Predictive regression modeling with MEG/EEG: from source power to signals and cognitive states

Machine learning of brain-specific biomarkers from EEG

BACKGROUND

Electroencephalography (EEG) has a long history as a clinical tool to study brain function, and its potential to derive biomarkers for various applications is far from exhausted. Machine learning (ML) can guide future innovation by harnessing the wealth of complex EEG signals to isolate relevant brain activity. Yet, ML studies in EEG tend to ignore physiological artefacts, which may cause problems for deriving biomarkers specific to the central nervous system (CNS).

METHODS

We present a framework for conceptualising machine learning from CNS versus peripheral signals measured with EEG. A signal representation based on Morlet wavelets allowed us to define traditional brain activity features (e.g. log power) and alternative inputs used by state-of-the-art ML approaches based on covariance matrices. Using more than 2600 EEG recordings from large public databases (TUAB, TDBRAIN), we studied the impact of peripheral signals and artefact removal techniques on ML models in age and sex prediction analyses.

FINDINGS

Across benchmarks, basic artefact rejection improved model performance, whereas further removal of peripheral signals using ICA decreased performance. Our analyses revealed that peripheral signals enable age and sex prediction. However, they explained only a fraction of the performance provided by brain signals.

INTERPRETATION

We show that brain signals and body signals, both present in the EEG, allow for prediction of personal characteristics. While these results may depend on specific applications, our work suggests that great care is needed to separate these signals when the goal is to develop CNS-specific biomarkers using ML.

FUNDING

All authors have been working for F. Hoffmann-La Roche Ltd.

Neurophysiologically Meaningful Motor Imagery EEG Simulation With Applications to Data Augmentation

Motor imagery-based Brain-Computer Interfaces (MI-BCIs) have gained a lot of attention due to their potential usability in neurorehabilitation and neuroprosthetics. However, the accurate recognition of MI patterns in electroencephalography signals (EEG) is hindered by several data-related limitations, which restrict the practical utilization of these systems. Moreover, leveraging deep learning (DL) models for MI decoding is challenged by the difficulty of accessing user-specific MI-EEG data on large scales. Simulated MI-EEG signals can be useful to address these issues, providing well-defined data for the validation of decoding models and serving as a data augmentation approach to improve the training of DL models. While substantial efforts have been dedicated to implementing effective data augmentation strategies and model-based EEG signal generation, the simulation of neurophysiologically plausible EEG-like signals has not yet been exploited in the context of data augmentation. Furthermore, none of the existing approaches have integrated user-specific neurophysiological information during the data generation process. Here, we present PySimMIBCI, a framework for generating realistic MI-EEG signals by integrating neurophysiologically meaningful activity into biophysical forward models. By means of PySimMIBCI, different user capabilities to control an MI-BCI can be simulated and fatigue effects can be included in the generated EEG. Results show that our simulated data closely resemble real data. Moreover, a proposed data augmentation strategy based on our simulated user-specific data significantly outperforms other state-of-the-art augmentation approaches, enhancing DL models performance by up to 15%.

EEG functional connectivity as a Riemannian mediator: An application to malnutrition and cognition

Mediation analysis assesses whether an exposure directly produces changes in cognitive behavior or is influenced by intermediate "mediators". Electroencephalographic (EEG) spectral measurements have been previously used as effective mediators representing diverse aspects of brain function. However, it has been necessary to collapse EEG measures onto a single scalar using standard mediation methods. In this article, we overcome this limitation and examine EEG frequency-resolved functional connectivity measures as a mediator using the full EEG cross-spectral tensor (CST). Since CST samples do not exist in Euclidean space but in the Riemannian manifold of positive-definite tensors, we transform the problem, allowing for the use of classic multivariate statistics. Toward this end, we map the data from the original manifold space to the Euclidean tangent space, eliminating redundant information to conform to a "compressed CST." The resulting object is a matrix with rows corresponding to frequencies and columns to cross spectra between channels. We have developed a novel matrix mediation approach that leverages a nuclear norm regularization to determine the matrix-valued regression parameters. Furthermore, we introduced a global test for the overall CST mediation and a test to determine specific channels and frequencies driving the mediation. We validated the method through simulations and applied it to our well-studied 50+-year Barbados Nutrition Study dataset by comparing EEGs collected in school-age children (5-11 years) who were malnourished in the first year of life with those of healthy classmate controls. We hypothesized that the CST mediates the effect of malnutrition on cognitive performance. We can now explicitly pinpoint the frequencies (delta, theta, alpha, and beta bands) and regions (frontal, central, and occipital) in which functional connectivity was altered in previously malnourished children, an improvement to prior studies. Understanding the specific networks impacted by a history of postnatal malnutrition could pave the way for developing more targeted and personalized therapeutic interventions. Our methods offer a versatile framework applicable to mediation studies encompassing matrix and Hermitian 3D tensor mediators alongside scalar exposures and outcomes, facilitating comprehensive analyses across diverse research domains.

