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 ∼4000 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.

A reusable benchmark of brain-age prediction from M/EEG resting-state signals.

Population-level modeling can define quantitative measures of individual aging by applying machine learning to large volumes of brain images. These measures of brain age, obtained from the general population, helped characterize disease severity in neurological populations, improving estimates of diagnosis or prognosis. Magnetoencephalography (MEG) and Electroencephalography (EEG) have the potential to further generalize this approach towards prevention and public health by enabling assessments of brain health at large scales in socioeconomically diverse environments. However, more research is needed to define methods that can handle the complexity and diversity of M/EEG signals across diverse real-world contexts. To catalyse this effort, here we propose reusable benchmarks of competing machine learning approaches for brain age modeling. We benchmarked popular classical machine learning pipelines and deep learning architectures previously used for pathology decoding or brain age estimation in 4 international M/EEG cohorts from diverse countries and cultural contexts, including recordings from more than 2500 participants. Our benchmarks were built on top of the M/EEG adaptations of the BIDS standard, providing tools that can be applied with minimal modification on any M/EEG dataset provided in the BIDS format. Our results suggest that, regardless of whether classical machine learning or deep learning was used, the highest performance was reached by pipelines and architectures involving spatially aware representations of the M/EEG signals, leading to R2 scores between 0.60-0.74. Hand-crafted features paired with random forest regression provided robust benchmarks even in situations in which other approaches failed. Taken together, this set of benchmarks, accompanied by open-source software and high-level Python scripts, can serve as a starting point and quantitative reference for future efforts at developing M/EEG-based measures of brain aging. The generality of the approach renders this benchmark reusable for other related objectives such as modeling specific cognitive variables or clinical endpoints.

Brain age as a surrogate marker for cognitive performance in multiple sclerosis.

BACKGROUND AND PURPOSE: Data from neuro-imaging techniques allow us to estimate a brain's age. Brain age is easily interpretable as 'how old the brain looks' and could therefore be an attractive communication tool for brain health in clinical practice. This study aimed to investigate its clinical utility by investigating the relationship between brain age and cognitive performance in multiple sclerosis (MS). METHODS: A linear regression model was trained to predict age from brain magnetic resonance imaging volumetric features and sex in a healthy control dataset (HC_train, n = 1673). This model was used to predict brain age in two test sets: HC_test (n = 50) and MS_test (n = 201). Brain-predicted age difference (BPAD) was calculated as BPAD = brain age minus chronological age. Cognitive performance was assessed by the Symbol Digit Modalities Test (SDMT). RESULTS: Brain age was significantly related to SDMT scores in the MS_test dataset (r = -0.46, p < 0.001) and contributed uniquely to variance in SDMT beyond chronological age, reflected by a significant correlation between BPAD and SDMT (r = -0.24, p < 0.001) and a significant weight (-0.25, p = 0.002) in a multivariate regression equation with age. CONCLUSIONS: Brain age is a candidate biomarker for cognitive dysfunction in MS and an easy to grasp metric for brain health.

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

Predicting biomedical outcomes from Magnetoencephalography and Electroencephalography (M/EEG) is central to applications like decoding, brain-computer-interfaces (BCI) or biomarker development and is facilitated by supervised machine learning. Yet, most of the literature is concerned with classification of outcomes defined at the event-level. Here, we focus on predicting continuous outcomes from M/EEG signal defined at the subject-level, and analyze about 600 MEG recordings from Cam-CAN dataset and about 1000 EEG recordings from TUH dataset. Considering different generative mechanisms for M/EEG signals and the biomedical outcome, we propose statistically-consistent predictive models that avoid source-reconstruction based on the covariance as representation. Our mathematical analysis and ground-truth simulations demonstrated that consistent function approximation can be obtained with supervised spatial filtering or by embedding with Riemannian geometry. Additional simulations revealed that Riemannian methods were more robust to model violations, in particular geometric distortions induced by individual anatomy. To estimate the relative contribution of brain dynamics and anatomy to prediction performance, we propose a novel model inspection procedure based on biophysical forward modeling. Applied to prediction of outcomes at the subject-level, the analysis revealed that the Riemannian model better exploited anatomical information while sensitivity to brain dynamics was similar across methods. We then probed the robustness of the models across different data cleaning options. Environmental denoising was globally important but Riemannian models were strikingly robust and continued performing well even without preprocessing. Our results suggest each method has its niche: supervised spatial filtering is practical for event-level prediction while the Riemannian model may enable simple end-to-end learning.

