AND MONITORING FRAMEWORK IN IEEG RECORDED USING BRAIN INTERCHANGE DEVICEBesheli, Behrang Fazli; Ayyoubi, Amir Hossein; Okkabaz, Jhan Luke; Swamy, Chandra Prakash; Quach, Michael M.; Miller, Kai J.; Worrell, Gregory; Ince, Nuri Firat; 10.3217/978-3-99161-014-4-075In this study, we developed and validated an online analysis framework in MATLAB Simulink for recording and analysis of intracranial electroencephalography (iEEG). This framework aims to detect interictal spikes in patients with epilepsy as the data is being recorded. An online spike detection was performed over 10-minute interictal iEEG data recorded with Brain Interchange CorTec in three human subjects. A pool of detected spikes is then broadcasted using User Datagram Protocol (UDP) to an external graphical user interface for further post-processing and visualization. The real-time spike detector demonstrated a 99% similarity index with the previously published offline detector, identifying interictal spikes. Furthermore, our findings indicated that channels with highest spike rates, captured with Brain Interchange CorTec, were in the epileptogenic focus. By enabling the detection of interictal spikes in an online fashion, this work provides early feedback on the probable seizure onset zone (SOZ) and suggests a promising direction for enhancing SOZ localization accuracy to clinicians, which is crucial for the surgical treatment of epilepsy.
• ENHANCING COMPUTATIONAL EFFICIENCY OF MOTOR IMAGERY BCI CLASSIFICATION WITH BLOCK-TOEPLITZ AUGMENTED COVARIANCE MATRICES AND SIEGEL METRICCarrara, Igor; Papadopoulo, Theodore; 10.3217/978-3-99161-014-4-076Electroencephalographic signals are represented as multidimensional datasets. We introduce an enhancement to the augmented covariance method (ACM), exploiting more thoroughly its mathematical properties, in order to improve motor imagery classification. Standard ACM emerges as a combination of phase space reconstruction of dynamical systems and of Riemannian geometry. Indeed, it is based on the construction of a Symmetric Positive Definite matrix to improve classification. But this matrix also has a Block-Toeplitz structure that was previously ignored. This work treats such matrices in the real manifold to which they belong: the set of Block-Toeplitz SPD matrices. After some manipulation, this set is can be seen as the product of an SPD manifold and a Siegel Disk Space. The proposed methodology was tested using the MOABB framework with a within-session evaluation procedure. It achieves a similar classification performance to ACM, which is typically better than – or at worse comparable to – state-of-the-art methods. But, it also improves consequently the computational efficiency over ACM, making it even more suitable for real time experiments.
• SYNTHESIZING EEG SIGNALS FROM EVENT-RELATED POTENTIAL PARADIGMS WITH CONDITIONAL DIFFUSION MODELSKlein, Guido; Guetschel, Pierre; Silvestri, Gianluigi; Tangermann, Michael; 10.3217/978-3-99161-014-4-077Generative models, specifically diffusion models, can alleviate data scarcity in the brain-computer interface field. While diffusion models have previously been successfully applied to electroencephalogram (EEG) data, existing models lack flexibility regarding sampling or require alternative representations of the EEG data. To overcome these limitations, we introduce a novel approach to conditional diffusion models that utilizes classifier-free guidance to directly generate subject-, session-, and classspecific EEG data. In addition to commonly used metrics, domain-specific metrics are employed to evaluate the specificity of the generated samples. The results indicate that the proposed model can generate EEG data that resembles real data for each subject, session, and class.
• EEG SINGLE-TRIAL DECODING OF VISUAL ART PREFERENCEWelter, Marc; Casal Martínez, Jesu; Redmond, Erin; Baum, Jonathan; Ward, Tomas; Lotte, Fabien; 10.3217/978-3-99161-014-4-078Brain-Computer-Interfaces (BCIs) able to decode aesthetic preference could improve user experience in digital spaces by personalizing aesthetic stimuli selection without requiring explicit user feedback that might interrupt aesthetic experience. However, neuroscientific understanding of aesthetic experience remains lacking, while the tried and tested BCI classification algorithms have not yet been applied to decode aesthetic preferences from EEG signals. We thus conducted an experiment in which participants where exposed to visual artworks in a virtual museum and requested to grade their preferences for each of them, all this while their EEG was being measured. Previous neuroaesthetic research suggested that oscillatory modulations in different neural frequency bands could be informative of aesthetic preference. Therefore, we tested a time-frequency fea- ture classification method widely used in BCIs, i.e. Filterbank Common Spatial Patterns feature extraction together with shrinkage Linear Discriminant Analysis, in a 2-class aesthetic preference classification problem. We report promising aesthetic preference decoding accuracies significantly and substantially above chance level.