Age-Related Characteristics of Resting-State Electroencephalographic Signals and the Corresponding Analytic Approaches: A Review

The study of the effects of aging on neural activity in the human brain has attracted considerable attention in neurophysiological, neuropsychiatric, and neurocognitive research, as it is directly linked to an understanding of the neural mechanisms underlying the disruption of the brain structures and functions that lead to age-related pathological disorders. Electroencephalographic (EEG) signals recorded during resting-state conditions have been widely used because of the significant advantage of non-invasive signal acquisition with higher temporal resolution. These advantages include the capability of a variety of linear and nonlinear signal analyses and state-of-the-art machine-learning and deep-learning techniques. Advances in artificial intelligence (AI) can not only reveal the neural mechanisms underlying aging but also enable the assessment of brain age reliably by means of the age-related characteristics of EEG signals. This paper reviews the literature on the age-related features, available analytic methods, large-scale resting-state EEG databases, interpretations of the resulting findings, and recent advances in age-related AI models.

Detecting cognitive traits and occupational proficiency using EEG and statistical inference

Machine learning (ML) is widely used in classification tasks aimed at detecting various cognitive states or neurological diseases using noninvasive electroencephalogram (EEG) time series. However, successfully detecting specific cognitive skills in a healthy population, independent of subject, remains challenging. This study compared the subject-independent classification performance of three different pipelines: supervised and Riemann projections with logistic regression and handcrafted power spectral features with light gradient boosting machine (LightGBM). 128-channel EEGs were recorded from 26 healthy volunteers while they solved arithmetic, logical, and verbal tasks. The participants were divided into two groups based on their higher education and occupation: specialists in mathematics and humanities. The balanced accuracy of the education type was significantly above chance for all pipelines: 0.84-0.89, 0.85-0.88, and 0.86-0.88 for each type of task, respectively. All three pipelines allowed us to distinguish mathematical proficiency based on learning experience with different trade-offs between performance and explainability. Our results suggest that ML approaches could also be effective for recognizing individual cognitive traits using EEG.

Sparse Bayesian Learning for End-to-End EEG Decoding

Decoding brain activity from non-invasive electroencephalography (EEG) is crucial for brain-computer interfaces (BCIs) and the study of brain disorders. Notably, end-to-end EEG decoding has gained widespread popularity in recent years owing to the remarkable advances in deep learning research. However, many EEG studies suffer from limited sample sizes, making it difficult for existing deep learning models to effectively generalize to highly noisy EEG data. To address this fundamental limitation, this paper proposes a novel end-to-end EEG decoding algorithm that utilizes a low-rank weight matrix to encode both spatio-temporal filters and the classifier, all optimized under a principled sparse Bayesian learning (SBL) framework. Importantly, this SBL framework also enables us to learn hyperparameters that optimally penalize the model in a Bayesian fashion. The proposed decoding algorithm is systematically benchmarked on five motor imagery BCI EEG datasets ( N=192) and an emotion recognition EEG dataset ( N=45), in comparison with several contemporary algorithms, including end-to-end deep-learning-based EEG decoding algorithms. The classification results demonstrate that our algorithm significantly outperforms the competing algorithms while yielding neurophysiologically meaningful spatio-temporal patterns. Our algorithm therefore advances the state-of-the-art by providing a novel EEG-tailored machine learning tool for decoding brain activity.

STSNet: a novel spatio-temporal-spectral network for subject-independent EEG-based emotion recognition

How to use the characteristics of EEG signals to obtain more complementary and discriminative data representation is an issue in EEG-based emotion recognition. Many studies have tried spatio-temporal or spatio-spectral feature fusion to obtain higher-level representations of EEG data. However, these studies ignored the complementarity between spatial, temporal and spectral domains of EEG signals, thus limiting the classification ability of models. This study proposed an end-to-end network based on ManifoldNet and BiLSTM networks, named STSNet. The STSNet first constructed a 4-D spatio-temporal-spectral data representation and a spatio-temporal data representation based on EEG signals in manifold space. After that, they were fed into the ManifoldNet network and the BiLSTM network respectively to calculate higher-level features and achieve spatio-temporal-spectral feature fusion. Finally, extensive comparative experiments were performed on two public datasets, DEAP and DREAMER, using the subject-independent leave-one-subject-out cross-validation strategy. On the DEAP dataset, the average accuracy of the valence and arousal are 69.38% and 71.88%, respectively; on the DREAMER dataset, the average accuracy of the valence and arousal are 78.26% and 82.37%, respectively. Experimental results show that the STSNet model has good emotion recognition performance.

A supervised data-driven spatial filter denoising method for speech artifacts in intracranial electrophysiological recordings

Neurosurgical procedures that enable direct brain recordings in awake patients offer unique opportunities to explore the neurophysiology of human speech. The scarcity of these opportunities and the altruism of participating patients compel us to apply the highest rigor to signal analysis. Intracranial electroencephalography (iEEG) signals recorded during overt speech can contain a speech artifact that tracks the fundamental frequency (F0) of the participant's voice, involving the same high-gamma frequencies that are modulated during speech production and perception. To address this artifact, we developed a spatial-filtering approach to identify and remove acoustic-induced contaminations of the recorded signal. We found that traditional reference schemes jeopardized signal quality, whereas our data-driven method denoised the recordings while preserving underlying neural activity.

