Brain Topography

In this study, we subject the conventional physiological interpretation of independent component analysis (ICA) applied to EEG, the small-patch model, to systematic falsification, and propose an alternative large-patch model. The small-patch model assumes that ICs correspond to localized cortical patches with < 1 cm{superscript 2}. However, this assumption has remained unvalidated. The small-patch model predicts that approximately 70% of sources are localized within sulci up to 15 mm deep, with rapidly changing dipole orientations across the cortex. In contrast, the large-patch model (>6-10 cm{superscript 2}) predicts relatively stable radial orientations accompanied by physiologically implausible source depths due to depth bias. First, we conducted a stimulation study using a forward-inverse modeling framework with a four-layer head conductor model. We confirmed that depth bias emerges when a single equivalent dipole is fitted to a potential field generated by a broad array of parallel dipoles. This observation led to the key hypothesis that the presence of depth bias in empirical data would favor the large-patch model. Second, we analyzed resting-state EEG from two European open datasets comprising 820 recordings (62-64 channels), yielding dipole depth and orientation distributions for nearly 15,000 qualified brain ICs. Results showed that more than 80% of ICs were localized at physiologically implausible depths (19-26 mm), favoring the large-patch model. A novel dipole-orientation analysis revealed broad, low-spatial-frequency structure in dipole orientations, further supporting the large-patch model. We conclude that the revised physiological interpretation of ICA aligns with electrophysiological literature and computational insights into EEG-specific spatial scaling laws. Significance statementIndependent component analysis (ICA) has been proposed as a promising tool for computational neuroscience using human scalp EEG. One of the original proponents introduced a physiological model suggesting that anatomically accurate neural sources could be directly recovered by applying ICA to EEG data. However, we found that this model assumes EEG generation within cortical patches smaller than 1 cm{superscript 2}, which has remained unvalidated for over a decade and requires revision. Using both simulation and empirical EEG datasets, we demonstrated that our alternative model, involving larger cortical patches (>6-10 cm{superscript 2}), better fits the electrophysiological generative model of scalp EEG signals. We conclude that our large-patch model provides an updated, more physiologically plausible interpretation of ICA results.

Both episodic memory and word retrieval have been linked to power decreases in the alpha and beta oscillatory bands, but these patterns have rarely been related to each other, partly due to a lack of methodological approaches available. In this explorative study, we investigate the similarities and dissimilarities in the oscillatory fingerprints of the retrieval of words and episodes by directly comparing the activity patterns across time, frequency, and space. We acquired electroencephalography (EEG) data of participants performing a language and an episodic memory task based on the same stimulus material. With a newly developed approach, we directly compared the source-reconstructed oscillatory activity using mutual information and a feature-impact analysis. While left temporal and frontal regions showed dissimilarities between the tasks, right-hemispheric parietal regions exhibited similarities. We speculate that this could indicate a homologous function of these regions, potentially sharing less-specific representations between the tasks. We further uncovered a dissociation of the alpha and beta bands regarding the similarity across tasks. While the beta band was dissimilar between word and episodic memory retrieval, the alpha band seemed to contribute to the similarity we observed in right parietal regions. Whether this points to a task-unspecific function of the alpha band or a functional role in the retrieval process of the presumed representations, remains to be determined. In summary, we present an approach to study similarity across tasks using the temporal, spectral, and spatial dimensions of EEG data, and present results of exploring the shared oscillatory fingerprints between episodic memory and word retrieval.

Interest in broadband aperiodic brain activity (1/f phenomenon) has increased exponentially over recent years, partly fueled by the development of tools to parameterize it (i.e., estimate its offset/intercept and exponent/slope) using the M/EEG power spectrum. Broadband aperiodic activity needs to be separated from narrowband periodic activity before its parameters are computed. A popular method, the fooof toolbox (Donoghue et al., 2020), is based on the data-driven detection of narrowband-periodic peaks, whose maximum number is set by the user. While increasing analytic flexibility, variability in the number of detected peaks may increase sensitivity to noise and reduce the reliability of aperiodic parameter estimates and the power of analytic pipelines. Here, we present an investigation of the effects of analytic choices (e.g., number of peaks, spectral estimation method) on metrics indicating the adequacy of spectral parametrization. These include the internal consistency (odd-even reliability) of aperiodic estimates, the number of outliers generated, and their ability to detect effects. Across two different data sets (resting state and task-based) we found a decrease in the reliability of intercept and slope estimates as more peaks were allowed to be extracted. To ameliorate this problem, we propose a theory-driven modification of fooof labelled censored regression, whereby a theory-driven range of frequencies expected to contain periodic activity is removed from all spectra, and the remaining power values are regressed on the remaining frequencies to obtain parameter estimates. This method shows more reliable and robust estimates compared to fooof, while avoiding overfitting.