• EXPLORING EOG MARKERS OF FATIGUE DURING MOTOR IMAGERY BCI USEDreyer, Pauline; Roc, Aline; Trocellier, David; Welter, Marc; Roy, Raphaëlle N.; Lotte, Fabien; 10.3217/978-3-99161-014-4-079Brain-Computer Interface (BCI) performance suffer from various variability sources, including intra-subject factors such as mental fatigue. While frequently measured using subjective reports, mental fatigue can also be assessed via blink parameters extracted from electro-oculography signals. To our knowledge, no study has yet evaluated blink parameters during motor imagery (MI) BCI use to assess the potential development of mental fatigue. In this study, the blinks of 23 MI-BCI participants were analyzed con- currently with subjective reports and BCI performance. Our results showed that blink parameters were correlated with neither MI-BCI performance nor subjective reports. However, they revealed a positive correlation between time-on-task and both blinks number and mean duration. Similarly, subjective fatigue was correlated with time-on- task. This suggests that blinks parameters may be useful for BCI user monitoring, although their relationship with BCI performance and fatigue needs further studies. Alto- gether, this study paves the way towards a better understanding of mental fatigue during BCI use, and in finding solutions to mitigate it.
• SHOULD ATTEMPTED MOVEMENTS REPLACE MOTOR IMAGERY IN BCI? THE ISSUE OF COMPATIBILITY WITH GAZE USEShishkin, Sergei L.; Yashin, Artem S.; Shevtsova, Yulia G.; Vasilyev, Anatoly N.; 10.3217/978-3-99161-014-4-080Attempted movements have recently become common in invasive studies as a way to send commands via BCIs and have been successfully employed in some studies of neurorehabilitation using noninvasive BCIs. Nevertheless, they are still far less common in noninvasive BCIs than motor imagery. We proposed a hypothesis that attempted movements can be more compatible with the interaction with the external world than imaginary movements and therefore may help to use BCIs more effectively. The hypothesis was tested in 15 healthy participants who were asked to make prosaccades, which represented an external task, and quasi-movements (movement attempts minimized down to complete extinction of related muscle activation), which were used as a model of attempted movements. Preliminary results of the study were mostly in line with the predictions, although more studies are required for more definite conclusions. The study also may be considered as a new demonstration of the potential of quasi-movements, a very little explored phenomenon, for BCI research.
• USING A CNN-LSTM ARCHITECTURE WITH DATA AUGMENTATION TO IMPROVE HD-ECOG SPOKEN SYLLABLE CLASSIFICATIONMirehkoohi, Mehdi Javani; Freudenburg, Zachary; Neumann, Amira; Ramsey, Nick F.; 10.3217/978-3-99161-014-4-081Brain-Computer Interfaces (BCIs) have emerged as vital tools in understanding and assisting individuals with LIS due to neurological diseases such as ALS. This study focuses on the and feasibility of recognizing spoken syllables from implanted HD-ECoG signals as a platform for Speech BCIs. We propose a hybrid deep learning model, which uses a modified EEGNet as a feature extractor coupled with an LSTM. A primary challenge in this domain is the limited quantity of ECoG data. To address this challenge, we employ window clipping as a data augmentation technique, effectively increasing the amount of training data available for the model. Using a dataset comprising recordings from six subjects implanted with HD-ECoG, we evaluate our proposed method. Results indicate a notable improvement in classification accuracy achieved through the designed hybrid DL model. Furthermore, our findings elucidate the distinctive impact of data augmentation methods in further enhancing the performance of our designed model.
