Introduction to Large Language Models (LLMs) for dementia care and research.

Introduction: Dementia is a progressive neurodegenerative disorder that affects cognitive abilities including memory, reasoning, and communication skills, leading to gradual decline in daily activities and social engagement. In light of the recent advent of Large Language Models (LLMs) such as ChatGPT, this paper aims to thoroughly analyse their potential applications and usefulness in dementia care and research. Method: To this end, we offer an introduction into LLMs, outlining the key features, capabilities, limitations, potential risks, and practical considerations for deployment as easy-to-use software (e.g., smartphone apps). We then explore various domains related to dementia, identifying opportunities for LLMs to enhance understanding, diagnostics, and treatment, with a broader emphasis on improving patient care. For each domain, the specific contributions of LLMs are examined, such as their ability to engage users in meaningful conversations, deliver personalized support, and offer cognitive enrichment. Potential benefits encompass improved social interaction, enhanced cognitive functioning, increased emotional well-being, and reduced caregiver burden. The deployment of LLMs in caregiving frameworks also raises a number of concerns and considerations. These include privacy and safety concerns, the need for empirical validation, user-centered design, adaptation to the user's unique needs, and the integration of multimodal inputs to create more immersive and personalized experiences. Additionally, ethical guidelines and privacy protocols must be established to ensure responsible and ethical deployment of LLMs. Results: We report the results on a questionnaire filled in by people with dementia (PwD) and their supporters wherein we surveyed the usefulness of different application scenarios of LLMs as well as the features that LLM-powered apps should have. Both PwD and supporters were largely positive regarding the prospect of LLMs in care, although concerns were raised regarding bias, data privacy and transparency. Discussion: Overall, this review corroborates the promising utilization of LLMs to positively impact dementia care by boosting cognitive abilities, enriching social interaction, and supporting caregivers. The findings underscore the importance of further research and development in this field to fully harness the benefits of LLMs and maximize their potential for improving the lives of individuals living with dementia.

Targeted memory reactivation in human REM sleep elicits detectable reactivation.

It is now well established that memories can reactivate during non-rapid eye movement (non-REM) sleep, but the question of whether equivalent reactivation can be detected in rapid eye movement (REM) sleep is hotly debated. To examine this, we used a technique called targeted memory reactivation (TMR) in which sounds are paired with learned material in wake, and then re-presented in subsequent sleep, in this case REM, to trigger reactivation. We then used machine learning classifiers to identify reactivation of task-related motor imagery from wake in REM sleep. Interestingly, the strength of measured reactivation positively predicted overnight performance improvement. These findings provide the first evidence for memory reactivation in human REM sleep after TMR that is directly related to brain activity during wakeful task performance.

Sleep is crucial for rest and recovery, but it also allows the brain to process things it has learned while awake. This is why a person may go to bed frustrated with learning a tune on the piano but wake up the next morning ready to play it without fumbling. For this to happen, it is thought that memories must be reactivated during sleep ­ something which can be studied by monitoring brain activity. While it has been shown that memory reactivation occurs in some stages of human sleep, it was unclear whether it occurred in a specific stage known as REM sleep ­ which is important for learning. To study memory reactivation during REM sleep, Abdellahi et al. recruited volunteers and monitored their brain activity during an 'adaptation night' when certain sounds played as they slept. The following day, memories ­ such as an image or pressing a certain button ­ were paired with the sounds, which were replayed during REM sleep the following night to trigger memory reactivation (experimental night). Abdellahi et al. measured how strongly brain activity during each night related to the waking activity when the sound pairing tasks were imagined and compared the adaptation and experimental nights. The experimental night showed clear signs of memory reactivation after the sounds were played during REM sleep, suggesting that the sounds triggered memories of the associated images or buttons. These findings show that in humans, brain activity patterns that indicate memory reactivation can be identified during REM sleep. The work paves the way for future studies into the characteristics of this memory reactivation and how to trigger it in a way that leads to improvements in memory.

A Skewed Loss Function for Correcting Predictive Bias in Brain Age Prediction.

