A generative model of electrophysiological brain responses to stimulation.

Each brain response to a stimulus is, to a large extent, unique. However this variability, our perceptual experience feels stable. Standard decoding models, which utilise information across several areas to tap into stimuli representation and processing, are fundamentally based on averages. Therefore, they can focus precisely on the features that are most stable across stimulus presentations. But which are these features exactly is difficult to address in the absence of a generative model of the signal. Here, I introduce genephys, a generative model of brain responses to stimulation publicly available as a Python package that, when confronted with a decoding algorithm, can reproduce the structured patterns of decoding accuracy that we observe in real data. Using this approach, I characterise how these patterns may be brought about by the different aspects of the signal, which in turn may translate into distinct putative neural mechanisms. In particular, the model shows that the features in the data that support successful decoding-and, therefore, likely reflect stable mechanisms of stimulus representation-have an oscillatory component that spans multiple channels, frequencies, and latencies of response; and an additive, slower response with a specific (cross-frequency) relation to the phase of the oscillatory component. At the individual trial level, still, responses are found to be highly variable, which can be due to various factors including phase noise and probabilistic activations.

osl-dynamics, a toolbox for modeling fast dynamic brain activity.

Neural activity contains rich spatiotemporal structure that corresponds to cognition. This includes oscillatory bursting and dynamic activity that span across networks of brain regions, all of which can occur on timescales of tens of milliseconds. While these processes can be accessed through brain recordings and imaging, modeling them presents methodological challenges due to their fast and transient nature. Furthermore, the exact timing and duration of interesting cognitive events are often a priori unknown. Here, we present the OHBA Software Library Dynamics Toolbox (osl-dynamics), a Python-based package that can identify and describe recurrent dynamics in functional neuroimaging data on timescales as fast as tens of milliseconds. At its core are machine learning generative models that are able to adapt to the data and learn the timing, as well as the spatial and spectral characteristics, of brain activity with few assumptions. osl-dynamics incorporates state-of-the-art approaches that can be, and have been, used to elucidate brain dynamics in a wide range of data types, including magneto/electroencephalography, functional magnetic resonance imaging, invasive local field potential recordings, and electrocorticography. It also provides novel summary measures of brain dynamics that can be used to inform our understanding of cognition, behavior, and disease. We hope osl-dynamics will further our understanding of brain function, through its ability to enhance the modeling of fast dynamic processes.

Toward stability of dynamic FC estimates in neuroimaging and electrophysiology: Solutions and limits.

Time-varying functional connectivity (FC) methods are used to map the spatiotemporal organization of brain activity. However, their estimation can be unstable, in the sense that different runs of the inference may yield different solutions. But to draw meaningful relations to behavior, estimates must be robust and reproducible. Here, we propose two solutions using the hidden Markov model (HMM) as a descriptive model of time-varying FC. The first, best ranked HMM, involves running the inference multiple times and selecting the best model based on a quantitative measure combining fitness and model complexity. The second, hierarchical-clustered HMM, generates stable cluster state time series by applying hierarchical clustering to the state time series obtained from multiple runs. Experimental results on fMRI and magnetoencephalography data demonstrate that these approaches substantially improve the stability of time-varying FC estimations. Overall, hierarchical-clustered HMM is preferred when the inference variability is high, while the best ranked HMM performs better otherwise.

A data-driven network decomposition of the temporal, spatial, and spectral dynamics underpinning visual-verbal working memory processes.

The brain dynamics underlying working memory (WM) unroll via transient frequency-specific large-scale brain networks. This multidimensionality (time, space, and frequency) challenges traditional analyses. Through an unsupervised technique, the time delay embedded-hidden Markov model (TDE-HMM), we pursue a functional network analysis of magnetoencephalographic data from 38 healthy subjects acquired during an n-back task. Here we show that this model inferred task-specific networks with unique temporal (activation), spectral (phase-coupling connections), and spatial (power spectral density distribution) profiles. A theta frontoparietal network exerts attentional control and encodes the stimulus, an alpha temporo-occipital network rehearses the verbal information, and a broad-band frontoparietal network with a P300-like temporal profile leads the retrieval process and motor response. Therefore, this work provides a unified and integrated description of the multidimensional working memory dynamics that can be interpreted within the neuropsychological multi-component model of WM, improving the overall neurophysiological and neuropsychological comprehension of WM functioning.