• NEURAL CORRELATES OF EXPERTISE DURING KINESTHETIC MOTOR IMAGERY: SHOULD WE REWARD MAXIMUM SMR-ERD?Izac, Margaux; Rossignol, Eléa; Pierrieau, Emeline; Grechukhin, Natalia; Coudroy, Elina; N'Kaoua, Bernard; Pillette, Léa; Jeunet-Kelway, Camille; 10.3217/978-3-99161-014-4-082Athletes practice Kinesthetic Motor Imagery (KMI) for its many benefits. However, lack of feedback impairs regular practice. To optimise KMI efficiency, athletes can use BCIs. Whereas current BCI protocols targeting KMI abilities reward maximum desyn- chronisation (ERD) of sensorimotor rhythms (SMRs, 12- 15Hz), the neural efficiency hypothesis raises the question “what neurophysiological markers should we reinforce?”. We hypothesised that experts’ SMR-ERDs would differ from novices’, in particular when imagining a mastered task. To test this hypothesis, EEG activity was recorded during KMI of bio-mechanically similar tasks: one mastered by experts only and one requiring no specific expertise. Self-reported measures based on validated questionnaires were collected to assess KMI ability and MI frequency of use and to measure their potential impact on SMR-ERD. Experts (basketball players) reported higher perceived KMI abilities than novices, but similar MI practice frequency. In addition, experts showed a stronger SMR-ERD than novices. This effect was only weakly mediated by perceived KMI ability, seeming mainly driven by sport expertise.
• BAYESIAN MODEL OF INDIVIDUAL LEARNING TO CONTROL A MOTOR IMAGERY BCIAnnicchiarico, Côme; Lotte, Fabien; Mattout, Jérémie; 10.3217/978-3-99161-014-4-083The cognitive mechanisms underlying subjects' self-regulation in Brain-Computer Interface (BCI) and neurofeedback (NF) training remain poorly understood. Yet, a mechanistic computational model of each individual learning trajectory is required to improve the reliability of BCI applications. The few existing attempts mostly rely on model-free (reinforcement learning) approaches. Hence, they cannot capture the strategy developed by each subject and neither finely predict their learning curve. In this study, we propose an alternative, model-based approach rooted in cognitive skill learning within the Active Inference framework. We show how BCI training may be framed as an inference problem under high uncertainties. We illustrate the proposed approach on a previously published synthetic Motor Imagery ERD laterality training. We show how simple changes in model parameters allow us to qualitatively match experimental results and account for various subject. In the near future, this approach may provide a powerful computational to model individual skill learning and thus optimize and finely characterize BCI training.
• USING TRANSFORMER NETWORKS FOR STREAMING SPEECH SYNTHESIS FROM INTRACRANIAL EEGAmigó-Vega, Joaquín; Verwoert, Maxime; Ottenhoff, Maarten C.; Kubben, Pieter L.; Herff, Christian; 10.3217/978-3-99161-014-4-084peech Neuroprostheses have the potential to enable users to communicate without the need for overt muscle movement. Several recent approaches have demonstrated the feasibility of decoding textual and acoustic representations of speech from invasively measured neural activity. However, most approaches decode or synthesize speech after several seconds or complete utterances. While this provides tremendous communicative ability to patients, it lacks the full expressive power of natural conversations. Ideally, a speech neuroprosthesis would synthesize speech without a noticeable delay. Here, we present a real-time speech decoding pipeline that generates speech output in a streaming fashion, i.e., with delays of less than 40 ms. Intracranial EEG data is measured, processed, decoded, and synthesized into an audio waveform using our fast and modular framework. Notably, we employ a Transformer architecture for the decoding step from neural features to a spectral representation of speech.