In neuroimaging, the difference between predicted brain age and chronological age, known as brain age delta, has shown its potential as a biomarker related to various pathological phenotypes. There is a frequently observed bias when estimating brain age delta using regression models. This bias manifests as an overestimation of brain age for young participants and an underestimation of brain age for older participants. Therefore, the brain age delta is negatively correlated with chronological age, which can be problematic when evaluating relationships between brain age delta and other age-associated variables. This paper proposes a novel bias correction method for regression models by introducing a skewed loss function to replace the normal symmetric loss function. The regression model then behaves differently depending on whether it makes overestimations or underestimations. Our approach works with any type of MR image and no specific preprocessing is required, as long as the image is sensitive to age-related changes. The proposed approach has been validated using three classic deep learning models, namely ResNet, VGG, and GoogleNet on publicly available neuroimaging aging datasets. It shows flexibility across different model architectures and different choices of hyperparameters. The corrected brain age delta from our approach then has no linear relationship with chronological age and achieves higher predictive accuracy than a commonly-used two-stage approach.

Targeting targeted memory reactivation: Characteristics of cued reactivation in sleep.

Targeted memory reactivation (TMR) is a technique in which sensory cues associated with memories during wake are used to trigger memory reactivation during subsequent sleep. The characteristics of such cued reactivation, and the optimal placement of TMR cues, remain to be determined. We built an EEG classification pipeline that discriminated reactivation of right- and left-handed movements and found that cues which fall on the up-going transition of the slow oscillation (SO) are more likely to elicit a classifiable reactivation. We also used a novel machine learning pipeline to predict the likelihood of eliciting a classifiable reactivation after each TMR cue using the presence of spindles and features of SOs. Finally, we found that reactivations occurred either immediately after the cue or one second later. These findings greatly extend our understanding of memory reactivation and pave the way for development of wearable technologies to efficiently enhance memory through cueing in sleep.

Quality assessment of anatomical MRI images from generative adversarial networks: Human assessment and image quality metrics.

BACKGROUND: Generative Adversarial Networks (GANs) can synthesize brain images from image or noise input. So far, the gold standard for assessing the quality of the generated images has been human expert ratings. However, due to limitations of human assessment in terms of cost, scalability, and the limited sensitivity of the human eye to more subtle statistical relationships, a more automated approach towards evaluating GANs is required. NEW METHOD: We investigated to what extent visual quality can be assessed using image quality metrics and we used group analysis and spatial independent components analysis to verify that the GAN reproduces multivariate statistical relationships found in real data. Reference human data was obtained by recruiting neuroimaging experts to assess real Magnetic Resonance (MR) images and images generated by a GAN. Image quality was manipulated by exporting images at different stages of GAN training. RESULTS: Experts were sensitive to changes in image quality as evidenced by ratings and reaction times, and the generated images reproduced group effects (age, gender) and spatial correlations moderately well. We also surveyed a number of image quality metrics. Overall, Fréchet Inception Distance (FID), Maximum Mean Discrepancy (MMD) and Naturalness Image Quality Evaluator (NIQE) showed sensitivity to image quality and good correspondence with the human data, especially for lower-quality images (i.e., images from early stages of GAN training). However, only a Deep Quality Assessment (QA) model trained on human ratings was able to reproduce the subtle differences between higher-quality images. CONCLUSIONS: We recommend a combination of group analyses, spatial correlation analyses, and both distortion metrics (FID, MMD, NIQE) and perceptual models (Deep QA) for a comprehensive evaluation and comparison of brain images produced by GANs.

Towards an objective evaluation of EEG/MEG source estimation methods - The linear approach.

The spatial resolution of EEG/MEG source estimates, often described in terms of source leakage in the context of the inverse problem, poses constraints on the inferences that can be drawn from EEG/MEG source estimation results. Software packages for EEG/MEG data analysis offer a large choice of source estimation methods but few tools to experimental researchers for methods evaluation and comparison. Here, we describe a framework and tools for objective and intuitive resolution analysis of EEG/MEG source estimation based on linear systems analysis, and apply those to the most widely used distributed source estimation methods such as L2-minimum-norm estimation (L2-MNE) and linearly constrained minimum variance (LCMV) beamformers. Within this framework it is possible to define resolution metrics that define meaningful aspects of source estimation results (such as localization accuracy in terms of peak localization error, PLE, and spatial extent in terms of spatial deviation, SD) that are relevant to the task at hand and can easily be visualized. At the core of this framework is the resolution matrix, which describes the potential leakage from and into point sources (point-spread and cross-talk functions, or PSFs and CTFs, respectively). Importantly, for linear methods these functions allow generalizations to multiple sources or complex source distributions. This paper provides a tutorial-style introduction into linear EEG/MEG source estimation and resolution analysis aimed at experimental (rather than methods-oriented) researchers. We used this framework to demonstrate how L2-MNE-type as well as LCMV beamforming methods can be evaluated in practice using software tools that have only recently become available for routine use. Our novel methods comparison includes PLE and SD for a larger number of methods than in similar previous studies, such as unweighted, depth-weighted and normalized L2-MNE methods (including dSPM, sLORETA, eLORETA) and two LCMV beamformers. The results demonstrate that some methods can achieve low and even zero PLE for PSFs. However, their SD as well as both PLE and SD for CTFs are far less optimal for all methods, in particular for deep cortical areas. We hope that our paper will encourage EEG/MEG researchers to apply this approach to their own tasks at hand.