Dissecting unsupervised learning through hidden Markov modeling in electrophysiological data.

Unsupervised, data-driven methods are commonly used in neuroscience to automatically decompose data into interpretable patterns. These patterns differ from one another depending on the assumptions of the models. How these assumptions affect specific data decompositions in practice, however, is often unclear, which hinders model applicability and interpretability. For instance, the hidden Markov model (HMM) automatically detects characteristic, recurring activity patterns (so-called states) from time series data. States are defined by a certain probability distribution, whose state-specific parameters are estimated from the data. But what specific features, from all of those that the data contain, do the states capture? That depends on the choice of probability distribution and on other model hyperparameters. Using both synthetic and real data, we aim to better characterize the behavior of two HMM types that can be applied to electrophysiological data. Specifically, we study which differences in data features (such as frequency, amplitude, or signal-to-noise ratio) are more salient to the models and therefore more likely to drive the state decomposition. Overall, we aim at providing guidance for the appropriate use of this type of analysis on one- or two-channel neural electrophysiological data and an informed interpretation of its results given the characteristics of the data and the purpose of the analysis.NEW & NOTEWORTHY Compared with classical supervised methods, unsupervised methods of analysis have the advantage to be freer of subjective biases. However, it is not always clear what aspects of the data these methods are most sensitive to, which complicates interpretation. Focusing on the hidden Markov model, commonly used to describe electrophysiological data, we explore in detail the nature of its estimates through simulations and real data examples, providing important insights about what to expect from these models.

WITHDRAWN: Dynamic functional connectivity: Why the controversy?

This article has been withdrawn at the request of the author(s) and/or editor. The Publisher apologizes for any inconvenience this may cause. The full Elsevier Policy on Article Withdrawal can be found at https://www.elsevier.com/about/policies/article-withdrawal.

The relationship between frequency content and representational dynamics in the decoding of neurophysiological data.

Decoding of high temporal resolution, stimulus-evoked neurophysiological data is increasingly used to test theories about how the brain processes information. However, a fundamental relationship between the frequency spectra of the neural signal and the subsequent decoding accuracy timecourse is not widely recognised. We show that, in commonly used instantaneous signal decoding paradigms, each sinusoidal component of the evoked response is translated to double its original frequency in the subsequent decoding accuracy timecourses. We therefore recommend, where researchers use instantaneous signal decoding paradigms, that more aggressive low pass filtering is applied with a cut-off at one quarter of the sampling rate, to eliminate representational alias artefacts. However, this does not negate the accompanying interpretational challenges. We show that these can be resolved by decoding paradigms that utilise both a signal's instantaneous magnitude and its local gradient information as features for decoding. On a publicly available MEG dataset, this results in decoding accuracy metrics that are higher, more stable over time, and free of the technical and interpretational challenges previously characterised. We anticipate that a broader awareness of these fundamental relationships will enable stronger interpretations of decoding results by linking them more clearly to the underlying signal characteristics that drive them.

Effective psychological therapy for PTSD changes the dynamics of specific large-scale brain networks.