Simulation-based inference for efficient identification of generative models in computational connectomics

Recent advances in connectomics research enable the acquisition of increasing amounts of data about the connectivity patterns of neurons. How can we use this wealth of data to efficiently derive and test hypotheses about the principles underlying these patterns? A common approach is to simulate neuronal networks using a hypothesized wiring rule in a generative model and to compare the resulting synthetic data with empirical data. However, most wiring rules have at least some free parameters, and identifying parameters that reproduce empirical data can be challenging as it often requires manual parameter tuning. Here, we propose to use simulation-based Bayesian inference (SBI) to address this challenge. Rather than optimizing a fixed wiring rule to fit the empirical data, SBI considers many parametrizations of a rule and performs Bayesian inference to identify the parameters that are compatible with the data. It uses simulated data from multiple candidate wiring rule parameters and relies on machine learning methods to estimate a probability distribution (the 'posterior distribution over parameters conditioned on the data') that characterizes all data-compatible parameters. We demonstrate how to apply SBI in computational connectomics by inferring the parameters of wiring rules in an in silico model of the rat barrel cortex, given in vivo connectivity measurements. SBI identifies a wide range of wiring rule parameters that reproduce the measurements. We show how access to the posterior distribution over all data-compatible parameters allows us to analyze their relationship, revealing biologically plausible parameter interactions and enabling experimentally testable predictions. We further show how SBI can be applied to wiring rules at different spatial scales to quantitatively rule out invalid wiring hypotheses. Our approach is applicable to a wide range of generative models used in connectomics, providing a quantitative and efficient way to constrain model parameters with empirical connectivity data.

Repurposing electroencephalogram monitoring of general anaesthesia for building biomarkers of brain ageing: an exploratory study

Background

Electroencephalography (EEG) is increasingly used for monitoring the depth of general anaesthesia, but EEG data from general anaesthesia monitoring are rarely reused for research. Here, we explored repurposing EEG monitoring from general anaesthesia for brain-age modelling using machine learning. We hypothesised that brain age estimated from EEG during general anaesthesia is associated with perioperative risk.

Methods

We reanalysed four-electrode EEGs of 323 patients under stable propofol or sevoflurane anaesthesia to study four EEG signatures (95% of EEG power <8-13 Hz) for age prediction: total power, alpha-band power (8-13 Hz), power spectrum, and spatial patterns in frequency bands. We constructed age-prediction models from EEGs of a healthy reference group (ASA 1 or 2) during propofol anaesthesia. Although all signatures were informative, state-of-the-art age-prediction performance was unlocked by parsing spatial patterns across electrodes along the entire power spectrum (mean absolute error=8.2 yr; R2=0.65).

Results

Clinical exploration in ASA 1 or 2 patients revealed that brain age was positively correlated with intraoperative burst suppression, a risk factor for general anaesthesia complications. Surprisingly, brain age was negatively correlated with burst suppression in patients with higher ASA scores, suggesting hidden confounders. Secondary analyses revealed that age-related EEG signatures were specific to propofol anaesthesia, reflected by limited model generalisation to anaesthesia maintained with sevoflurane.

Conclusions

Although EEG from general anaesthesia may enable state-of-the-art age prediction, differences between anaesthetic drugs can impact the effectiveness and validity of brain-age models. To unleash the dormant potential of EEG monitoring for clinical research, larger datasets from heterogeneous populations with precisely documented drug dosage will be essential.

An extended clinical EEG dataset with 15,300 automatically labelled recordings for pathology decoding

Automated clinical EEG analysis using machine learning (ML) methods is a growing EEG research area. Previous studies on binary EEG pathology decoding have mainly used the Temple University Hospital (TUH) Abnormal EEG Corpus (TUAB) which contains approximately 3,000 manually labelled EEG recordings. To evaluate and eventually even improve the generalisation performance of machine learning methods for EEG pathology, decoding larger, publicly available datasets is required. A number of studies addressed the automatic labelling of large open-source datasets as an approach to create new datasets for EEG pathology decoding, but little is known about the extent to which training on larger, automatically labelled dataset affects decoding performances of established deep neural networks. In this study, we automatically created additional pathology labels for the Temple University Hospital (TUH) EEG Corpus (TUEG) based on the medical reports using a rule-based text classifier. We generated a dataset of 15,300 newly labelled recordings, which we call the TUH Abnormal Expansion EEG Corpus (TUABEX), and which is five times larger than the TUAB. Since the TUABEX contains more pathological (75%) than non-pathological (25%) recordings, we then selected a balanced subset of 8,879 recordings, the TUH Abnormal Expansion Balanced EEG Corpus (TUABEXB). To investigate how training on a larger, automatically labelled dataset affects the decoding performance of deep neural networks, we applied four established deep convolutional neural networks (ConvNets) to the task of pathological versus non-pathological classification and compared the performance of each architecture after training on different datasets. The results show that training on the automatically labelled TUABEXB dataset rather than training on the manually labelled TUAB dataset increases accuracies on TUABEXB and even for TUAB itself for some architectures. We argue that automatically labelling of large open-source datasets can be used to efficiently utilise the massive amount of EEG data stored in clinical archives. We make the proposed TUABEXB available open source and thus offer a new dataset for EEG machine learning research.

Association of sleep and anaesthesia EEG biomarkers with preoperative MoCA score: A pilot study

INTRODUCTION

Preoperative cognitive impairments increase the risk of postoperative complications. The electroencephalogram (EEG) could provide information on cognitive vulnerability. The feasibility and clinical relevance of sleep EEG (EEG_sleep ) compared to intraoperative EEG (EEG_intraop ) in cognitive risk stratification remains to be explored. We investigated similarities between EEG_sleep and EEG_intraop vis-a-vis preoperative cognitive impairments.

METHODS

Pilot study including 27 patients (63 year old [53.5, 70.0]) to whom Montreal cognitive assessment (MoCA) and EEG_sleep were administered 1 day before a propofol-based general anaesthesia, in addition to EEG_intraop acquisition from depth-of-anaesthesia monitors. Sleep spindles on EEG_sleep and intraoperative alpha-band power on EEG_intraop were particularly explored.