Emergence of ß and γ networks following multisensory training.

Our perceptual reality relies on inferences about the causal structure of the world given by multiple sensory inputs. In ecological settings, multisensory events that cohere in time and space benefit inferential processes: hearing and seeing a speaker enhances speech comprehension, and the acoustic changes of flapping wings naturally pace the motion of a flock of birds. Here, we asked how a few minutes of (multi)sensory training could shape cortical interactions in a subsequent unisensory perceptual task. For this, we investigated oscillatory activity and functional connectivity as a function of individuals' sensory history during training. Human participants performed a visual motion coherence discrimination task while being recorded with magnetoencephalography. Three groups of participants performed the same task with visual stimuli only, while listening to acoustic textures temporally comodulated with the strength of visual motion coherence, or with auditory noise uncorrelated with visual motion. The functional connectivity patterns before and after training were contrasted to resting-state networks to assess the variability of common task-relevant networks, and the emergence of new functional interactions as a function of sensory history. One major finding is the emergence of a large-scale synchronization in the high γ (gamma: 60-120Hz) and ß (beta: 15-30Hz) bands for individuals who underwent comodulated multisensory training. The post-training network involved prefrontal, parietal, and visual cortices. Our results suggest that the integration of evidence and decision-making strategies become more efficient following congruent multisensory training through plasticity in network routing and oscillatory regimes.

Robust EEG-based cross-site and cross-protocol classification of states of consciousness.

Determining the state of consciousness in patients with disorders of consciousness is a challenging practical and theoretical problem. Recent findings suggest that multiple markers of brain activity extracted from the EEG may index the state of consciousness in the human brain. Furthermore, machine learning has been found to optimize their capacity to discriminate different states of consciousness in clinical practice. However, it is unknown how dependable these EEG markers are in the face of signal variability because of different EEG configurations, EEG protocols and subpopulations from different centres encountered in practice. In this study we analysed 327 recordings of patients with disorders of consciousness (148 unresponsive wakefulness syndrome and 179 minimally conscious state) and 66 healthy controls obtained in two independent research centres (Paris Pitié-Salpêtrière and Liège). We first show that a non-parametric classifier based on ensembles of decision trees provides robust out-of-sample performance on unseen data with a predictive area under the curve (AUC) of ~0.77 that was only marginally affected when using alternative EEG configurations (different numbers and positions of sensors, numbers of epochs, average AUC = 0.750 ± 0.014). In a second step, we observed that classifiers based on multiple as well as single EEG features generalize to recordings obtained from different patient cohorts, EEG protocols and different centres. However, the multivariate model always performed best with a predictive AUC of 0.73 for generalization from Paris 1 to Paris 2 datasets, and an AUC of 0.78 from Paris to Liège datasets. Using simulations, we subsequently demonstrate that multivariate pattern classification has a decisive performance advantage over univariate classification as the stability of EEG features decreases, as different EEG configurations are used for feature-extraction or as noise is added. Moreover, we show that the generalization performance from Paris to Liège remains stable even if up to 20% of the diagnostic labels are randomly flipped. Finally, consistent with recent literature, analysis of the learned decision rules of our classifier suggested that markers related to dynamic fluctuations in theta and alpha frequency bands carried independent information and were most influential. Our findings demonstrate that EEG markers of consciousness can be reliably, economically and automatically identified with machine learning in various clinical and acquisition contexts.

Brain-heart interactions reveal consciousness in noncommunicating patients.