In posttraumatic stress disorder (PTSD), re-experiencing of the trauma is a hallmark symptom proposed to emerge from a de-contextualised trauma memory. Cognitive therapy for PTSD (CT-PTSD) addresses this de-contextualisation through different strategies. At the brain level, recent research suggests that the dynamics of specific large-scale brain networks play an essential role in both the healthy response to a threatening situation and the development of PTSD. However, very little is known about how these dynamics are altered in the disorder and rebalanced after treatment and successful recovery. Using a data-driven approach and fMRI, we detected recurring large-scale brain functional states with high temporal precision in a population of healthy trauma-exposed and PTSD participants before and after successful CT-PTSD. We estimated the total amount of time that each participant spent on each of the states while being exposed to trauma-related and neutral pictures. We found that PTSD participants spent less time on two default mode subnetworks involved in different forms of self-referential processing in contrast to PTSD participants after CT-PTSD (mtDMN+ and dmDMN+ ) and healthy trauma-exposed controls (only mtDMN+ ). Furthermore, re-experiencing severity was related to decreased time spent on the default mode subnetwork involved in contextualised retrieval of autobiographical memories, and increased time spent on the salience and visual networks. Overall, our results support the hypothesis that PTSD involves an imbalance in the dynamics of specific large-scale brain network states involved in self-referential processes and threat detection, and suggest that successful CT-PTSD might rebalance this dynamic aspect of brain function.

Spatiotemporally resolved multivariate pattern analysis for M/EEG.

An emerging goal in neuroscience is tracking what information is represented in brain activity over time as a participant completes some task. While electroencephalography (EEG) and magnetoencephalography (MEG) offer millisecond temporal resolution of how activity patterns emerge and evolve, standard decoding methods present significant barriers to interpretability as they obscure the underlying spatial and temporal activity patterns. We instead propose the use of a generative encoding model framework that simultaneously infers the multivariate spatial patterns of activity and the variable timing at which these patterns emerge on individual trials. An encoding model inversion maps from these parameters to the equivalent decoding model, allowing predictions to be made about unseen test data in the same way as in standard decoding methodology. These SpatioTemporally Resolved MVPA (STRM) models can be flexibly applied to a wide variety of experimental paradigms, including classification and regression tasks. We show that these models provide insightful maps of the activity driving predictive accuracy metrics; demonstrate behaviourally meaningful variation in the timing of pattern emergence on individual trials; and achieve predictive accuracies that are either equivalent or surpass those achieved by more widely used methods. This provides a new avenue for investigating the brain's representational dynamics and could ultimately support more flexible experimental designs in the future.

Multi-dynamic modelling reveals strongly time-varying resting fMRI correlations.

The activity of functional brain networks is responsible for the emergence of time-varying cognition and behaviour. Accordingly, time-varying correlations (Functional Connectivity) in resting fMRI have been shown to be predictive of behavioural traits, and psychiatric and neurological conditions. Typically, methods that measure time varying Functional Connectivity (FC), such as sliding windows approaches, do not separately model when changes occur in the mean activity levels from when changes occur in the FC, therefore conflating these two distinct types of modulation. We show that this can bias the estimation of time-varying FC to appear more stable over time than it actually is. Here, we propose an alternative approach that models changes in the mean brain activity and in the FC as being able to occur at different times to each other. We refer to this method as the Multi-dynamic Adversarial Generator Encoder (MAGE) model, which includes a model of the network dynamics that captures long-range time dependencies, and is estimated on fMRI data using principles of Generative Adversarial Networks. We evaluated the approach across several simulation studies and resting fMRI data from the Human Connectome Project (1003 subjects), as well as from UK Biobank (13301 subjects). Importantly, we find that separating fluctuations in the mean activity levels from those in the FC reveals much stronger changes in FC over time, and is a better predictor of individual behavioural variability.

Balance between competing spectral states in subthalamic nucleus is linked to motor impairment in Parkinson's disease.