RESULTS

In total, 11 (41%) patients had a MoCA <25 points. These patients had a significantly lower sleep spindle power on EEG_sleep (25 vs. 40 μv² /Hz, p = .035) and had a weaker intraoperative alpha-band power on EEG_intraop (85 vs. 150 μv² /Hz, p = .001) compared to patients with normal MoCA. Correlation between sleep spindle and intraoperative alpha-band power was positive and significant (r = 0.544, p = .003).

CONCLUSION

Preoperative cognitive impairment appears to be detectable by both EEG_sleep and EEG_intraop . Preoperative sleep EEG to assess perioperative cognitive risk is feasible but more data are needed to demonstrate its benefit compared to intraoperative EEG.

Dynamic flexibility and controllability of network communities in juvenile myoclonic epilepsy

Juvenile myoclonic epilepsy (JME) is the most common syndrome within the idiopathic generalized epilepsy spectrum, manifested by myoclonic and generalized tonic-clonic seizures and spike-and-wave discharges (SWDs) on electroencephalography (EEG). Currently, the pathophysiological concepts addressing SWD generation in JME are still incomplete. In this work, we characterize the temporal and spatial organization of functional networks and their dynamic properties as derived from high-density EEG (hdEEG) recordings and MRI in 40 JME patients (25.4 ± 7.6 years, 25 females). The adopted approach allows for the construction of a precise dynamic model of ictal transformation in JME at the cortical and deep brain nuclei source levels. We implement Louvain algorithm to attribute brain regions with similar topological properties to modules during separate time windows before and during SWD generation. Afterwards, we quantify how modular assignments evolve and steer through different states towards the ictal state by measuring characteristics of flexibility and controllability. We find antagonistic dynamics of flexibility and controllability within network modules as they evolve towards and undergo ictal transformation. Prior to SWD generation, we observe concomitantly increasing flexibility (F(1,39) = 25.3, corrected p < 0.001) and decreasing controllability (F(1,39) = 55.3, p < 0.001) within the fronto-parietal module in γ-band. On a step further, during interictal SWDs as compared to preceding time windows, we notice decreasing flexibility (F(1,39) = 11.9, p < 0.001) and increasing controllability (F(1,39) = 10.1, p < 0.001) within the fronto-temporal module in γ-band. During ictal SWDs as compared to prior time windows, we demonstrate significantly decreasing flexibility (F(1,14) = 31.6; p < 0.001) and increasing controllability (F(1,14) = 44.7, p < 0.001) within the basal ganglia module. Furthermore, we show that flexibility and controllability within the fronto-temporal module of the interictal SWDs relate to seizure frequency and cognitive performance in JME patients. Our results demonstrate that detection of network modules and quantification of their dynamic properties is relevant to track the generation of SWDs. The observed flexibility and controllability dynamics reflect the reorganization of de-/synchronized connections and the ability of evolving network modules to reach a seizure-free state, respectively. These findings may advance the elaboration of network-based biomarkers and more targeted therapeutic neuromodulatory approaches in JME.

Revisiting Riemannian geometry-based EEG decoding through approximate joint diagonalization

Objective.The wider adoption of Riemannian geometry in electroencephalography (EEG) processing is hindered by two factors: (a) it involves the manipulation of complex mathematical formulations and, (b) it leads to computationally demanding tasks. The main scope of this work is to simplify particular notions of Riemannian geometry and provide an efficient and comprehensible scheme for neuroscientific explorations.Approach.To overcome the aforementioned shortcomings, we exploit the concept of approximate joint diagonalization in order to reconstruct the spatial covariance matrices assuming the existence of (and identifying) a common eigenspace in which the application of Riemannian geometry is significantly simplified.Main results.The employed reconstruction process abides to physiologically plausible assumptions, reduces the computational complexity in Riemannian geometry schemes and bridges the gap between rigorous mathematical procedures and computational neuroscience. Our approach is both formally established and experimentally validated by employing real and synthetic EEG data.Significance.The implications of the introduced reconstruction process are highlighted by reformulating and re-introducing two signal processing methodologies, namely the 'Symmetric Positive Definite (SPD) Matrix Quantization' and the 'Coding over SPD Atoms'. The presented approach paves the way for robust and efficient neuroscientific explorations that exploit Riemannian geometry schemes.

Decoding continuous variables from event-related potential (ERP) data with linear support vector regression using the Decision Decoding Toolbox (DDTBOX)

Multivariate classification analysis for event-related potential (ERP) data is a powerful tool for predicting cognitive variables. However, classification is often restricted to categorical variables and under-utilises continuous data, such as response times, response force, or subjective ratings. An alternative approach is support vector regression (SVR), which uses single-trial data to predict continuous variables of interest. In this tutorial-style paper, we demonstrate how SVR is implemented in the Decision Decoding Toolbox (DDTBOX). To illustrate in more detail how results depend on specific toolbox settings and data features, we report results from two simulation studies resembling real EEG data, and one real ERP-data set, in which we predicted continuous variables across a range of analysis parameters. Across all studies, we demonstrate that SVR is effective for analysis windows ranging from 2 to 100 ms, and relatively unaffected by temporal averaging. Prediction is still successful when only a small number of channels encode true information, and the analysis is robust to temporal jittering of the relevant information in the signal. Our results show that SVR as implemented in DDTBOX can reliably predict continuous, more nuanced variables, which may not be well-captured by classification analysis. In sum, we demonstrate that linear SVR is a powerful tool for the investigation of single-trial EEG data in relation to continuous variables, and we provide practical guidance for users.

