Efficient Low-Frequency SSVEP Detection with Wearable EEG Using Normalized Canonical Correlation Analysis.

Recent studies show that the integrity of core perceptual and cognitive functions may be tested in a short time with Steady-State Visual Evoked Potentials (SSVEP) with low stimulation frequencies, between 1 and 10 Hz. Wearable EEG systems provide unique opportunities to test these brain functions on diverse populations in out-of-the-lab conditions. However, they also pose significant challenges as the number of EEG channels is typically limited, and the recording conditions might induce high noise levels, particularly for low frequencies. Here we tested the performance of Normalized Canonical Correlation Analysis (NCCA), a frequency-normalized version of CCA, to quantify SSVEP from wearable EEG data with stimulation frequencies ranging from 1 to 10 Hz. We validated NCCA on data collected with an 8-channel wearable wireless EEG system based on BioWolf, a compact, ultra-light, ultra-low-power recording platform. The results show that NCCA correctly and rapidly detects SSVEP at the stimulation frequency within a few cycles of stimulation, even at the lowest frequency (4 s recordings are sufficient for a stimulation frequency of 1 Hz), outperforming a state-of-the-art normalized power spectral measure. Importantly, no preliminary artifact correction or channel selection was required. Potential applications of these results to research and clinical studies are discussed.

Adaptable and Robust EEG Bad Channel Detection Using Local Outlier Factor (LOF).

Electroencephalogram (EEG) data are typically affected by artifacts. The detection and removal of bad channels (i.e., with poor signal-to-noise ratio) is a crucial initial step. EEG data acquired from different populations require different cleaning strategies due to the inherent differences in the data quality, the artifacts' nature, and the employed experimental paradigm. To deal with such differences, we propose a robust EEG bad channel detection method based on the Local Outlier Factor (LOF) algorithm. Unlike most existing bad channel detection algorithms that look for the global distribution of channels, LOF identifies bad channels relative to the local cluster of channels, which makes it adaptable to any kind of EEG. To test the performance and versatility of the proposed algorithm, we validated it on EEG acquired from three populations (newborns, infants, and adults) and using two experimental paradigms (event-related and frequency-tagging). We found that LOF can be applied to all kinds of EEG data after calibrating its main hyperparameter: the LOF threshold. We benchmarked the performance of our approach with the existing state-of-the-art (SoA) bad channel detection methods. We found that LOF outperforms all of them by improving the F1 Score, our chosen performance metric, by about 40% for newborns and infants and 87.5% for adults.

NEAR: An artifact removal pipeline for human newborn EEG data.

Electroencephalography (EEG) is arising as a valuable method to investigate neurocognitive functions shortly after birth. However, obtaining high-quality EEG data from human newborn recordings is challenging. Compared to adults and older infants, datasets are typically much shorter due to newborns' limited attentional span and much noisier due to non-stereotyped artifacts mainly caused by uncontrollable movements. We propose Newborn EEG Artifact Removal (NEAR), a pipeline for EEG artifact removal designed explicitly for human newborns. NEAR is based on two key steps: 1) A novel bad channel detection tool based on the Local Outlier Factor (LOF), a robust outlier detection algorithm; 2) A parameter calibration procedure for adapting to newborn EEG data the algorithm Artifacts Subspace Reconstruction (ASR), developed for artifact removal in mobile adult EEG. Tests on simulated data showed that NEAR outperforms existing methods in removing representative newborn non-stereotypical artifacts. NEAR was validated on two developmental populations (newborns and 9-month-old infants) recorded with two different experimental designs (frequency-tagging and ERP). Results show that NEAR artifact removal successfully reproduces established EEG responses from noisy datasets, with a higher statistical significance than the one obtained by existing artifact removal methods. The EEGLAB-based NEAR pipeline is freely available at https://github.com/vpKumaravel/NEAR.

Efficient Artifact Removal from Low-Density Wearable EEG using Artifacts Subspace Reconstruction.