Ongoing neural oscillations predict the post-stimulus outcome of closed loop auditory stimulation during slow-wave sleep.

Large slow oscillations (SO, 0.5-2 Hz) characterise slow-wave sleep and are crucial to memory consolidation and other physiological functions. Manipulating slow oscillations may enhance sleep and memory, as well as benefitting the immune system. Closed-loop auditory stimulation (CLAS) has been demonstrated to increase the SO amplitude and to boost fast sleep spindle activity (11-16 Hz). Nevertheless, not all such stimuli are effective in evoking SOs, even when they are precisely phase locked. Here, we studied what factors of the ongoing activity patterns may help to determine what oscillations to stimulate to effectively enhance SOs or SO-locked spindle activity. Hence, we trained classifiers using the morphological characteristics of the ongoing SO, as measured by electroencephalography (EEG), to predict whether stimulation would lead to a benefit in terms of the resulting SO and spindle amplitude. Separate classifiers were trained using trials from spontaneous control and stimulated datasets, and we evaluated their performance by applying them to held-out data both within and across conditions. We were able to predict both when large SOs occurred spontaneously, and whether a phase-locked auditory click effectively enlarged them with good accuracy for predicting the SO trough (∼70%) and SO peak values (∼80%). Also, we were able to predict when stimulation would elicit spindle activity with an accuracy of ∼60%. Finally, we evaluate the importance of the various SO features used to make these predictions. Our results offer new insight into SO and spindle dynamics and may suggest techniques for developing future methods for online optimization of stimulation.

Text Mining of Adverse Events in Clinical Trials: Deep Learning Approach.

BACKGROUND: Pharmacovigilance and safety reporting, which involve processes for monitoring the use of medicines in clinical trials, play a critical role in the identification of previously unrecognized adverse events or changes in the patterns of adverse events. OBJECTIVE: This study aims to demonstrate the feasibility of automating the coding of adverse events described in the narrative section of the serious adverse event report forms to enable statistical analysis of the aforementioned patterns. METHODS: We used the Unified Medical Language System (UMLS) as the coding scheme, which integrates 217 source vocabularies, thus enabling coding against other relevant terminologies such as the International Classification of Diseases-10th Revision, Medical Dictionary for Regulatory Activities, and Systematized Nomenclature of Medicine). We used MetaMap, a highly configurable dictionary lookup software, to identify the mentions of the UMLS concepts. We trained a binary classifier using Bidirectional Encoder Representations from Transformers (BERT), a transformer-based language model that captures contextual relationships, to differentiate between mentions of the UMLS concepts that represented adverse events and those that did not. RESULTS: The model achieved a high F1 score of 0.8080, despite the class imbalance. This is 10.15 percent points lower than human-like performance but also 17.45 percent points higher than that of the baseline approach. CONCLUSIONS: These results confirmed that automated coding of adverse events described in the narrative section of serious adverse event reports is feasible. Once coded, adverse events can be statistically analyzed so that any correlations with the trialed medicines can be estimated in a timely fashion.

The hippocampus as the switchboard between perception and memory.

Adaptive memory recall requires a rapid and flexible switch from external perceptual reminders to internal mnemonic representations. However, owing to the limited temporal or spatial resolution of brain imaging modalities used in isolation, the hippocampal-cortical dynamics supporting this process remain unknown. We thus employed an object-scene cued recall paradigm across two studies, including intracranial electroencephalography (iEEG) and high-density scalp EEG. First, a sustained increase in hippocampal high gamma power (55 to 110 Hz) emerged 500 ms after cue onset and distinguished successful vs. unsuccessful recall. This increase in gamma power for successful recall was followed by a decrease in hippocampal alpha power (8 to 12 Hz). Intriguingly, the hippocampal gamma power increase marked the moment at which extrahippocampal activation patterns shifted from perceptual cue toward mnemonic target representations. In parallel, source-localized EEG alpha power revealed that the recall signal progresses from hippocampus to posterior parietal cortex and then to medial prefrontal cortex. Together, these results identify the hippocampus as the switchboard between perception and memory and elucidate the ensuing hippocampal-cortical dynamics supporting the recall process.