Wavelet Based Filters for Artifact Elimination in Electroencephalography Signal: A Review

Electroencephalography (EEG) is a diagnostic test that records and measures the electrical activity of the human brain. Research investigating human behaviors and conditions using EEG has increased from year to year. Therefore, an efficient approach is vital to process the EEG dataset to improve the output signal quality. The wavelet is one of the well-known approaches for processing the EEG signal in time-frequency domain analysis. The wavelet is better than the traditional Fourier Transform because it has good time-frequency localized properties and multi-resolution analysis where the transient information of an EEG signal can be extracted efficiently. Thus, this review article aims to comprehensively describe the application of the wavelet method in denoising the EEG signal based on recent research. This review begins with a brief overview of the basic theory and characteristics of EEG and the wavelet transform method. Then, several wavelet-based methods commonly applied in EEG dataset denoising are described and a considerable number of the latest published EEG research works with wavelet applications are reviewed. Besides, the challenges that exist in current EEG-based wavelet method research are discussed. Finally, alternative solutions to mitigate the issues are recommended.

A framework to analyze opinion formation models

Comparing model predictions with real data is crucial to improve and validate a model. For opinion formation models, validation based on real data is uncommon and difficult to obtain, also due to the lack of systematic approaches for a meaningful comparison. We introduce a framework to assess opinion formation models, which can be used to determine the qualitative outcomes that an opinion formation model can produce, and compare model predictions with real data. The proposed approach relies on a histogram-based classification algorithm, and on transition tables. The algorithm classifies an opinion distribution as perfect consensus, consensus, polarization, clustering, or dissensus; these qualitative categories were identified from World Values Survey data. The transition tables capture the qualitative evolution of the opinion distribution between an initial and a final time. We compute the real transition tables based on World Values Survey data from different years, as well as the predicted transition tables produced by the French-DeGroot, Weighted-Median, Bounded Confidence, and Quantum Game models, and we compare them. Our results provide insight into the evolution of real-life opinions and highlight key directions to improve opinion formation models.

Late combination shows that MEG adds to MRI in classifying MCI versus controls

Early detection of Alzheimer's disease (AD) is essential for developing effective treatments. Neuroimaging techniques like Magnetic Resonance Imaging (MRI) have the potential to detect brain changes before symptoms emerge. Structural MRI can detect atrophy related to AD, but it is possible that functional changes are observed even earlier. We therefore examined the potential of Magnetoencephalography (MEG) to detect differences in functional brain activity in people with Mild Cognitive Impairment (MCI) - a state at risk of early AD. We introduce a framework for multimodal combination to ask whether MEG data from a resting-state provides complementary information beyond structural MRI data in the classification of MCI versus controls. More specifically, we used multi-kernel learning of support vector machines to classify 163 MCI cases versus 144 healthy elderly controls from the BioFIND dataset. When using the covariance of planar gradiometer data in the low Gamma range (30-48 Hz), we found that adding a MEG kernel improved classification accuracy above kernels that captured several potential confounds (e.g., age, education, time-of-day, head motion). However, accuracy using MEG alone (68%) was worse than MRI alone (71%). When simply concatenating (normalized) features from MEG and MRI into one kernel (Early combination), there was no advantage of combining MEG with MRI versus MRI alone. When combining kernels of modality-specific features (Intermediate combination), there was an improvement in multimodal classification to 74%. The biggest multimodal improvement however occurred when we combined kernels from the predictions of modality-specific classifiers (Late combination), which achieved 77% accuracy (a reliable improvement in terms of permutation testing). We also explored other MEG features, such as the variance versus covariance of magnetometer versus planar gradiometer data within each of 6 frequency bands (delta, theta, alpha, beta, low gamma, or high gamma), and found that they generally provided complementary information for classification above MRI. We conclude that MEG can improve on the MRI-based classification of MCI.

Harmonized-Multinational qEEG norms (HarMNqEEG)

This paper extends frequency domain quantitative electroencephalography (qEEG) methods pursuing higher sensitivity to detect Brain Developmental Disorders. Prior qEEG work lacked integration of cross-spectral information omitting important functional connectivity descriptors. Lack of geographical diversity precluded accounting for site-specific variance, increasing qEEG nuisance variance. We ameliorate these weaknesses. (i) Create lifespan Riemannian multinational qEEG norms for cross-spectral tensors. These norms result from the HarMNqEEG project fostered by the Global Brain Consortium. We calculate the norms with data from 9 countries, 12 devices, and 14 studies, including 1564 subjects. Instead of raw data, only anonymized metadata and EEG cross-spectral tensors were shared. After visual and automatic quality control, developmental equations for the mean and standard deviation of qEEG traditional and Riemannian DPs were calculated using additive mixed-effects models. We demonstrate qEEG "batch effects" and provide methods to calculate harmonized z-scores. (ii) We also show that harmonized Riemannian norms produce z-scores with increased diagnostic accuracy predicting brain dysfunction produced by malnutrition in the first year of life and detecting COVID induced brain dysfunction. (iii) We offer open code and data to calculate different individual z-scores from the HarMNqEEG dataset. These results contribute to developing bias-free, low-cost neuroimaging technologies applicable in various health settings.