Light-weight, minimally-obtrusive mobile EEG systems with a small number of electrodes (i.e., low-density) allow for convenient monitoring of the brain activity in out-of-the-lab conditions. However, they pose a higher risk for signal contamination with non-stereotypical artifacts due to hardware limitations and the challenging environment where signals are collected. A promising solution is Artifacts Subspace Reconstruction (ASR), a component-based approach that can automatically remove non-stationary transient-like artifacts in EEG data. Since ASR has only been validated with high-density systems, it is unclear whether it is equally efficient on low-density portable EEG. This paper presents a complete analysis of ASR performance based on clean and contaminated datasets acquired with BioWolf, an Ultra-Low-Power system featuring only eight channels, during SSVEP sessions recorded from six adults. Empirical results show that even with such few channels, ASR efficiently corrects artifacts, enabling an overall enhancement of up to 40% in SSVEP response. Furthermore, by choosing the optimal ASR parameters on a single-subject basis, SSVEP response can be further increased to more than 45%. These results suggest that ASR is a viable and robust method for online automatic artifact correction with low-density BCI systems in real-life scenarios.

The neural representation of absolute direction during mental navigation in conceptual spaces.

When humans mentally "navigate" bidimensional uniform conceptual spaces, they recruit the same grid-like and distance codes typically evoked when exploring the physical environment. Here, using fMRI, we show evidence that conceptual navigation also elicits another kind of spatial code: that of absolute direction. This code is mostly localized in the medial parietal cortex, where its strength predicts participants' comparative semantic judgments. It may provide a complementary mechanism for conceptual navigation outside the hippocampal formation.

Grid-like and distance codes for representing word meaning in the human brain.

Relational information about items in memory is thought to be represented in our brain thanks to an internal comprehensive model, also referred to as a "cognitive map". In the human neuroimaging literature, two signatures of bi-dimensional cognitive maps have been reported: the grid-like code and the distance-dependent code. While these kinds of representation were previously observed during spatial navigation and, more recently, during processing of perceptual stimuli, it is still an open question whether they also underlie the representation of the most basic items of language: words. Here we taught human participants the meaning of novel words as arbitrary labels for a set of audiovisual objects varying orthogonally in size and sound. The novel words were therefore conceivable as points in a navigable 2D map of meaning. While subjects performed a word comparison task, we recorded their brain activity using functional magnetic resonance imaging (fMRI). By applying a combination of representational similarity and fMRI-adaptation analyses, we found evidence of (i) a grid-like code, in the right postero-medial entorhinal cortex, representing the relative angular positions of words in the word space, and (ii) a distance-dependent code, in medial prefrontal, orbitofrontal, and mid-cingulate cortices, representing the Euclidean distance between words. Additionally, we found evidence that the brain also separately represents the single dimensions of word meaning: their implied size, encoded in visual areas, and their implied sound, in Heschl's gyrus/Insula. These results support the idea that the meaning of words, when they are organized along two dimensions, is represented in the human brain across multiple maps of different dimensionality. SIGNIFICANT STATEMENT: How do we represent the meaning of words and perform comparative judgements on them in our brain? According to influential theories, concepts are conceivable as points of an internal map (where distance represents similarity) that, as the physical space, can be mentally navigated. Here we use fMRI to show that when humans compare newly learnt words, they recruit a grid-like and a distance code, the same types of neural codes that, in mammals, represent relations between locations in the environment and support physical navigation between them.

Cortical route for facelike pattern processing in human newborns.

Humans are endowed with an exceptional ability for detecting faces, a competence that, in adults, is supported by a set of face-specific cortical patches. Human newborns, already shortly after birth, preferentially orient to faces, even when they are presented in the form of highly schematic geometrical patterns vs. perceptually equivalent nonfacelike stimuli. The neural substrates underlying this early preference are still largely unexplored. Is the adult face-specific cortical circuit already active at birth, or does its specialization develop slowly as a function of experience and/or maturation? We measured EEG responses in 1- to 4-day-old awake, attentive human newborns to schematic facelike patterns and nonfacelike control stimuli, visually presented with slow oscillatory "peekaboo" dynamics (0.8 Hz) in a frequency-tagging design. Despite the limited duration of newborns' attention, reliable frequency-tagged responses could be estimated for each stimulus from the peak of the EEG power spectrum at the stimulation frequency. Upright facelike stimuli elicited a significantly stronger frequency-tagged response than inverted facelike controls in a large set of electrodes. Source reconstruction of the underlying cortical activity revealed the recruitment of a partially right-lateralized network comprising lateral occipitotemporal and medial parietal areas overlapping with the adult face-processing circuit. This result suggests that the cortical route specialized in face processing is already functional at birth.