• NEUROPHONE: REAL-TIME BRAIN-MOBILE PHONE INTERFACEAbdelhafez, Norhan; Tantawy, Manal; Sayed, Abdelrahman; Ekramy, Nora; Nour-Eldin, Mohammed; 10.3217/978-3-99161-014-4-085The extensively studied P300 component of the human event-related potential in cognitive neuroscience has significant applications, including constructing BCI systems for individuals with motor disabilities. However, accurately and efficiently identifying the P300 component in EEG data poses challenges due to the low signal-to-noise ratio and biological diversity among subjects. To address this, cutting-edge deep learning architectures were developed and employed. Initially, digital signal processing techniques were applied, followed by training and evaluation of DL models like Chrononet, EEGNet, DCRNN, CNNs, and RNNs. Results revealed that our lightweight CNN model, combined with K-fold crossvalidation and weighted class, achieved the highest average classification accuracy of 98% surpassing other models for subject-dependent P300 classification. This high-performing CNN model facilitated the creation of NeuroPhone, a communication application grounded in the core principles of BCI systems.
• NOVEL MATERIALS FOR BRAIN COMPUTER INTERFACES: PERSPECTIVES AND ASPECTS OF COMBINATION OF A MAGNETOELECTRIC STIMULATOR AND A GRAPHENE MICROTRANSISTOR ARRAY RECORDING SYSTEMMatsoukis, Stratis; Scharinger, Josef; Covelo, Joana; Cancino-Fuentes, Nathalia; Sanchez-Vives, Maria V.; Edlinger, Guenter; Guger, Christoph; 10.3217/978-3-99161-014-4-086In this paper, we explore the innovative combination of magneto-electric nanoparticles (MENPs) and graphene solution-gated field-effect transistors (gSGFETs) to advance brain-computer interfaces (BCIs). ME materials, known for their wireless and minimally invasive brain stimulation capabilities, are combined with gSGFETs, known for their high-resolution neural recording. Our research explores the potential benefits of this hybrid approach, including reduced artifacts, enhanced spatial resolution, and improved detection of subthreshold phenomena and DC potentials. A hardware and software setup is proposed and possible data analysis methods that will assist in the further development of the system are reviewed. This combined technology offers a promising direction for advanced BCIs and represents a significant advance in neural engineering.
• COMPARING FINGERS AND GESTURES FOR BCI CONTROL USING AN OPTIMIZED CLASSICAL MACHINE LEARNING DECODERKeller, Dirk; Vansteensel, Mariska J.; Mehrkanoon, Siamak; Branco, Mariana P.; 10.3217/978-3-99161-014-4-087Severe impairment of the central motor network can result in loss of motor function, clinically recognized as Locked-in Syndrome. Advances in BrainComputer Interfaces offer a promising avenue for partially restoring compromised communicative abilities by decoding different types of hand movements from the sensorimotor cortex. In this study, we collected ECoG recordings from 8 epilepsy patients and compared the de- codability of individual finger flexion and hand gestures with the resting state, as a proxy for a one-dimensional brain-click. The results show that all individual finger flexion and hand gestures are equally decodable across multiple models and subjects (>98.0%). In particular, hand movements, involving index finger flexion, emerged as promising candidates for brain-clicks. When decoding among multiple hand movements, finger flexion appears to outperform hand gestures (96.2% and 92.5% respectively) and exhibit greater robustness against misclassi- fication errors when all hand movements are included. These findings highlight that optimized classical machine learning models with feature engineering are viable decoder designs for communication-assistive systems.
• REFERENCING SCHEMES AND THEIR EFFECT ON OSCILLATIONS AND BROADBAND POWER SPECTRAL SHIFTS IN STEREOELECTROENCEPHALOGRAPHYRockhill, Alexander P.; Jensen, Michael A.; Swann, Nicole C.; Raslan, Ahmed M.; Hermes, Dora; Miller, Kai J.; 10.3217/978-3-99161-014-4-088Choosing a referencing scheme for stereoelectroencephalography (SEEG) is complicated by the varying depth of contact locations and, consequently, the different tissue that is being recorded from. In order to better understand how changes in electrophysiology related to movement are affected by the choice of reference, we examined how 16 different referencing schemes effected alpha (8 - 13 Hz) and beta (13 - 30 Hz) oscillations and high-frequency broadband (HFB) power (65 - 115 Hz). We found the choice of referencing scheme has more complicated effects than previously described and recommend using different referencing schemes as a methodological tool to optimize brain-computer interface (BCI) performace.