When participants are engaged with auditory narratives, physiological and neural signals exhibit temporal correlations between subjects. The intersubject correlation (ISC) increases when attention is directed to the stories, suggesting that shared neural and bodily dynamics arise from a similar processing of the narratives. Identifying the factors that drive these common responses is clinically relevant for interpreting EEG ISC exhibited in unresponsive patients. In this study, we investigated whether the ISC of the EEG elicited by auditory narratives is driven by low-level acoustic (envelope, spectrogram) and/or higher-level linguistic information (word onset, word surprisal) in two groups of healthy participants during passive, attentive and distracted listening. We use temporal response functions (TRFs) for acoustic, and linguistic features to assess the contribution of each feature to the ISC, measured using correlated component analysis (CorrCA). TRFs derived for acoustic features explained a larger fraction of variance in the EEG than linguistic features and were the main contributors to the ISC. The attention-related increase in ISC was driven by all features. Importantly, word surprisal had an effect on ISC only during active story engagement, with timing and scalp distribution consistent with language processing. Notably, the linear responses captured by TRFs only explained a small amount of the overall ISC, suggesting that ISC is largely driven by nonlinear responses to the narratives. We propose that the combined use of ISC and TRFs has the potential to provide meaningful markers of language processing in patients with disorders of consciousness, and we suggest practical recommendations for their implementation.

An increasing number of studies are currently focusing on personality neuroscience, a term denoting the research aimed at neuroimaging correlates of inter-individual temperament and character variability. Among other methods, a graph theoretical analysis of the functional connectivity in resting-state functional magnetic resonance imaging data was applied in a study by Gao et al. (2013), reporting novel functional connectivity correlates of personality traits. The current paper presents a conceptual replication of the results of this study and discusses the related challenges, including an extension of the original statistical methods in order to illustrate the effect of the multiple comparison problem. Five personality dimensions were obtained using the revised Big Five Personality Inventory, including scores of Extraversion and Neuroticism covered in the original paper. Using a larger sample (84 subjects) with adequate statistical power (ranging from 0.75 to 0.95 across analyses), we failed to replicate any of the nine specific neuroimaging correlates of personality presented by Gao et al. While acknowledging differences in the experimental procedures, we discuss that the lack of replication might be caused by the relatively liberal control of false positives in the original study. Indeed, the original testing scheme leads to an expected count of about 10 false positive observations among all tests; applying this scheme to our data we observed a similar number of positive tests, albeit for different relations. No significant correlations were found in our data when standard family-wise error control was applied. These results illustrate the importance of combining exploration with independent validation, use of large datasets, as well as appropriate control of multiple comparison problem in order to prevent false alarms in research into neural substrates of personality differences. Importantly, our findings do not disprove the existence of a link between personality and the brains intrinsic functional architecture; but rather suggest that such a link might be even more subtle and elusive than previously reported.

Transcranial magnetic stimulation (TMS) combined with electroencephalography (EEG) has strong potential for recording cortical reactivity and connectivity. However, this promise is hampered by TMS-induced EEG artifacts. Here, we examine the origins of these artifacts with phantom TMS-EEG recordings and simulations. We focus on two major types of artifacts: (1) the TMS pulse artifact during each [~]0.2 ms TMS pulse and (2) the decay artifact that may last tens of milliseconds. We examine how these artifacts change as a function of the relative position between TMS coil windings and EEG electrode leads. We also examine the hypothesis that certain EEG lead configurations may reduce or even cancel out these artifacts. In experimental results across 23 different TMS coil / EEG lead configurations, the amplitudes between the TMS pulse artifact and the decay artifact were highly correlated (Spearman {rho} = 0.86, p < 0.001), suggesting that the decay artifact is caused by the TMS pulse artifact. As predicted, in certain EEG lead configurations, both the TMS pulse and decay artifacts were minimized. The simulations confirmed that the TMS pulse artifacts depended on the electromagnetic induction from the TMS coil windings to the EEG leads. These results illuminate the generator mechanisms of--and possible means to reduce--both artifacts.