Feel Your Reach: An EEG-Based Framework to Continuously Detect Goal-Directed Movements and Error Processing to Gate Kinesthetic Feedback Informed Artificial Arm Control

Establishing the basic knowledge, methodology, and technology for a framework for the continuous decoding of hand/arm movement intention was the aim of the ERC-funded project "Feel Your Reach". In this work, we review the studies and methods we performed and implemented in the last 6 years, which build the basis for enabling severely paralyzed people to non-invasively control a robotic arm in real-time from electroencephalogram (EEG). In detail, we investigated goal-directed movement detection, decoding of executed and attempted movement trajectories, grasping correlates, error processing, and kinesthetic feedback. Although we have tested some of our approaches already with the target populations, we still need to transfer the "Feel Your Reach" framework to people with cervical spinal cord injury and evaluate the decoders' performance while participants attempt to perform upper-limb movements. While on the one hand, we made major progress towards this ambitious goal, we also critically discuss current limitations.

A practical guide to applying machine learning to infant EEG data

Electroencephalography (EEG) has been widely adopted by the developmental cognitive neuroscience community, but the application of machine learning (ML) in this domain lags behind adult EEG studies. Applying ML to infant data is particularly challenging due to the low number of trials, low signal-to-noise ratio, high inter-subject variability, and high inter-trial variability. Here, we provide a step-by-step tutorial on how to apply ML to classify cognitive states in infants. We describe the type of brain attributes that are widely used for EEG classification and also introduce a Riemannian geometry based approach for deriving connectivity estimates that account for inter-trial and inter-subject variability. We present pipelines for learning classifiers using trials from a single infant and from multiple infants, and demonstrate the application of these pipelines on a standard infant EEG dataset of forty 12-month-old infants collected under an auditory oddball paradigm. While we classify perceptual states induced by frequent versus rare stimuli, the presented pipelines can be easily adapted for other experimental designs and stimuli using the associated code that we have made publicly available.

Functional connectivity ensemble method to enhance BCI performance (FUCONE)

OBJECTIVE

Relying on the idea that functional connectivity provides important insights on the underlying dynamic of neuronal interactions, we propose a novel framework that combines functional connectivity estimators and covariance-based pipelines to improve the classification of mental states, such as motor imagery.

METHODS

A Riemannian classifier is trained for each estimator and an ensemble classifier combines the decisions in each feature space. A thorough assessment of the functional connectivity estimators is provided and the best performing pipeline among those tested, called FUCONE, is evaluated on different conditions and datasets.

RESULTS

Using a meta-analysis to aggregate results across datasets, FUCONE performed significantly better than all state-of-the-art methods.

CONCLUSION

The performance gain is mostly imputable to the improved diversity of the feature spaces, increasing the robustness of the ensemble classifier with respect to the inter- and intra-subject variability.

SIGNIFICANCE

Our results offer new insights into the need to consider functional connectivity-based methods to improve the BCI performance.

Robust learning from corrupted EEG with dynamic spatial filtering

Building machine learning models using EEG recorded outside of the laboratory setting requires methods robust to noisy data and randomly missing channels. This need is particularly great when working with sparse EEG montages (1-6 channels), often encountered in consumer-grade or mobile EEG devices. Neither classical machine learning models nor deep neural networks trained end-to-end on EEG are typically designed or tested for robustness to corruption, and especially to randomly missing channels. While some studies have proposed strategies for using data with missing channels, these approaches are not practical when sparse montages are used and computing power is limited (e.g., wearables, cell phones). To tackle this problem, we propose dynamic spatial filtering (DSF), a multi-head attention module that can be plugged in before the first layer of a neural network to handle missing EEG channels by learning to focus on good channels and to ignore bad ones. We tested DSF on public EEG data encompassing ∼4,000 recordings with simulated channel corruption and on a private dataset of ∼100 at-home recordings of mobile EEG with natural corruption. Our proposed approach achieves the same performance as baseline models when no noise is applied, but outperforms baselines by as much as 29.4% accuracy when significant channel corruption is present. Moreover, DSF outputs are interpretable, making it possible to monitor the effective channel importance in real-time. This approach has the potential to enable the analysis of EEG in challenging settings where channel corruption hampers the reading of brain signals.

Functional connectivity using high density EEG shows competitive reliability and agreement across test/retest sessions

BACKGROUND

Electrophysiological resting state functional connectivity using high density electroencephalography (hdEEG) is gaining momentum. The increased resolution offered by hdEEG, usually either 128 or 256 channels, permits source localization of EEG signals on the cortical surface. However, the number of methodological options for the acquisition and analysis of resting state hdEEG is extremely large. These include acquisition duration, eyes open/closed, channel density, source localization methods, and functional connectivity metric.

NEW METHODS

We undertake an extensive examination of the test-retest reliability and methodological agreement of all these options for regional measures of functional connectivity.

RESULTS

Power envelope connectivity shows larger test-retest reliability than imaginary coherence across all bands. While channel density doesn't strongly impact reliability or agreement, source localization methods produce systematically different functional connectivity, highlighting an important obstacle for replicating results in the literature. Most importantly, reliability and agreement often plateaus at or after 6 minutes of acquisition, well beyond the typical duration of 3 minutes. Finally, our study demonstrates that resting EEG can be as or more reliable than resting fMRI acquired in the same individuals.