OBJECTIVE: We here aimed at characterizing heart-brain interactions in patients with disorders of consciousness. We tested how this information impacts data-driven classification between unresponsive and minimally conscious patients. METHODS: A cohort of 127 patients in vegetative state/unresponsive wakefulness syndrome (VS/UWS; n = 70) and minimally conscious state (MCS; n = 57) were presented with the local-global auditory oddball paradigm, which distinguishes 2 levels of processing: short-term deviation of local auditory regularities and global long-term rule violations. In addition to previously validated markers of consciousness extracted from electroencephalograms (EEG), we computed autonomic cardiac markers, such as heart rate (HR) and HR variability (HRV), and cardiac cycle phase shifts triggered by the processing of the auditory stimuli. RESULTS: HR and HRV were similar in patients across groups. The cardiac cycle was not sensitive to the processing of local regularities in either the VS/UWS or MCS patients. In contrast, global regularities induced a phase shift of the cardiac cycle exclusively in the MCS group. The interval between the auditory stimulation and the following R peak was significantly shortened in MCS when the auditory rule was violated. When the information for the cardiac cycle modulations and other consciousness-related EEG markers were combined, single patient classification performance was enhanced compared to classification with solely EEG markers. INTERPRETATION: Our work shows a link between residual cognitive processing and the modulation of autonomic somatic markers. These results open a new window to evaluate patients with disorders of consciousness via the embodied paradigm, according to which body-brain functions contribute to a holistic approach to conscious processing. Ann Neurol 2017;82:578-591.

Survival and consciousness recovery are better in the minimally conscious state than in the vegetative state.

BACKGROUND: The prognosis value of early clinical diagnosis of consciousness impairment is documented by an extremely limited number of studies, whereas it may convey important information to guide medical decisions. OBJECTIVE: We aimed at determining if patients diagnosed at an early stage (<90 days after brain injury) as being in the minimally conscious state (MCS) have a better prognosis than patients in the vegetative state/Unresponsive Wakefulness syndrome (VS/UWS), independent of care limitations or withdrawal decisions. METHODS: Patients hospitalized in ICUs of the Pitié-Salpêtrière Hospital (Paris, France) from November 2008 to January 2011 were included and evaluated behaviourally with standardized assessment and with the Coma Recovery Scale-Revised as being either in the VS/UWS or in the MCS. They were then prospectively followed until 1July 2011 to evaluate their outcome with the GOSE. We compared survival function and outcomes of these two groups. RESULTS: Both survival function and outcomes, including consciousness recovery, were significantly better in the MCS group. This difference of outcome still holds when considering only patients still alive at the end of the study. CONCLUSIONS: Early accurate clinical diagnosis of VS/UWS or MCS conveys a strong prognostic value of survival and of consciousness recovery.

Autoreject: Automated artifact rejection for MEG and EEG data.

We present an automated algorithm for unified rejection and repair of bad trials in magnetoencephalography (MEG) and electroencephalography (EEG) signals. Our method capitalizes on cross-validation in conjunction with a robust evaluation metric to estimate the optimal peak-to-peak threshold - a quantity commonly used for identifying bad trials in M/EEG. This approach is then extended to a more sophisticated algorithm which estimates this threshold for each sensor yielding trial-wise bad sensors. Depending on the number of bad sensors, the trial is then repaired by interpolation or by excluding it from subsequent analysis. All steps of the algorithm are fully automated thus lending itself to the name Autoreject. In order to assess the practical significance of the algorithm, we conducted extensive validation and comparisons with state-of-the-art methods on four public datasets containing MEG and EEG recordings from more than 200 subjects. The comparisons include purely qualitative efforts as well as quantitatively benchmarking against human supervised and semi-automated preprocessing pipelines. The algorithm allowed us to automate the preprocessing of MEG data from the Human Connectome Project (HCP) going up to the computation of the evoked responses. The automated nature of our method minimizes the burden of human inspection, hence supporting scalability and reliability demanded by data analysis in modern neuroscience.

Assessing and tuning brain decoders: Cross-validation, caveats, and guidelines.

Decoding, i.e. prediction from brain images or signals, calls for empirical evaluation of its predictive power. Such evaluation is achieved via cross-validation, a method also used to tune decoders' hyper-parameters. This paper is a review on cross-validation procedures for decoding in neuroimaging. It includes a didactic overview of the relevant theoretical considerations. Practical aspects are highlighted with an extensive empirical study of the common decoders in within- and across-subject predictions, on multiple datasets -anatomical and functional MRI and MEG- and simulations. Theory and experiments outline that the popular "leave-one-out" strategy leads to unstable and biased estimates, and a repeated random splits method should be preferred. Experiments outline the large error bars of cross-validation in neuroimaging settings: typical confidence intervals of 10%. Nested cross-validation can tune decoders' parameters while avoiding circularity bias. However we find that it can be favorable to use sane defaults, in particular for non-sparse decoders.