The human visual system represents stimuli in a rich and detailed manner. Traditional methods of studying visual representations in humans, such as event-related potentials (ERP), revealed numerous distinctions between the brain activity elicited by different categories of stimuli. However, these methods miss the information embedded in the spatial distributions of brain activity, or patterns, and are not always sensitive to study visual representations of different stimuli at the single participant or single trial level. Time-resolved multivariate pattern classification analysis (MVPA), or Decoding, efficiently extracts the visual representations of stimuli from the EEG topography without a-priori assumptions about the location of the effect in time and space at the single participant level. The rich information this method provides has increased its popularity dramatically in recent years. Yet, different participants show variable quality of decoding performance, and it is unclear if the accuracy of decoding is maintained within participants across multiple sessions, tasks, attentional conditions and visual features. In the current study, participants performed three visual tasks, over two sessions (1-7 days apart). We examined the correlation of decoding accuracy: within the cross-validation set, between sessions, between features (color and category) and to different measurements of the ERP signal and behavioral performance. We also examined how models generalized to different tasks and different attention conditions. We found that decoding accuracies varied substantially across participants, and that decoding accuracy was reliable within participant, over sessions, attention condition and task. This suggests the decodability behaves like an individual trait. Moreover, the spatial patterns underlying the decoding (classification weights) generalized across different tasks, attentional conditions and sessions. This suggests minimal representational drift at the resolution allowed by the EEG. We conclude that EEG decoding is a reliable method, and that visual representations are stable.

Functional brain network (FBN) dysconnectivity has been repeatedly reported in bipolar disorder (BD). However, it remains unclear how this dysconnectivity manifests from the perspective of oscillatory FBNs, that is, which network measures and frequency bands most reliably capture this alteration. Moreover, it is unknown whether this dysconnection is predominantly expressed through linear or nonlinear interactions. Here, we investigated properties of oscillatory FBNs in individuals with euthymic BD. Networks were constructed using linear and nonlinear connectivity measures applied to source-localized resting-state electroencephalography (EEG) current density signals. We then quantified whole-FBN and nodal features using conventional and spectral graph theory methods to characterize disorder-related network mechanisms and evaluate their potential as biomarkers. Significant group differences between BD and control groups were observed in the theta and alpha1 bands. Dynamical whole-FBN alterations were detected primarily in linear oscillatory FBNs, with reduced Shannon entropy and energy in the BD group. These effects were replicated using machine learning, achieving 85% classification accuracy with entropy and energy as the most informative features. In contrast, nodal-level differences emerged mainly in nonlinear FBNs, revealing increased centrality in frontal, and decreased centrality across temporal, and limbic regions. These findings emphasize distinct, frequency-specific roles of linear and nonlinear oscillatory FBNs in BD, with global dysconnectivity reflected in linear FBNs and local alterations captured by nonlinear connectivity. Moreover, network measures related to synchronization stability and complexity more effectively capture BD-related dysconnectivity, suggesting that dynamic features of oscillatory FBNs may serve as potential biomarkers for BD.