CONCLUSIONS

The competitive reliability and agreement of power envelope connectivity greatly increases our confidence in measuring resting state connectivity using EEG and its capacity to find individual differences.

A tutorial on generalized eigendecomposition for denoising, contrast enhancement, and dimension reduction in multichannel electrophysiology

The goal of this paper is to present a theoretical and practical introduction to generalized eigendecomposition (GED), which is a robust and flexible framework used for dimension reduction and source separation in multichannel signal processing. In cognitive electrophysiology, GED is used to create spatial filters that maximize a researcher-specified contrast. For example, one may wish to exploit an assumption that different sources have different frequency content, or that sources vary in magnitude across experimental conditions. GED is fast and easy to compute, performs well in simulated and real data, and is easily adaptable to a variety of specific research goals. This paper introduces GED in a way that ties together myriad individual publications and applications of GED in electrophysiology, and provides sample MATLAB and Python code that can be tested and adapted. Practical considerations and issues that often arise in applications are discussed.

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

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

Decoding subjective emotional arousal from EEG during an immersive virtual reality experience

Immersive virtual reality (VR) enables naturalistic neuroscientific studies while maintaining experimental control, but dynamic and interactive stimuli pose methodological challenges. We here probed the link between emotional arousal, a fundamental property of affective experience, and parieto-occipital alpha power under naturalistic stimulation: 37 young healthy adults completed an immersive VR experience, which included rollercoaster rides, while their EEG was recorded. They then continuously rated their subjective emotional arousal while viewing a replay of their experience. The association between emotional arousal and parieto-occipital alpha power was tested and confirmed by (1) decomposing the continuous EEG signal while maximizing the comodulation between alpha power and arousal ratings and by (2) decoding periods of high and low arousal with discriminative common spatial patterns and a long short-term memory recurrent neural network. We successfully combine EEG and a naturalistic immersive VR experience to extend previous findings on the neurophysiology of emotional arousal towards real-world neuroscience.

Fatigue Detection with Spatial-Temporal Fusion Method on Covariance Manifolds of Electroencephalography

With the increasing pressure of current life, fatigue caused by high-pressure work has deeply affected people and even threatened their lives. In particular, fatigue driving has become a leading cause of traffic accidents and deaths. This paper investigates electroencephalography (EEG)-based fatigue detection for driving by mining the latent information through the spatial-temporal changes in the relations between EEG channels. First, EEG data are partitioned into several segments to calculate the covariance matrices of each segment, and then we feed these matrices into a recurrent neural network to obtain high-level temporal information. Second, the covariance matrices of whole signals are leveraged to extract two kinds of spatial features, which will be fused with temporal characteristics to obtain comprehensive spatial-temporal information. Experiments on an open benchmark showed that our method achieved an excellent classification accuracy of 93.834% and performed better than several novel methods. These experimental results indicate that our method enables better reliability and feasibility in the detection of fatigued driving.

Toward Studying Cognition in a Dish

Bioengineered neural tissues help advance our understanding of neurodevelopment, regeneration, and neural disease; however, it remains unclear whether they can replicate higher-order functions including cognition. Building upon technical achievements in the fields of biomaterials, tissue engineering, and cell biology, investigators have generated an assortment of artificial brain structures and cocultured circuits. Though they have displayed basic electrochemical signaling, their capacities to generate minimal patterns of information processing suggestive of high-order cognitive analogues have not yet been explored. Here, we review the current state of neural tissue engineering and consider the possibility of a study of cognition in vitro. We adopt a practical definition of minimal cognition, anticipate problems of measurement, and discuss solutions toward a study of cognition in a dish.

Heterogeneity and Classification of Recent Onset Psychosis and Depression: A Multimodal Machine Learning Approach

Diagnostic heterogeneity within and across psychotic and affective disorders challenges accurate treatment selection, particularly in the early stages. Delineation of shared and distinct illness features at the phenotypic and brain levels may inform the development of more precise differential diagnostic tools. We aimed to identify prototypes of depression and psychosis to investigate their heterogeneity, with common, comorbid transdiagnostic symptoms. Analyzing clinical/neurocognitive and grey matter volume (GMV) data from the PRONIA database, we generated prototypic models of recent-onset depression (ROD) vs. recent-onset psychosis (ROP) by training support-vector machines to separate patients with ROD from patients with ROP, who were selected for absent comorbid features (pure groups). Then, models were applied to patients with comorbidity, ie, ROP with depressive symptoms (ROP+D) and ROD participants with sub-threshold psychosis-like features (ROD+P), to measure their positions within the affective-psychotic continuum. All models were independently validated in a replication sample. Comorbid patients were positioned between pure groups, with ROP+D patients being more frequently classified as ROD compared to pure ROP patients (clinical/neurocognitive model: χ2 = 14.874; P < .001; GMV model: χ2 = 4.933; P = .026). ROD+P patient classification did not differ from ROD (clinical/neurocognitive model: χ2 = 1.956; P = 0.162; GMV model: χ2 = 0.005; P = .943). Clinical/neurocognitive and neuroanatomical models demonstrated separability of prototypic depression from psychosis. The shift of comorbid patients toward the depression prototype, observed at the clinical and biological levels, suggests that psychosis with affective comorbidity aligns more strongly to depressive rather than psychotic disease processes. Future studies should assess how these quantitative measures of comorbidity predict outcomes and individual responses to stratified therapeutic interventions.