Exaggerated local field potential bursts of activity at frequencies in the low beta band are a well-established phenomenon in the subthalamic nucleus of patients with Parkinson's disease. However, such activity is only moderately correlated with motor impairment. Here we test the hypothesis that beta bursts are just one of several dynamic states in the subthalamic nucleus local field potential in Parkinson's disease, and that together these different states predict motor impairment with high fidelity. Local field potentials were recorded in 32 patients (64 hemispheres) undergoing deep brain stimulation surgery targeting the subthalamic nucleus. Recordings were performed following overnight withdrawal of anti-parkinsonian medication, and after administration of levodopa. Local field potentials were analysed using hidden Markov modelling to identify transient spectral states with frequencies under 40 Hz. Findings in the low beta frequency band were similar to those previously reported; levodopa reduced occurrence rate and duration of low beta states, and the greater the reductions, the greater the improvement in motor impairment. However, additional local field potential states were distinguished in the theta, alpha and high beta bands, and these behaved in an opposite manner. They were increased in occurrence rate and duration by levodopa, and the greater the increases, the greater the improvement in motor impairment. In addition, levodopa favoured the transition of low beta states to other spectral states. When all local field potential states and corresponding features were considered in a multivariate model it was possible to predict 50% of the variance in patients' hemibody impairment OFF medication, and in the change in hemibody impairment following levodopa. This only improved slightly if signal amplitude or gamma band features were also included in the multivariate model. In addition, it compares with a prediction of only 16% of the variance when using beta bursts alone. We conclude that multiple spectral states in the subthalamic nucleus local field potential have a bearing on motor impairment, and that levodopa-induced shifts in the balance between these states can predict clinical change with high fidelity. This is important in suggesting that some states might be upregulated to improve parkinsonism and in suggesting how local field potential feedback can be made more informative in closed-loop deep brain stimulation systems.

Frequency modulation of entorhinal cortex neuronal activity drives distinct frequency-dependent states of brain-wide dynamics.

Human neuroimaging studies have shown that, during cognitive processing, the brain undergoes dynamic transitions between multiple, frequency-tuned states of activity. Although different states may emerge from distinct sources of neural activity, it remains unclear whether single-area neuronal spiking can also drive multiple dynamic states. In mice, we ask whether frequency modulation of the entorhinal cortex activity causes dynamic states to emerge and whether these states respond to distinct stimulation frequencies. Using hidden Markov modeling, we perform unsupervised detection of transient states in mouse brain-wide fMRI fluctuations induced via optogenetic frequency modulation of excitatory neurons. We unveil the existence of multiple, frequency-dependent dynamic states, invisible through standard static fMRI analyses. These states are linked to different anatomical circuits and disrupted in a frequency-dependent fashion in a transgenic model of cognitive disease directly related to entorhinal cortex dysfunction. These findings provide cross-scale insight into basic neuronal mechanisms that may underpin flexibility in brain-wide dynamics.

Breakdown of random matrix universality in Markov models.

Biological systems need to react to stimuli over a broad spectrum of timescales. If and how this ability can emerge without external fine-tuning is a puzzle. We consider here this problem in discrete Markovian systems, where we can leverage results from random matrix theory. Indeed, generic large transition matrices are governed by universal results, which predict the absence of long timescales unless fine-tuned. We consider an ensemble of transition matrices and motivate a temperature-like variable that controls the dynamic range of matrix elements, which we show plays a crucial role in the applicability of the large matrix limit: as the dynamic range increases, a phase transition occurs whereby the random matrix theory result is avoided, and long relaxation times ensue, in the entire "ordered" phase. We furthermore show that this phase transition is accompanied by a drop in the entropy rate and a peak in complexity, as measured by predictive information [Bialek, Nemenman, and Tishby Neural Comput. 13, 2409 (2001)NEUCEB0899-766710.1162/089976601753195969]. Extending the Markov model to a Hidden Markov model (HMM), we show that observable sequences inherit properties of the hidden sequences, allowing HMMs to be understood in terms of more accessible Markov models. We then apply our findings to fMRI data from 820 human subjects scanned at wakeful rest. We show that the data can be quantitatively understood in terms of the random model, and that brain activity lies close to the phase transition when engaged in unconstrained, task-free cognition-supporting the brain criticality hypothesis in this context.

Dissociable Components of Information Encoding in Human Perception.