Electroencephalography and magnetoencephalography (EEG/MEG) provide non-invasive measurements of large-scale neural activity but do not directly reveal the underlying cortical sources, motivating the use of source localisation algorithms. However, objective evaluation of these methods remains challenging due to the absence of an experimentally verifiable ground truth. This study presents a simulation framework for generating biologically plausible ictal dynamics and their corresponding EEG signals to enable systematic benchmarking of source imaging approaches. Cortical seizure initiation and propagation were simulated using network-coupled neural mass (Epileptor) models, and combined with realistic forward models of the human head to produce macroscopic, electrophysiological data with known ground truth under varying conditions. Using this dataset, we evaluated established source localisation methods across idealised and realistic scenarios. Existing approaches achieved reasonable spatial accuracy under high-density, noise-free conditions; however, performance degraded substantially with reduced sensor coverage and added noise. This degradation was driven primarily by failures to recover source polarity, even when spatial localisation remained relatively accurate. These results suggest that current methods may be sufficient for identifying epileptogenic regions or tracking regional recruitment, but highlight polarity reconstruction as a key limitation for studies of seizure dynamics and network organisation. The proposed framework provides a reproducible and biologically grounded testbed for the development and evaluation of electrophysiological source localisation techniques.

Accurate transcranial electrical stimulation (TES), electroconvulsive therapy (ECT), and electroencephalography (EEG) forward modeling requires resolving numerical singularities in the charge density near electrodes and tissue interfaces. We present an adaptive mesh refinement (AMR) strategy for the charge based boundary element method (BEM) accelerated by the fast multiple method (BEM-FMM) including electrode and interface singularities. We derive a new error estimator which considers both local and nonlocal contributions of the single-layer potential operator and construct a refinement criterion based on the difference in charge solution across AMR iterations. We evaluate this approach on a 5-layer sphere model and on multiple subject-specific head models derived from the 7-tissue SimNIBS (headreco) and 40-tissue Sim4Life (head40) segmentations, using both voltage-controlled and current-controlled electrode formulations. Through convergence analysis on the white matter and deep hippocampal targets, we find electric fields with relative residual errors below 1% and 0.1% for SimNIBS and Sim4Life models, respectively. Our results indicate that the residual based AMR applied to BEM-FMM leads to numerically stable TES and EEG forward solutions in realistic head models.

The complex interplay between brain regions that support emotional experience and their link to individual differences is a topic of active research. Additionally, there has been growing interest in using functional gradients to investigate human cortical organization during both rest and film fMRI. Among these, several studies demonstrated improved brain fingerprinting performance, reflecting greater neural identification capability of film fMRI against rest fMRI despite higher subject synchronization during film-watching than in rest. Comparably, in this work we study the relation between individual differences, in particular, state anxiety and openness scores, and brain activity during the processing of various emotional scenes in films, through functional gradients. Next to including subcortical areas, we also propose a new approach of computing functional gradients based on a subset of frames selected using emotional annotation data of films, resulting in emotion-informed functional gradients. Then we evaluate the variance in emotion-informed gradients across subjects and employ these same gradients in the prediction of individual differences. For emotion-informed functional gradients, the highest predictability of state anxiety was found for scenes of negative valence and medium-high arousal, corresponding to the typical location of anxiety within the valence-arousal-power emotional space. Additionally, predictability of state anxiety was negatively correlated to inter-subject variability. In contrast, predictability of openness was found to be highest during scenes with low arousal and positively correlated to inter-subject variability. In essence, our results first show that macroscale brain organization is affected by emotional experience, and that frame selection based on the latter can be useful to remove non-subject-specific variability while extracting subject-specific information related to the emotion experience. It also demonstrates that frame selection increases inter-subject variability allowing the extraction of more subject-specific information. Thus, expanding on the idea of brain fingerprint in film fMRI, we argue that emotional experiences enhance disentanglement of various domain of individual differences. Moreover, depending on the individual difference of interest, fMRI acquired during more or less constrained paradigms would be more suitable to reveal different properties of brain function.

Many functional magnetic resonance imaging (fMRI) studies conclude that two conditions engage "overlapping, yet partly distinct" patterns of activation. Yet, there is currently no commonly accepted method for determining the extent of this overlap. While correlations between activation patterns can serve as a measure of their correspondence, empirical correlations are strongly biased towards zero due to measurement noise, preventing their use in testing hypotheses about the actual degree of pattern correspondence. In this paper, we derive the maximum-likelihood estimate for the correlation of the true (noise-less) activation patterns and examine its behavior in the low signal-to-noise regime that is typical for fMRI studies. We show that although the maximum-likelihood estimate corrects for much of the influence of measurement noise, it is ultimately biased. We examine different ways of drawing inferences about the size of the underlying true correlations. We find that a subject-wise bootstrap on the maximum-likelihood group estimate performs best over the tested conditions. We extend the proposed method to test more general hypotheses about the representational geometry of activation patterns for more conditions, and highlight best practices, as well as common pitfalls and problems, in testing such hypotheses.