Correlation Constraints for Regression Models: Controlling Bias in Brain Age Prediction.

In neuroimaging, the difference between chronological age and predicted brain age, also known as brain age delta, has been proposed as a pathology marker linked to a range of phenotypes. Brain age delta is estimated using regression, which involves a frequently observed bias due to a negative correlation between chronological age and brain age delta. In brain age prediction models, this correlation can manifest as an overprediction of the age of young brains and an underprediction for elderly ones. We show that this bias can be controlled for by adding correlation constraints to the model training procedure. We develop an analytical solution to this constrained optimization problem for Linear, Ridge, and Kernel Ridge regression. The solution is optimal in the least-squares sense i.e., there is no other model that satisfies the correlation constraints and has a better fit. Analyses on the PAC2019 competition data demonstrate that this approach produces optimal unbiased predictive models with a number of advantages over existing approaches. Finally, we introduce regression toolboxes for Python and MATLAB that implement our algorithm.

MVPA-Light: A Classification and Regression Toolbox for Multi-Dimensional Data.

MVPA-Light is a MATLAB toolbox for multivariate pattern analysis (MVPA). It provides native implementations of a range of classifiers and regression models, using modern optimization algorithms. High-level functions allow for the multivariate analysis of multi-dimensional data, including generalization (e.g., time x time) and searchlight analysis. The toolbox performs cross-validation, hyperparameter tuning, and nested preprocessing. It computes various classification and regression metrics and establishes their statistical significance, is modular and easily extendable. Furthermore, it offers interfaces for LIBSVM and LIBLINEAR as well as an integration into the FieldTrip neuroimaging toolbox. After introducing MVPA-Light, example analyses of MEG and fMRI datasets, and benchmarking results on the classifiers and regression models are presented.

Evidence that neural information flow is reversed between object perception and object reconstruction from memory.

Remembering is a reconstructive process, yet little is known about how the reconstruction of a memory unfolds in time in the human brain. Here, we used reaction times and EEG time-series decoding to test the hypothesis that the information flow is reversed when an event is reconstructed from memory, compared to when the same event is initially being perceived. Across three experiments, we found highly consistent evidence supporting such a reversed stream. When seeing an object, low-level perceptual features were discriminated faster behaviourally, and could be decoded from brain activity earlier, than high-level conceptual features. This pattern reversed during associative memory recall, with reaction times and brain activity patterns now indicating that conceptual information was reconstructed more rapidly than perceptual details. Our findings support a neurobiologically plausible model of human memory, suggesting that memory retrieval is a hierarchical, multi-layered process that prioritises semantically meaningful information over perceptual details.

Activity and Connectivity Differences Underlying Inhibitory Control Across the Adult Life Span.

Inhibitory control requires precise regulation of activity and connectivity within multiple brain networks. Previous studies have typically evaluated age-related changes in regional activity or changes in interregional interactions. Instead, we test the hypothesis that activity and connectivity make distinct, complementary contributions to performance across the life span and the maintenance of successful inhibitory control systems. A representative sample of healthy human adults in a large, population-based life span cohort performed an integrated Stop-Signal (SS)/No-Go task during functional magnetic resonance imaging (n = 119; age range, 18-88 years). Individual differences in inhibitory control were measured in terms of the SS reaction time (SSRT), using the blocked integration method. Linear models and independent components analysis revealed that individual differences in SSRT correlated with both activity and connectivity in a distributed inhibition network, comprising prefrontal, premotor, and motor regions. Importantly, this pattern was moderated by age, such that the association between inhibitory control and connectivity, but not activity, differed with age. Multivariate statistics and out-of-sample validation tests of multifactorial functional organization identified differential roles of activity and connectivity in determining an individual's SSRT across the life span. We propose that age-related differences in adaptive cognitive control are best characterized by the joint consideration of multifocal activity and connectivity within distributed brain networks. These insights may facilitate the development of new strategies to support cognitive ability in old age.SIGNIFICANCE STATEMENT The preservation of cognitive and motor control is crucial for maintaining well being across the life span. We show that such control is determined by both activity and connectivity within distributed brain networks. In a large, population-based cohort, we used a novel whole-brain multivariate approach to estimate the functional components of inhibitory control, in terms of their activity and connectivity. Both activity and connectivity in the inhibition network changed with age. But only the association between performance and connectivity, not activity, differed with age. The results suggest that adaptive control is best characterized by the joint consideration of multifocal activity and connectivity. These insights may facilitate the development of new strategies to maintain cognitive ability across the life span in health and disease.