• FUNCTIONAL REPRESENTATION OF SOMATOSENSORY, VISUAL, AND REINFORCEMENT PROCESSING ON THE CANINE BRAIN SURFACELampert, Frederik; Mivalt, Filip; Kim, Inyong; Ince, Nuri F.; Kim, Jiwon; Okkabaz, Jhan L.; Van den Boom, Max A.; Kremen, Vaclav; Ali, Rushna; Coenen, Volker A.; Schalk, Gerwin; Brunner, Peter; Worrell, Gregory A.; Miller, Kai J.; 10.3217/978-3-99161-014-4-089Implantable brain-computer interface (BCI) systems, promising for neurological disorder treatment, often encounter high technical barriers. Our fully-implanted CorTec BrainInterchange-BCI2000 ecosystem, aimed for widespread open-source adoption, demonstrates functionality through a year-post-implant canine study, using a brain surface electrocorticography (ECoG) construct. Broadband power-spectrum increases have been shown to track neural population activity in humans, and we find that they reveal distinct functional representation for processing of visual, somatosensory, and auditory reinforcement stimuli in the canine (captured at 65- 150Hz). Canine visual and somatosensory rhythms resemble human alpha and beta rhythms but at different frequencies: a ∼15Hz visual rhythm in the occipital analog (marginal gyrus) suppresses with light exposure, and a ∼24Hz somatosensory rhythm diminishes upon petting. These findings indicate a unique canine neurophysiology and confirm the BCI2000-BrainInterchange ecosystem’s robustness a year after the implantation. This ecosystem holds promise for developing open-source BCI devices to assist patients with neurological conditions.
• MOVEMENT ASSOCIATED INCREASE IN THALAMIC BROADBAND SPECTRAL POWER IS A POTENTIAL FEATURE FOR BCI CONTROLKlassen, Bryan T.; Baker, Matthew R.; Ojeda Valencia, Gabriela; Miller, Kai J.; 10.3217/978-3-99161-014-4-090Signals within the subcortical brain regions may be useful as control signals for brain computer interfaces (BCI). In this study we show, using a simple hand movement task, that focal increases in broadband spectral power, which are commonly used to control cortically-based BCI interfaces, may also be observed in the ventralis intermedius (VIM) thalamic nucleus, a key relay for the cerebellar outputs to the motor cortex that help to regulate voluntary movement.
• DYNAMIC SUPPRESSION OF THE CORTEX THROUGH SYNCHRONISATION DURING BRAIN COMPUTER INTERFACINGPermezel, Fiona E.; Jensen, Michael A.; Hermes, Dora; Miller, Kai J.; 10.3217/978-3-99161-014-4-091This study investigates the neural dynamics of motor imagery and brain-computer interface (BCI) feedback through electrocorticography (ECoG). It focuses on how 12-20Hz rhythm entrainment with broadband power indicates cortical synchronization and suppression. The research examines 12-20Hz rhythm entrainment across rest and active phases in a motor task and BCI imagery feedback task. Using speech-associated broadband power increases in a speech motor area, a patient controlled a BCI system with word repetition imagery. The study examined broadband power shifts between rest and active task conditions, revealing increased power shifts in the speech motor area during the BCI imagery task as well as a unique activation in the dorsal motor area. Notably, it found increased broadband power to 12-20Hz rhythm coupling, indicating suppression of cortical activity, in the dorsal motor cortex during the BCI imagery feedback task's rest phase compared to the motor task rest phase, which may be suggestive of "cognitive control" over cortical suppression.
• WAVELET PACKET DECOMPOSITION TO EXTRACT FREQUENCY FEATURES FROM SPEECH IMAGERYTates, Alberto; Matran-Fernandez, Ana; Halder, Sebastian; Daly, Ian; 10.3217/978-3-99161-014-4-092Speech Imagery (SI) is considered an intuitive paradigm for Brain-Computer Interface designs in particular for communication applications. In this work, we use Electroencephalography (EEG) for offline SI decoding. We recorded covert speech from 17 participants. We tested two types of wavelet decomposition techniques. Specifically, we considered coefficients from 6 decomposition levels with Discrete Wavelet Transform (DWT) and multiple 2 Hz spaced packets with Wavelet Packet Decomposition (WPD), we computed different statistical features from such coefficients to form vector inputs for our binary-class classification approach. We approached the issue of feature/sample gap by using the Maximum Relevance and Minimum Redundancy (MRMR) feature selector algorithm to select the most informative features. We achieved a mean accuracy of 76.6% ± 16 and demonstrated the potential of WPD to extract narrow-band features, and how its refined representation outperforms DWT in SI decoding.