Voltage perturbations to a repetitively firing Hodgkin-Huxley (HH) model of neuronal spiking in the bistable regime with coexisting limit cycle and stable steady node can either lead to the spikes phase resetting or collapse to the stable steady state. The latter describes a non-firing hyperpolarized quiescent state of the neuron despite the presence of constant external current. Using asymptotic phase response curve (PRC), the impact of voltage perturbations on a repetitively firing HH model is studied here while it is diffusively coupled to another HH model under identical external stimulation. It is observed that the pre-perturbation state of synchronization and the coupling strength critically determine the PRC response of the perturbed HH dynamics. Higher coupling strengths of perfectly in-phase (anti-phase) synchronized HH models shrink (expand) the combinatorial space of perturbation strengths and the oscillation phases causing collapse to the quiescent state. This indicates reduced (enlarged) basin of attraction, viz. the null space, associated with the steady state in the HH phase space. The findings bear important implications to the spiking dynamics of diverse interneurons, as well as special cases of pyramidal neurons, coupled through electrical synapses via. gap junctions, and suggest the role of gap junction plasticity in tuning vulnerability to quiescent state in the presence of biological noise and spikelets.

Seizure propagation - how epileptogenic brain regions recruit less excitable regions - is poorly understood. Previous studies have used dynamical modeling to study seizure propagation and to create patient-specific whole-brain models of seizure spread. However, these studies focused on seizures of a single dynamotype (onset and offset bifurcation pair). Here, we implement a novel coupling method to investigate seizure propagation in a diverse array of dynamotypes. We utilize the Multiclass Epileptor, a recently proposed model that captures a wide range of seizure dynamotypes in a cortical mass ("node"). We consider two nodes: the seizure onset zone (node 1), which bursts autonomously, and the potential propagation zone (node 2), which is not independently epileptogenic but can be recruited by node 1. We examine the impact of intrinsic and coupling factors on the likelihood and speed of recruitment, with particular attention to the onset bifurcation of node 1. We also measure the range of onset behaviors observed in node 2 with respect to the onset behavior of node 1. The model predicted that seizures that display baseline shifts at onset are less likely to spread, and spread more slowly, compared to seizures that do not exhibit baseline shifts at onset. Seizures that present with amplitude scaling at onset were unlikely to propagate. Further, the model predicted the potential for unusual combinations of onset dynamics, such as a baseline shift in node 2 but not node 1. We confirmed the possibility for several of these unusual recruitment behaviors in humans using intracranial electroencephalography data. The results of the study provide a theoretical framework for seizure propagation, establishing a basis for innovations in characterization of patients seizure networks and identification of the seizure onset zone. Author SummaryIn this work, we examined how a seizure spreads from one part of the brain to another using a computational model. We modeled two brain regions using the Multiclass Epileptor, which reproduces a range of brain activity patterns associated with seizures. In the model, the first brain node was able to recruit the second brain node into a seizure. The model predicted that the likelihood and speed of seizure spread differ depending on the pattern of brain activity observed at the start of the seizure. We also found that the pattern of brain activity at seizure onset is not necessarily the same pattern seen when the seizure spreads. We confirmed this possibility for mismatched patterns in recordings from human brain. The findings of the study improve our understanding of seizure spread, which lays the groundwork for development of tools to quantify seizure spread and may inform future work in patient-specific brain modeling.