Data-driven re-referencing of intracranial EEG based on independent component analysis (ICA).

BACKGROUND: Intracranial recordings from patients implanted with depth electrodes are a valuable source of information in neuroscience. They allow for the unique opportunity to record brain activity with high spatial and temporal resolution. A common pre-processing choice in stereotactic EEG (S-EEG) is to re-reference the data with a bipolar montage. In this, each channel is subtracted from its neighbor, to reduce commonalities between channels and isolate activity that is spatially confined. NEW METHOD: We challenge the assumption that bipolar reference effectively performs this task. To extract local activity, the distribution of the signal source of interest, interfering distant signals, and noise need to be considered. Referencing schemes with fixed coefficients can decrease the signal to noise ratio (SNR) of the data, they can lead to mislocalization of activity and consequently to misinterpretation of results. We propose to use Independent Component Analysis (ICA), to derive filter coefficients that reflect the statistical dependencies of the data at hand. RESULTS: We describe and demonstrate this on human S-EEG recordings. In a simulation with real data, we quantitatively show that ICA outperforms the bipolar referencing operation in sensitivity and importantly in specificity when revealing local time series from the superposition of neighboring channels. COMPARISON WITH EXISTING METHOD(S): We argue that ICA already performs the same task that bipolar referencing pursues, namely undoing the linear superposition of activity and will identify activity that is local. CONCLUSIONS: When investigating local sources in human S-EEG, ICA should be preferred over re-referencing the data with a bipolar montage.

Problematic internet use as an age-related multifaceted problem: Evidence from a two-site survey.

BACKGROUND AND AIMS: Problematic internet use (PIU; otherwise known as Internet Addiction) is a growing problem in modern societies. There is scarce knowledge of the demographic variables and specific internet activities associated with PIU and a limited understanding of how PIU should be conceptualized. Our aim was to identify specific internet activities associated with PIU and explore the moderating role of age and gender in those associations. METHODS: We recruited 1749 participants aged 18 and above via media advertisements in an Internet-based survey at two sites, one in the US, and one in South Africa; we utilized Lasso regression for the analysis. RESULTS: Specific internet activities were associated with higher problematic internet use scores, including general surfing (lasso ß: 2.1), internet gaming (ß: 0.6), online shopping (ß: 1.4), use of online auction websites (ß: 0.027), social networking (ß: 0.46) and use of online pornography (ß: 1.0). Age moderated the relationship between PIU and role-playing-games (ß: 0.33), online gambling (ß: 0.15), use of auction websites (ß: 0.35) and streaming media (ß: 0.35), with older age associated with higher levels of PIU. There was inconclusive evidence for gender and gender × internet activities being associated with problematic internet use scores. Attention-deficit hyperactivity disorder (ADHD) and social anxiety disorder were associated with high PIU scores in young participants (age ≤ 25, ß: 0.35 and 0.65 respectively), whereas generalized anxiety disorder (GAD) and obsessive-compulsive disorder (OCD) were associated with high PIU scores in the older participants (age > 55, ß: 6.4 and 4.3 respectively). CONCLUSIONS: Many types of online behavior (e.g. shopping, pornography, general surfing) bear a stronger relationship with maladaptive use of the internet than gaming supporting the diagnostic classification of problematic internet use as a multifaceted disorder. Furthermore, internet activities and psychiatric diagnoses associated with problematic internet use vary with age, with public health implications.

Problematic internet use (PIU): Associations with the impulsive-compulsive spectrum. An application of machine learning in psychiatry.