Conceptual and Perceptual Dimensions of Word Meaning Are Recovered Rapidly and in Parallel during Reading.

A single word (the noun "elephant") encapsulates a complex multidimensional meaning, including both perceptual ("big", "gray", "trumpeting") and conceptual ("mammal", "can be found in India") features. Opposing theories make different predictions as to whether different features (also conceivable as dimensions of the semantic space) are stored in similar neural regions and recovered with similar temporal dynamics during word reading. In this magnetoencephalography study, we tracked the brain activity of healthy human participants while reading single words varying orthogonally across three semantic dimensions: two perceptual ones (i.e., the average implied real-world size and the average strength of association with a prototypical sound) and a conceptual one (i.e., the semantic category). The results indicate that perceptual and conceptual representations are supported by partially segregated neural networks: Whereas visual and auditory dimensions are encoded in the phase coherence of low-frequency oscillations of occipital and superior temporal regions, respectively, semantic features are encoded in the power of low-frequency oscillations of anterior temporal and inferior parietal areas. However, despite the differences, these representations appear to emerge at the same latency: around 200 msec after stimulus onset. Taken together, these findings suggest that perceptual and conceptual dimensions of the semantic space are recovered automatically, rapidly, and in parallel during word reading.

A machine learning approach for automated wide-range frequency tagging analysis in embedded neuromonitoring systems.

EEG is a standard non-invasive technique used in neural disease diagnostics and neurosciences. Frequency-tagging is an increasingly popular experimental paradigm that efficiently tests brain function by measuring EEG responses to periodic stimulation. Recently, frequency-tagging paradigms have proven successful with low stimulation frequencies (0.5-6Hz), but the EEG signal is intrinsically noisy in this frequency range, requiring heavy signal processing and significant human intervention for response estimation. This limits the possibility to process the EEG on resource-constrained systems and to design smart EEG based devices for automated diagnostic. We propose an algorithm for artifact removal and automated detection of frequency tagging responses in a wide range of stimulation frequencies, which we test on a visual stimulation protocol. The algorithm is rooted on machine learning based pattern recognition techniques and it is tailored for a new generation parallel ultra low power processing platform (PULP), reaching performance of more that 90% accuracy in the frequency detection even for very low stimulation frequencies (<1Hz) with a power budget of 56mW.

Word meaning in the ventral visual path: a perceptual to conceptual gradient of semantic coding.

The meaning of words referring to concrete items is thought of as a multidimensional representation that includes both perceptual (e.g., average size, prototypical color) and conceptual (e.g., taxonomic class) dimensions. Are these different dimensions coded in different brain regions? In healthy human subjects, we tested the presence of a mapping between the implied real object size (a perceptual dimension) and the taxonomic categories at different levels of specificity (conceptual dimensions) of a series of words, and the patterns of brain activity recorded with functional magnetic resonance imaging in six areas along the ventral occipito-temporal cortical path. Combining multivariate pattern classification and representational similarity analysis, we found that the real object size implied by a word appears to be primarily encoded in early visual regions, while the taxonomic category and sub-categorical cluster in more anterior temporal regions. This anteroposterior gradient of information content indicates that different areas along the ventral stream encode complementary dimensions of the semantic space.

Disruption of hierarchical predictive coding during sleep.

When presented with an auditory sequence, the brain acts as a predictive-coding device that extracts regularities in the transition probabilities between sounds and detects unexpected deviations from these regularities. Does such prediction require conscious vigilance, or does it continue to unfold automatically in the sleeping brain? The mismatch negativity and P300 components of the auditory event-related potential, reflecting two steps of auditory novelty detection, have been inconsistently observed in the various sleep stages. To clarify whether these steps remain during sleep, we recorded simultaneous electroencephalographic and magnetoencephalographic signals during wakefulness and during sleep in normal subjects listening to a hierarchical auditory paradigm including short-term (local) and long-term (global) regularities. The global response, reflected in the P300, vanished during sleep, in line with the hypothesis that it is a correlate of high-level conscious error detection. The local mismatch response remained across all sleep stages (N1, N2, and REM sleep), but with an incomplete structure; compared with wakefulness, a specific peak reflecting prediction error vanished during sleep. Those results indicate that sleep leaves initial auditory processing and passive sensory response adaptation intact, but specifically disrupts both short-term and long-term auditory predictive coding.