Emotion is key to human communication, inferring emotion in a speakers voice is a cross-cultural and cross-linguistic capability. Electroencephalography (EEG) studies of neural mechanisms supporting emotion perception have reported that early components of the event-related potential (ERP) are modulated by emotion. However, the nature of emotions effect, especially on the P200 component, is disputed. We hypothesised that early acoustic features of emotional utterances might account for ERP modulations previously attributed to emotion. We recorded multi-channel EEG from healthy participants (n = 30) tasked with recognising the emotion of utterances. We used fifty vocalisations in five emotions - anger, happiness, neutral, sadness and pleasure - drawn from the Montreal Affective Voices dataset. We statistically quantified instantaneous associations between ERP amplitudes, emotion categories, and acoustic features, specifically, intensity, pitch, first formant, and second formant. We found that shortly after utterance onset (120-250 ms, i.e., P200, early P300) ERP amplitude for sad vocalisations was less than for other emotional categories. Moreover, ERP amplitude at around 180 ms for happy vocalisation was less than for anger, sadness, and pleasure. Our analysis showed that acoustic intensity explains most of these early-latency effects. We also found that, at longer latency (220-500 ms; late P200, P300) ERP amplitude for neutral vocalisations was less than for other emotional categories. Furthermore, there were also ERP differences between anger and happiness, anger and pleasure, anger and sadness, happiness and pleasure, as well as happiness and sadness in shorter windows during this late period. Acoustic pitch and, to a lesser degree, acoustic intensity explain most of these later effects. We conclude that acoustic features can account for early ERP modulations evoked by emotional utterances. Because previous studies used a variety of stimuli, our result likely resolves previous disputes on emotions effect on P200.

Compared with model-based and phenomenological descriptions of the spatiotemporal dynamics of resting-brain activity, statistical characterizations of resting-state fMRI (rs-fMRI) data remain relatively underexplored. Some sophisticated analysis techniques, such as Mapper-based topological data analysis (TDA) and innovation-driven coactivation pattern analysis (iCAP), can distinguish real data from phase-randomized (PR) surrogates, suggesting that rs-fMRI data are not as simple as stationary Gaussian processes. However, the exact statistical properties that distinguish real rs-fMRI data from PR surrogates have not yet been determined. In this study, we conducted system identification analysis and surrogate data analysis to specify key statistical properties that allow TDA and iCAP to discriminate real rs-fMRI data from PR surrogates. We first analyzed rs-fMRI data concatenated across scans using autoregressive (AR) modeling and found that the scan-concatenated rs-fMRI data were weakly non-Gaussian. However, non-Gaussianity alone was insufficient to reproduce realistic TDA and iCAP results because of non-stationarity across scans. AR modeling of single-scan data revealed that rs-fMRI data were statistically indistinguishable from a Gaussian distribution within a single scan, although TDA and iCAP results still differed between the real data and PR surrogates. A new surrogate dataset designed to preserve non-stationarity successfully reproduced realistic TDA and iCAP results, suggesting that TDA and iCAP likely capture the non-stationarity of rs-fMRI data to distinguish it from PR surrogates. Together, these results indicate approximate Gaussianity and non-stationarity in rs-fMRI data, providing a data-driven and statistical characterization of resting-state brain activity that can serve as a quantitative reference for whole brain simulations and generative models.

Accurate detection of interictal epileptiform discharges (IEDs) in electroencephalography (EEG) plays a crucial role in epilepsy diagnosis. Our work investigates the classification of IEDs using Artificial Neural Networks (ANNs) trained on EEG data represented in both signal and source space. Source waveforms were computed using equivalent current dipole models fitted using either a 1-parameter fixed-orientation or a 3-parameter projection approach, both localized to a single best-fit position during the rising flank of the IED. The ANN was trained on raw and feature-extracted versions of signal space and source space data. Feature extraction significantly improved performance across all domains. The highest accuracy (0.98) was achieved in signal space using Katz Fractional Dimension (KFD). In source space analyses, the 1-parameter and 3-parameter models achieved a maximum accuracy of 0.84, with statistical features performing best for the fixed-orientation model and KFD for the free orientation model. Additionally, annotations from three independent expert markers showed considerable variability, with ANN performance falling within the range of inter-expert agreement. These findings support the potential of ANN-based tools to assist expert evaluation in future clinical workflows.