• EEG CORRELATES OF ERROR-RELATED ACTIVITY DURING BALLISTIC COMPUTER MOUSE MOVEMENTSAmaunam, Idorenyin; Sultana, Mushfika; Rodriguez-Herreros, Borja; Tadi, Tej; Leeb, Robert; Perdikis, Serafeim; 10.3217/978-3-99161-014-4-093It has been repeatedly shown that processing of perceived errors in the human brain may elicit some type of evoked response in electroencephalography (EEG) collectively termed as Error-Related Potential (ErrP). The study of ErrP signatures offers a potential back door to better understanding how the brain encodes and reacts to errors and a useful tool for poking adaptation and learning, but also has several practical applications in Brain-Computer Interface (BCI) and general Human-Computer Interaction (HCI). The bulk of this literature has focused on so-called “interaction” ErrP, reflecting the response to discrete events occurring during self-paced, casual interaction of a subject with their en- vironment. Here we present a two-case study investigating the existence and characteristics of ErrP EEG correlates in an eye-hand coordination task consisting in “ballistic” computer mouse movements, where the action and reaction time constraints imposed on the subject are extremely tight. We show that clear EEG substrates of error processing can be retrieved for both subjects and bare strong similarities with the interaction ErrP waveforms. The findings of this work suggest the possibility of detecting, in real-time, errors committed during fast-paced interaction, thus potentially enabling automatic ErrP-based error correction in real-world BCI and HCI scenarios.
• FROM CUE-BASED TO SELF-PACED MOVEMENT DETECTION: INFLUENCE OF THE CUE ON TRAINING DATASuwandjieff, Patrick; Müller-Putz, Gernot R.; 10.3217/978-3-99161-014-4-094The utilization of a visual cue plays a significant role in enhancing the operational efficiency of brain-computer interface (BCI) systems for individuals with Locked-In Syndrome (LIS). This significance arises from the absence of a reliable method to discern the actual initiation of attempted movements in these patients. First, the decoders for identifying or classifying self-initiated movements need to be trained on cue-based paradigms. However, these cues can elicit neural activity (e.g., visual/auditory evoked potentials, cognitive processing, etc.) that obscures the neural dynamics of movement, thus negatively influencing the performance of the decoder. Therefore, we implemented four novel visual cues with the intention to reduce these effects to a minimum. Our research findings indicate that the effectiveness of classification performance in self-paced EEG recordings when the decoder is trained on cue-based data for movement tasks, is significantly impacted by the design of the cue.
• TOWARDS A MODEL-BASED PERSONALIZATION APPROACH FOR DRIVING A BCIDash, Adyasha; Wriessnegger, Selina Christin; 10.3217/978-3-99161-014-4-095A Brain-Computer Interface (BCI) translates a person’s intent, derived from brain signals, into control commands for various applications. This work focuses on Motor Imagery- based BCI (MI-BCI), specifically emphasizing sensorimotorrhythm (SMR) and MI as the relevant task. While improvements have been made in classification algorithms and signal acquisition, human factors influencing user-BCI compatibility remain underexplored. User performance in MI-BCI systems is impacted by personal, psychological, and neurophysiological factors, leading to a phenomenon termed “BCI illiteracy”. In this work, we aim to address BCI illiteracy through a systematic, standardized study, incorporating various human factors to enhance user performance by developing a neural network model predicting a trainability score and a training regime. To achieve this, the MI-BCI systems use population-specific indicators and task-based modulators, integrating anatomical, psychological, and neurophysiological information (EEG, biosignals). The proposed model-based personalization approach offers reproducible, innovative, and open-source training protocols to boost BCI performance avoiding prolonged and ineffective training sessions. The ultimate goal is to eliminate BCI illiteracy as a barrier to compatibility between users and BCI systems.