Automated model selection in covariance estimation and spatial whitening of MEG and EEG signals.

Magnetoencephalography and electroencephalography (M/EEG) measure non-invasively the weak electromagnetic fields induced by post-synaptic neural currents. The estimation of the spatial covariance of the signals recorded on M/EEG sensors is a building block of modern data analysis pipelines. Such covariance estimates are used in brain-computer interfaces (BCI) systems, in nearly all source localization methods for spatial whitening as well as for data covariance estimation in beamformers. The rationale for such models is that the signals can be modeled by a zero mean Gaussian distribution. While maximizing the Gaussian likelihood seems natural, it leads to a covariance estimate known as empirical covariance (EC). It turns out that the EC is a poor estimate of the true covariance when the number of samples is small. To address this issue the estimation needs to be regularized. The most common approach downweights off-diagonal coefficients, while more advanced regularization methods are based on shrinkage techniques or generative models with low rank assumptions: probabilistic PCA (PPCA) and factor analysis (FA). Using cross-validation all of these models can be tuned and compared based on Gaussian likelihood computed on unseen data. We investigated these models on simulations, one electroencephalography (EEG) dataset as well as magnetoencephalography (MEG) datasets from the most common MEG systems. First, our results demonstrate that different models can be the best, depending on the number of samples, heterogeneity of sensor types and noise properties. Second, we show that the models tuned by cross-validation are superior to models with hand-selected regularization. Hence, we propose an automated solution to the often overlooked problem of covariance estimation of M/EEG signals. The relevance of the procedure is demonstrated here for spatial whitening and source localization of MEG signals.

MNE software for processing MEG and EEG data.

Magnetoencephalography and electroencephalography (M/EEG) measure the weak electromagnetic signals originating from neural currents in the brain. Using these signals to characterize and locate brain activity is a challenging task, as evidenced by several decades of methodological contributions. MNE, whose name stems from its capability to compute cortically-constrained minimum-norm current estimates from M/EEG data, is a software package that provides comprehensive analysis tools and workflows including preprocessing, source estimation, time-frequency analysis, statistical analysis, and several methods to estimate functional connectivity between distributed brain regions. The present paper gives detailed information about the MNE package and describes typical use cases while also warning about potential caveats in analysis. The MNE package is a collaborative effort of multiple institutes striving to implement and share best methods and to facilitate distribution of analysis pipelines to advance reproducibility of research. Full documentation is available at http://martinos.org/mne.

Children's norm enforcement in their interactions with peers.

This study investigates how children negotiate social norms with peers. In Study 1, 48 pairs of 3- and 5-year-olds (N = 96) and in Study 2, 48 pairs of 5- and 7-year-olds (N = 96) were presented with sorting tasks with conflicting instructions (one child by color, the other by shape) or identical instructions. Three-year-olds differed from older children: They were less selective for the contexts in which they enforced norms, and they (as well as the older children to a lesser extent) used grammatical constructions objectifying the norms ("It works like this" rather than "You must do it like this"). These results suggested that children's understanding of social norms becomes more flexible during the preschool years.

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.

A systematic review of multimodal brain age studies: Uncovering a divergence between model accuracy and utility.

Brain aging is a complex, multifaceted process that can be challenging to model in ways that are accurate and clinically useful. One of the most common approaches has been to apply machine learning to neuroimaging data with the goal of predicting age in a data-driven manner. Building on initial brain age studies that were derived solely from T1-weighted scans (i.e., unimodal), recent studies have incorporated features across multiple imaging modalities (i.e., "multimodal"). In this systematic review, we show that unimodal and multimodal models have distinct advantages. Multimodal models are the most accurate and sensitive to differences in chronic brain disorders. In contrast, unimodal models from functional magnetic resonance imaging were most sensitive to differences across a broad array of phenotypes. Altogether, multimodal imaging has provided us valuable insight for improving the accuracy of brain age models, but there is still much untapped potential with regard to achieving widespread clinical utility.