Obsessive-compulsive disorder (OCD) is a disabling condition that is characterized by disruptions in distributed brain circuit dynamics. However, current network studies predominantly evaluate these circuits by measuring functional synchrony (connectivity) between pairs of regions of interest, potentially overlooking complex higher-order interactions. In this study, we applied a Hodge Laplacian topological framework to investigate these higher-order interactions in OCD. Using a large-scale resting-state fMRI dataset from the ENIGMA-OCD consortium (1,024 OCD patients and 1,028 healthy controls across 28 sites worldwide), we identified significant disruptions in topological loops spanning frontoparietal, default mode, and sensorimotor networks. Crucially, the edges constituting these abnormal loops largely lacked significant pairwise differences, highlighting higher-order multi-nodal disturbances. Subgroup analyses revealed that these disruptions were most pronounced in adult, medicated, and high-severity OCD patients. Our findings suggest that OCD pathology involves abnormal recurrent higher-order multi-region interactions, providing new insights into the brains functional organization and offering potential biomarkers for clinical application.

An increasingly popular approach to investigating the neural bases of speech processing is forward modeling via a multivariate temporal-response function (mTRF). This approach uses stimulus characteristics to predict neural responses, especially in EEG and MEG. A central question in this regard is how best to represent the input stimulus. In the case of speech processing, established representations include the speech envelope or spectrogram, as well as feature-based linguistic representations of phonetic content. However, when multiple representations are used as input, a key challenge is how best to isolate their relative effects. This is particularly challenging because such representations have nonvanishing mutual information. To address this problem, we propose optimizations to the mTRF framework via a novel statistical approach of cyclic permutation. Additionally, we propose methodological improvements to the mTRF model targeting three key challenges: effectively managing spatial and temporal autocorrelations endemic to multi-sensor EEG data; mitigating the effects of endogenous drift; and introducing robust artifact rejection to enhance data quality. To demonstrate the effectiveness of this approach, the novel method was applied to a novel EEG data set of natural language listening in 27 adults with normal hearing. Our data showed that including ICA decomposition, artifact rejection, and cyclic permutations in an mTRF analysis improves the isolation of neural responses specific to phonetic and acoustic input variables. Author SummarySpeech processing happens in different stages. It starts with recognizing basic sounds, then categorizes them into discrete categories called phonemes, and goes on to understanding words and sentences. The multivariate temporal response function (mTRF) is a method for predicting brain activity from different features of the speech stimulus. Features that can be used as input to the mTRF model include acoustic features, such as sound envelopes, as well as more abstract language features, such as phonemes, which are a fundamental building block of words. One problem in speech research is distinguishing neural responses to different features. This is challenging because knowing one feature of the speech stimulus enables educated guesses about others and educated predictions about how this feature will behave in the future. Both of these properties of speech make multivariate temporal statistical analysis more difficult. To address this, we propose changes to the preprocessing of the EEG recordings and a new mathematical model that uses a partially rearranged version of the features of the speech stimulus to isolate the predictive power of a particular type of speech feature.

Objective: Brain activity is measured using noninvasive electrophysiological techniques, such as electroencephalography (EEG) and magnetoencephalography (MEG). Data recorded from sensors outside the skull are regularly transformed into a virtual source space. Brain activity is typically parcellated into anatomical brain areas using an atlas. Then, functional connectivity (FC) is estimated between pairs of regions, with their brain activity characterized by a representative time series extracted from multiple voxel time series (multidimensional), using various techniques. Several FC estimators have been used to quantify FC between pairs of brain areas. In contrast, multivariate extensions of these estimators have been proposed, thereby eliminating the need for representative time series for each brain area. Approach: An appropriate framework for systematically evaluating FC estimators in the virtual MEG space and across multiple processing steps for brain network construction is missing. Here, we compared an exhaustive set of bivariate FC estimators with techniques for extracting representative time series, their multivariate extensions, and multivariate estimators for detecting MCI subjects versus healthy controls, using a k-NN classifier and an appropriate graph distance metric. Main Results: Our results demonstrate that the multivariate extension of bivariate FC estimators (representative-free approach), which summarizes pairwise FC strength across all voxels of two brain areas, and accurate multivariate estimators that consider pairs of region-wise voxel time series at once, clearly outperform bivariate FC estimators based on representative time series. Significance: Multivariate extension of bivariate FC estimators and multivariate FC estimators are the natural alternatives to the combination of representative time series per brain area and bivariate FC estimators.