Contribution of motor representations to action verb processing.

Electrophysiological and brain imaging studies show a somatotopic activation of the premotor cortex while subjects process action verbs. This somatotopic motor activation has been taken as an indication that the meaning of action verbs is embedded in motor representations. However, discrepancies in the literature led to the alternative hypothesis that motor representations are activated during the course of a mental imagery process emerging only after the meaning of the action has been accessed. In order to address this issue, we asked participants to decide whether a visually presented verb was concrete or abstract by pressing a button or a pedal (primary task) and then to provide a distinct vocal response to low and high sounds played soon after the verb display (secondary task). Manipulations of the visual display (lower vs. uppercase), verb imageability (concrete vs. abstract), verb meaning (hand vs. foot-related), and response effector (hand vs. foot) allowed us to trace the perceptual, semantic and response stages of verb processing. We capitalized on the psychological refractory period (PRP), which implies that the initiation of the secondary task should be delayed only by those factors that slow down the central decision process in the primary task. In line with this prediction, our results showed that the time cost resulting from the processing of abstract verbs, when compared to concrete verbs, was still observed in the subsequent response to the sounds, whereas the overall advantage of hand over foot responses did not influence sound judgments. Crucially, we also observed a verb-effector compatibility effect (i.e., foot-related verbs are responded faster with the foot and hand-related verbs with the hand) that contaminated the performance of the secondary task, providing clear evidence that motor interference from verb meaning occurred during the central decision stage. These results cannot be explained by a mental imagery process that would deploy only during the execution of the response to verb judgments. They rather indicate that the motor activation induced by action verbs accompanies the lexico-semantic processes leading to response selection.

Timing the impact of literacy on visual processing.

Learning to read requires the acquisition of an efficient visual procedure for quickly recognizing fine print. Thus, reading practice could induce a perceptual learning effect in early vision. Using functional magnetic resonance imaging (fMRI) in literate and illiterate adults, we previously demonstrated an impact of reading acquisition on both high- and low-level occipitotemporal visual areas, but could not resolve the time course of these effects. To clarify whether literacy affects early vs. late stages of visual processing, we measured event-related potentials to various categories of visual stimuli in healthy adults with variable levels of literacy, including completely illiterate subjects, early-schooled literate subjects, and subjects who learned to read in adulthood (ex-illiterates). The stimuli included written letter strings forming pseudowords, on which literacy is expected to have a major impact, as well as faces, houses, tools, checkerboards, and false fonts. To evaluate the precision with which these stimuli were encoded, we studied repetition effects by presenting the stimuli in pairs composed of repeated, mirrored, or unrelated pictures from the same category. The results indicate that reading ability is correlated with a broad enhancement of early visual processing, including increased repetition suppression, suggesting better exemplar discrimination, and increased mirror discrimination, as early as ∼ 100-150 ms in the left occipitotemporal region. These effects were found with letter strings and false fonts, but also were partially generalized to other visual categories. Thus, learning to read affects the magnitude, precision, and invariance of early visual processing.

ADJUST: An automatic EEG artifact detector based on the joint use of spatial and temporal features.

A successful method for removing artifacts from electroencephalogram (EEG) recordings is Independent Component Analysis (ICA), but its implementation remains largely user-dependent. Here, we propose a completely automatic algorithm (ADJUST) that identifies artifacted independent components by combining stereotyped artifact-specific spatial and temporal features. Features were optimized to capture blinks, eye movements, and generic discontinuities on a feature selection dataset. Validation on a totally different EEG dataset shows that (1) ADJUST's classification of independent components largely matches a manual one by experts (agreement on 95.2% of the data variance), and (2) Removal of the artifacted components detected by ADJUST leads to neat reconstruction of visual and auditory event-related potentials from heavily artifacted data. These results demonstrate that ADJUST provides a fast, efficient, and automatic way to use ICA for artifact removal.

Temporal integration in visual word recognition.