Brain decoding can predict visual perception from non-invasive electrophysiological data by combining information across multiple channels. However, decoding methods typically conflate the composite and distributed neural processes underlying perception that are together present in the signal, making it unclear what specific aspects of the neural computations involved in perception are reflected in this type of macroscale data. Using MEG data recorded while participants viewed a large number of naturalistic images, we analytically decomposed the brain signal into its oscillatory and non-oscillatory components, and used this decomposition to show that there are at least three dissociable stimulus-specific aspects to the brain data: a slow, non-oscillatory component, reflecting the temporally stable aspect of the stimulus representation; a global phase shift of the oscillation, reflecting the overall speed of processing of specific stimuli; and differential patterns of phase across channels, likely reflecting stimulus-specific computations. Further, we show that common cognitive interpretations of decoding analysis, in particular about how representations generalize across time, can benefit from acknowledging the multicomponent nature of the signal in the study of perception.

Differential dopaminergic modulation of spontaneous cortico-subthalamic activity in Parkinson's disease.

Pathological oscillations including elevated beta activity in the subthalamic nucleus (STN) and between STN and cortical areas are a hallmark of neural activity in Parkinson's disease (PD). Oscillations also play an important role in normal physiological processes and serve distinct functional roles at different points in time. We characterised the effect of dopaminergic medication on oscillatory whole-brain networks in PD in a time-resolved manner by employing a hidden Markov model on combined STN local field potentials and magnetoencephalography (MEG) recordings from 17 PD patients. Dopaminergic medication led to coherence within the medial and orbitofrontal cortex in the delta/theta frequency range. This is in line with known side effects of dopamine treatment such as deteriorated executive functions in PD. In addition, dopamine caused the beta band activity to switch from an STN-mediated motor network to a frontoparietal-mediated one. In contrast, dopamine did not modify local STN-STN coherence in PD. STN-STN synchrony emerged both on and off medication. By providing electrophysiological evidence for the differential effects of dopaminergic medication on the discovered networks, our findings open further avenues for electrical and pharmacological interventions in PD.

A new model for simultaneous dimensionality reduction and time-varying functional connectivity estimation.

An important question in neuroscience is whether or not we can interpret spontaneous variations in the pattern of correlation between brain areas, which we refer to as functional connectivity or FC, as an index of dynamic neuronal communication in fMRI. That is, can we measure time-varying FC reliably? And, if so, can FC reflect information transfer between brain regions at relatively fast-time scales? Answering these questions in practice requires dealing with the statistical challenge of having high-dimensional data and a comparatively lower number of time points or volumes. A common strategy is to use PCA to reduce the dimensionality of the data, and then apply some model, such as the hidden Markov model (HMM) or a mixture model of Gaussian distributions, to find a set of distinct FC patterns or states. The distinct spatial properties of these FC states together with the time-resolved switching between them offer a flexible description of time-varying FC. In this work, I show that in this context PCA can suffer from systematic biases and loss of sensitivity for the purposes of finding time-varying FC. To get around these issues, I propose a novel variety of the HMM, named HMM-PCA, where the states are themselves PCA decompositions. Since PCA is based on the data covariance, the state-specific PCA decompositions reflect distinct patterns of FC. I show, theoretically and empirically, that fusing dimensionality reduction and time-varying FC estimation in one single step can avoid these problems and outperform alternative approaches, facilitating the quantification of transient communication in the brain.

Increased brain atrophy and lesion load is associated with stronger lower alpha MEG power in multiple sclerosis patients.