Problematic internet use is common, functionally impairing, and in need of further study. Its relationship with obsessive-compulsive and impulsive disorders is unclear. Our objective was to evaluate whether problematic internet use can be predicted from recognised forms of impulsive and compulsive traits and symptomatology. We recruited volunteers aged 18 and older using media advertisements at two sites (Chicago USA, and Stellenbosch, South Africa) to complete an extensive online survey. State-of-the-art out-of-sample evaluation of machine learning predictive models was used, which included Logistic Regression, Random Forests and Naïve Bayes. Problematic internet use was identified using the Internet Addiction Test (IAT). 2006 complete cases were analysed, of whom 181 (9.0%) had moderate/severe problematic internet use. Using Logistic Regression and Naïve Bayes we produced a classification prediction with a receiver operating characteristic area under the curve (ROC-AUC) of 0.83 (SD 0.03) whereas using a Random Forests algorithm the prediction ROC-AUC was 0.84 (SD 0.03) [all three models superior to baseline models p < 0.0001]. The models showed robust transfer between the study sites in all validation sets [p < 0.0001]. Prediction of problematic internet use was possible using specific measures of impulsivity and compulsivity in a population of volunteers. Moreover, this study offers proof-of-concept in support of using machine learning in psychiatry to demonstrate replicability of results across geographically and culturally distinct settings.

The LDA beamformer: Optimal estimation of ERP source time series using linear discriminant analysis.

We introduce a novel beamforming approach for estimating event-related potential (ERP) source time series based on regularized linear discriminant analysis (LDA). The optimization problems in LDA and linearly-constrained minimum-variance (LCMV) beamformers are formally equivalent. The approaches differ in that, in LCMV beamformers, the spatial patterns are derived from a source model, whereas in an LDA beamformer the spatial patterns are derived directly from the data (i.e., the ERP peak). Using a formal proof and MEG simulations, we show that the LDA beamformer is robust to correlated sources and offers a higher signal-to-noise ratio than the LCMV beamformer and PCA. As an application, we use EEG data from an oddball experiment to show how the LDA beamformer can be harnessed to detect single-trial ERP latencies and estimate connectivity between ERP sources. Concluding, the LDA beamformer optimally reconstructs ERP sources by maximizing the ERP signal-to-noise ratio. Hence, it is a highly suited tool for analyzing ERP source time series, particularly in EEG/MEG studies wherein a source model is not available.

Single-trial analysis of the neural correlates of speech quality perception.

OBJECTIVE: Assessing speech quality perception is a challenge typically addressed in behavioral and opinion-seeking experiments. Only recently, neuroimaging methods were introduced, which were used to study the neural processing of quality at group level. However, our electroencephalography (EEG) studies show that the neural correlates of quality perception are highly individual. Therefore, it became necessary to establish dedicated machine learning methods for decoding subject-specific effects. APPROACH: The effectiveness of our methods is shown by the data of an EEG study that investigates how the quality of spoken vowels is processed neurally. Participants were asked to indicate whether they had perceived a degradation of quality (signal-correlated noise) in vowels, presented in an oddball paradigm. MAIN RESULTS: We find that the P3 amplitude is attenuated with increasing noise. Single-trial analysis allows one to show that this is partly due to an increasing jitter of the P3 component. A novel classification approach helps to detect trials with presumably non-conscious processing at the threshold of perception. We show that this approach uncovers a non-trivial confounder between neural hits and neural misses. SIGNIFICANCE: The combined use of EEG signals and machine learning methods results in a significant 'neural' gain in sensitivity (in processing quality loss) when compared to standard behavioral evaluation; averaged over 11 subjects, this amounts to a relative improvement in sensitivity of 35%.

Online detection of error-related potentials boosts the performance of mental typewriters.

BACKGROUND: Increasing the communication speed of brain-computer interfaces (BCIs) is a major aim of current BCI-research. The idea to automatically detect error-related potentials (ErrPs) in order to veto erroneous decisions of a BCI has been existing for more than one decade, but this approach was so far little investigated in online mode. METHODS: In our study with eleven participants, an ErrP detection mechanism was implemented in an electroencephalography (EEG) based gaze-independent visual speller. RESULTS: Single-trial ErrPs were detected with a mean accuracy of 89.1% (AUC 0.90). The spelling speed was increased on average by 49.0% using ErrP detection. The improvement in spelling speed due to error detection was largest for participants with low spelling accuracy. CONCLUSION: The performance of BCIs can be increased by using an automatic error detection mechanism. The benefit for patients with motor disorders is potentially high since they often have rather low spelling accuracies compared to healthy people.