Event-related modulation of alpha rhythm explains the auditory P300-evoked response in EEG.

Evoked responses and oscillations represent two major electrophysiological phenomena in the human brain yet the link between them remains rather obscure. Here we show how most frequently studied EEG signals: the P300-evoked response and alpha oscillations (8-12 Hz) can be linked with the baseline-shift mechanism. This mechanism states that oscillations generate evoked responses if oscillations have a non-zero mean and their amplitude is modulated by the stimulus. Therefore, the following predictions should hold: (1) the temporal evolution of P300 and alpha amplitude is similar, (2) spatial localisations of the P300 and alpha amplitude modulation overlap, (3) oscillations are non-zero mean, (4) P300 and alpha amplitude correlate with cognitive scores in a similar fashion. To validate these predictions, we analysed the data set of elderly participants (N=2230, 60-82 years old), using (a) resting-state EEG recordings to quantify the mean of oscillations, (b) the event-related data, to extract parameters of P300 and alpha rhythm amplitude envelope. We showed that P300 is indeed linked to alpha rhythm, according to all four predictions. Our results provide an unifying view on the interdependency of evoked responses and neuronal oscillations and suggest that P300, at least partly, is generated by the modulation of alpha oscillations.

Cortico-Cerebellar neurodynamics during social interaction in Autism Spectrum Disorders.

BACKGROUND: Exploring neural network dynamics during social interaction could help to identify biomarkers of Autism Spectrum Disorders (ASD). A cerebellar involvement in autism has long been suspected and recent methodological advances now enable studying cerebellar functioning in a naturalistic setting. Here, we investigated the electrophysiological activity of the cerebro-cerebellar network during real-time social interaction in ASD. We focused our analysis on theta oscillations (3-8 Hz), which have been associated with large-scale coordination of distant brain areas and might contribute to interoception, motor control, and social event anticipation, all skills known to be altered in ASD. METHODS: We combined the Human Dynamic Clamp, a paradigm for studying realistic social interactions using a virtual avatar, with high-density electroencephalography (HD-EEG). Using source reconstruction, we investigated power in the cortex and the cerebellum, along with coherence between the cerebellum and three cerebral-cortical areas, and compared our findings in a sample of participants with ASD (n = 107) and with typical development (TD) (n = 33). We developed an open-source pipeline to analyse neural dynamics at the source level from HD-EEG data. RESULTS: Individuals with ASD showed a significant increase in theta band power over the cerebellum and the frontal and temporal cortices during social interaction compared to resting state, along with significant coherence increases between the cerebellum and the sensorimotor, frontal and parietal cortices. However, a phase-based connectivity measure did not support a strict activity increase in the cortico-cerebellar functional network. We did not find any significant differences between the ASD and the TD group. CONCLUSIONS: This exploratory study uncovered increases in the theta band activity of participants with ASD during social interaction, pointing at the presence of neural interactions between the cerebellum and cerebral networks associated with social cognition. It also emphasizes the need for complementary functional connectivity measures to capture network-level alterations. Future work will focus on optimizing artifact correction to include more participants with TD and increase the statistical power of group-level contrasts.

Predicting future cognitive decline from non-brain and multimodal brain imaging data in healthy and pathological aging.

Previous literature has focused on predicting a diagnostic label from structural brain imaging. Since subtle changes in the brain precede a cognitive decline in healthy and pathological aging, our study predicts future decline as a continuous trajectory instead. Here, we tested whether baseline multimodal neuroimaging data improve the prediction of future cognitive decline in healthy and pathological aging. Nonbrain data (demographics, clinical, and neuropsychological scores), structural MRI, and functional connectivity data from OASIS-3 (N = 662; age = 46-96 years) were entered into cross-validated multitarget random forest models to predict future cognitive decline (measured by CDR and MMSE), on average 5.8 years into the future. The analysis was preregistered, and all analysis code is publicly available. Combining non-brain with structural data improved the continuous prediction of future cognitive decline (best test-set performance: R2 = 0.42). Cognitive performance, daily functioning, and subcortical volume drove the performance of our model. Including functional connectivity did not improve predictive accuracy. In the future, the prognosis of age-related cognitive decline may enable earlier and more effective individualized cognitive, pharmacological, and behavioral interventions.