In multiple sclerosis, the interplay of neurodegeneration, demyelination and inflammation leads to changes in neurophysiological functioning. This study aims to characterize the relation between reduced brain volumes and spectral power in multiple sclerosis patients and matched healthy subjects. During resting-state eyes closed, we collected magnetoencephalographic data in 67 multiple sclerosis patients and 47 healthy subjects, matched for age and gender. Additionally, we quantified different brain volumes through magnetic resonance imaging (MRI). First, a principal component analysis of MRI-derived brain volumes demonstrates that atrophy can be largely described by two components: one overall degenerative component that correlates strongly with different cognitive tests, and one component that mainly captures degeneration of the cortical grey matter that strongly correlates with age. A multimodal correlation analysis indicates that increased brain atrophy and lesion load is accompanied by increased spectral power in the lower alpha (8-10 Hz) in the temporoparietal junction (TPJ). Increased lower alpha power in the TPJ was further associated with worse results on verbal and spatial working memory tests, whereas an increased lower/upper alpha power ratio was associated with slower information processing speed. In conclusion, multiple sclerosis patients with increased brain atrophy, lesion and thalamic volumes demonstrated increased lower alpha power in the TPJ and reduced cognitive abilities.

Revisiting the global workspace orchestrating the hierarchical organization of the human brain.

A central challenge in neuroscience is how the brain organizes the information necessary to orchestrate behaviour. Arguably, this whole-brain orchestration is carried out by a core subset of integrative brain regions, a 'global workspace', but its constitutive regions remain unclear. We quantified the global workspace as the common regions across seven tasks as well as rest, in a common 'functional rich club'. To identify this functional rich club, we determined the information flow between brain regions by means of a normalized directed transfer entropy framework applied to multimodal neuroimaging data from 1,003 healthy participants and validated in participants with retest data. This revealed a set of regions orchestrating information from perceptual, long-term memory, evaluative and attentional systems. We confirmed the causal significance and robustness of our results by systematically lesioning a generative whole-brain model. Overall, this framework describes a complex choreography of the functional hierarchical organization of the human brain.

Replay bursts in humans coincide with activation of the default mode and parietal alpha networks.

Our brains at rest spontaneously replay recently acquired information, but how this process is orchestrated to avoid interference with ongoing cognition is an open question. Here we investigated whether replay coincided with spontaneous patterns of whole-brain activity. We found, in two separate datasets, that replay sequences were packaged into transient bursts occurring selectively during activation of the default mode network (DMN) and parietal alpha networks. These networks are believed to support inwardly oriented attention and inhibit bottom-up sensory processing and were characterized by widespread synchronized oscillations coupled to increases in high frequency power, mechanisms thought to coordinate information flow between disparate cortical areas. Our data reveal a tight correspondence between two widely studied phenomena in neural physiology and suggest that the DMN may coordinate replay bursts in a manner that minimizes interference with ongoing cognition.

Learning patterns of the ageing brain in MRI using deep convolutional networks.

Both normal ageing and neurodegenerative diseases cause morphological changes to the brain. Age-related brain changes are subtle, nonlinear, and spatially and temporally heterogenous, both within a subject and across a population. Machine learning models are particularly suited to capture these patterns and can produce a model that is sensitive to changes of interest, despite the large variety in healthy brain appearance. In this paper, the power of convolutional neural networks (CNNs) and the rich UK Biobank dataset, the largest database currently available, are harnessed to address the problem of predicting brain age. We developed a 3D CNN architecture to predict chronological age, using a training dataset of 12,802 T1-weighted MRI images and a further 6,885 images for testing. The proposed method shows competitive performance on age prediction, but, most importantly, the CNN prediction errors ΔBrainAge=AgePredicted-AgeTrue correlated significantly with many clinical measurements from the UK Biobank in the female and male groups. In addition, having used images from only one imaging modality in this experiment, we examined the relationship between ΔBrainAge and the image-derived phenotypes (IDPs) from all other imaging modalities in the UK Biobank, showing correlations consistent with known patterns of ageing. Furthermore, we show that the use of nonlinearly registered images to train CNNs can lead to the network being driven by artefacts of the registration process and missing subtle indicators of ageing, limiting the clinical relevance. Due to the longitudinal aspect of the UK Biobank study, in the future it will be possible to explore whether the ΔBrainAge from models such as this network were predictive of any health outcomes.