When two displays are presented in close temporal succession at the same location, how does the brain assign them to one versus two conscious percepts? We investigate this issue using a novel reading paradigm in which the odd and even letters of a string are presented alternatively at a variable rate. The results reveal a window of temporal integration during reading, with a nonlinear boundary around approximately 80 msec of presentation duration. Below this limit, the oscillating stimulus is easily fused into a single percept, with all characteristics of normal reading. Above this limit, reading times are severely slowed and suffer from a word-length effect. ERPs indicate that, even at the fastest frequency, the oscillating stimulus elicits synchronous oscillations in posterior visual cortices, while late ERP components sensitive to lexical status vanish beyond the fusion threshold. Thus, the fusion/segregation dilemma is not resolved by retinal or subcortical filtering, but at cortical level by at most 300 msec. The results argue against theories of visual word recognition and letter binding that rely on temporal synchrony or other fine temporal codes.

Investigating the neural correlates of continuous speech computation with frequency-tagged neuroelectric responses.

In order to learn an oral language, humans have to discover words from a continuous signal. Streams of artificial monotonous speech can be readily segmented based on the statistical analysis of the syllables' distribution. This parsing is considerably improved when acoustic cues, such as subliminal pauses, are added suggesting that a different mechanism is involved. Here we used a frequency-tagging approach to explore the neural mechanisms underlying word learning while listening to continuous speech. High-density EEG was recorded in adults listening to a concatenation of either random syllables or tri-syllabic artificial words, with or without subliminal pauses added every three syllables. Peaks in the EEG power spectrum at the frequencies of one and three syllables occurrence were used to tag the perception of a monosyllabic or tri-syllabic structure, respectively. Word streams elicited the suppression of a one-syllable frequency peak, steadily present during random streams, suggesting that syllables are no more perceived as isolated segments but bounded to adjacent syllables. Crucially, three-syllable frequency peaks were only observed during word streams with pauses, and were positively correlated to the explicit recall of the detected words. This result shows that pauses facilitate a fast, explicit and successful extraction of words from continuous speech, and that the frequency-tagging approach is a powerful tool to track brain responses to different hierarchical units of the speech structure.

Chance vs. necessity in living systems: a false antinomy.

The concepts of order and randomness are crucial to understand 'living systems' structural and dynamical rules. In the history of biology, they lay behind the everlasting debate on the relative roles of chance and determinism in evolution. Jacques Monod [1970] built a theory where chance (randomness) and determinism (order) were considered as two complementary aspects of life. In the present paper, we will give an up to date version of the problem going beyond the dichotomy between chance and determinism. To this end, we will first see how the view on living systems has evolved from the mechanistic one of the 19th century to the one stemming from the most recent literature, where they emerge as complex systems continuously evolving through multiple interactions among their components and with the surrounding environment. We will then report on the ever increasing evidence of "friendly" co-existence in living beings between a number of "variability generators", fixed by evolution, and the "spontaneous order" derived from interactions between components. We will propose that the "disorder" generated is "benevolent" because it allows living systems to rapidly adapt to changes in the environment by continuously changing, while keeping their internal harmony.

The correlated nature of large-scale neural activity unveiled by the resting brain.

Most of functional neuroscience studies investigate the brain's response to a task or stimulus. However, the brain is very active even in the absence of an external or internal input. Recent neuroimaging, electrophysiological and optical imaging studies revealed that the neural activity at rest is structured in functionally specific, temporally correlated, spatially distributed patterns that explain a large part of the variability of event-related responses. These results importantly extend the focus of cognitive neuroscience from the stimulus-evoked responses to the spatial and temporal correlations between different neural populations. I argue that integrating the correlated nature of the neural activity to standard measures of evoked responses might significantly advance our understanding on the neural mechanisms underlying cognitive functions. This perspective may be tackled by identifying spontaneous spatiotemporal patterns of neural activity at rest and tracking their evolution and dynamical interactions during cognitive processing.

Variance normalisation: a key mechanism for temporal adaptation in natural vision?

A major problem in natural vision is how neurons in the early visual system encode the widely varying visual input with the limited dynamic range of their activity. Recent experiments suggest that retinal neurons adapt their response not only to the temporal mean but also to the temporal variance of the visual input. Inspired by these results, we propose a simple model in which temporal adaptation can be achieved by a transformation consisting of a linear filtering followed by a variance normalisation. We show that such transformation efficiently adapts to the temporal statistics of natural time series of intensities by removing most of its redundancy, while no linear transformation alone achieves the same goal. Results reproduce important features of temporal adaptation in real vision.