Decoding and interpreting cortical signals with a compact convolutional neural network

OBJECTIVE

Brain-computer interfaces (BCIs) decode information from neural activity and send it to external devices. The use of Deep Learning approaches for decoding allows for automatic feature engineering within the specific decoding task. Physiologically plausible interpretation of the network parameters ensures the robustness of the learned decision rules and opens the exciting opportunity for automatic knowledge discovery.

APPROACH

We describe a compact convolutional network-based architecture for adaptive decoding of electrocorticographic (ECoG) data into finger kinematics. We also propose a novel theoretically justified approach to interpreting the spatial and temporal weights in the architectures that combine adaptation in both space and time. The obtained spatial and frequency patterns characterizing the neuronal populations pivotal to the specific decoding task can then be interpreted by fitting appropriate spatial and dynamical models.

MAIN RESULTS

We first tested our solution using realistic Monte-Carlo simulations. Then, when applied to the ECoG data from Berlin BCI competition IV dataset, our architecture performed comparably to the competition winners without requiring explicit feature engineering. Using the proposed approach to the network weights interpretation we could unravel the spatial and the spectral patterns of the neuronal processes underlying the successful decoding of finger kinematics from an ECoG dataset. Finally we have also applied the entire pipeline to the analysis of a 32-channel EEG motor-imagery dataset and observed physiologically plausible patterns specific to the task.

SIGNIFICANCE

We described a compact and interpretable CNN architecture derived from the basic principles and encompassing the knowledge in the field of neural electrophysiology. For the first time in the context of such multibranch architectures with factorized spatial and temporal processing we presented theoretically justified weights interpretation rules. We verified our recipes using simulations and real data and demonstrated that the proposed solution offers a good decoder and a tool for investigating motor control neural mechanisms.

Propofol Requirement and EEG Alpha Band Power During General Anesthesia Provide Complementary Views on Preoperative Cognitive Decline

Background: Although cognitive decline (CD) is associated with increased post-operative morbidity and mortality, routinely screening patients remains difficult. The main objective of this prospective study is to use the EEG response to a Propofol-based general anesthesia (GA) to reveal CD.

Methods: 42 patients with collected EEG and Propofol target concentration infusion (TCI) during GA had a preoperative cognitive assessment using MoCA. We evaluated the performance of three variables to detect CD (MoCA < 25 points): age, Propofol requirement to induce unconsciousness (TCI at SEF₉₅: 8–13 Hz) and the frontal alpha band power (AP at SEF₉₅: 8–13 Hz).

Results: The 17 patients (40%) with CD were significantly older (p < 0.001), had lower TCI (p < 0.001), and AP (p < 0.001). We found using logistic models that TCI and AP were the best set of variables associated with CD (AUC: 0.89) and performed better than age (p < 0.05). Propofol TCI had a greater impact on CD probability compared to AP, although both were complementary in detecting CD.

Conclusion: TCI and AP contribute additively to reveal patient with preoperative cognitive decline. Further research on post-operative cognitive trajectory are necessary to confirm the interest of intra operative variables in addition or as a substitute to cognitive evaluation.

Uncovering the structure of clinical EEG signals with self-supervised learning

OBJECTIVE

Supervised learning paradigms are often limited by the amount of labeled data that is available. This phenomenon is particularly problematic in clinically-relevant data, such as electroencephalography (EEG), where labeling can be costly in terms of specialized expertise and human processing time. Consequently, deep learning architectures designed to learn on EEG data have yielded relatively shallow models and performances at best similar to those of traditional feature-based approaches. However, in most situations, unlabeled data is available in abundance. By extracting information from this unlabeled data, it might be possible to reach competitive performance with deep neural networks despite limited access to labels.

APPROACH

We investigated self-supervised learning (SSL), a promising technique for discovering structure in unlabeled data, to learn representations of EEG signals. Specifically, we explored two tasks based on temporal context prediction as well as contrastive predictive coding on two clinically-relevant problems: EEG-based sleep staging and pathology detection. We conducted experiments on two large public datasets with thousands of recordings and performed baseline comparisons with purely supervised and hand-engineered approaches.

MAIN RESULTS

Linear classifiers trained on SSL-learned features consistently outperformed purely supervised deep neural networks in low-labeled data regimes while reaching competitive performance when all labels were available. Additionally, the embeddings learned with each method revealed clear latent structures related to physiological and clinical phenomena, such as age effects.

SIGNIFICANCE

We demonstrate the benefit of SSL approaches on EEG data. Our results suggest that self-supervision may pave the way to a wider use of deep learning models on EEG data.

Combining magnetoencephalography with magnetic resonance imaging enhances learning of surrogate-biomarkers

Electrophysiological methods, that is M/EEG, provide unique views into brain health. Yet, when building predictive models from brain data, it is often unclear how electrophysiology should be combined with other neuroimaging methods. Information can be redundant, useful common representations of multimodal data may not be obvious and multimodal data collection can be medically contraindicated, which reduces applicability. Here, we propose a multimodal model to robustly combine MEG, MRI and fMRI for prediction. We focus on age prediction as a surrogate biomarker in 674 subjects from the Cam-CAN dataset. Strikingly, MEG, fMRI and MRI showed additive effects supporting distinct brain-behavior associations. Moreover, the contribution of MEG was best explained by cortical power spectra between 8 and 30 Hz. Finally, we demonstrate that the model preserves benefits of stacking when some data is missing. The proposed framework, hence, enables multimodal learning for a wide range of biomarkers from diverse types of brain signals.