Calibrating Deep Learning Classifiers for Patient-Independent Electroencephalogram Seizure Forecasting.

The recent scientific literature abounds in proposals of seizure forecasting methods that exploit machine learning to automatically analyze electroencephalogram (EEG) signals. Deep learning algorithms seem to achieve a particularly remarkable performance, suggesting that the implementation of clinical devices for seizure prediction might be within reach. However, most of the research evaluated the robustness of automatic forecasting methods through randomized cross-validation techniques, while clinical applications require much more stringent validation based on patient-independent testing. In this study, we show that automatic seizure forecasting can be performed, to some extent, even on independent patients who have never been seen during the training phase, thanks to the implementation of a simple calibration pipeline that can fine-tune deep learning models, even on a single epileptic event recorded from a new patient. We evaluate our calibration procedure using two datasets containing EEG signals recorded from a large cohort of epileptic subjects, demonstrating that the forecast accuracy of deep learning methods can increase on average by more than 20%, and that performance improves systematically in all independent patients. We further show that our calibration procedure works best for deep learning models, but can also be successfully applied to machine learning algorithms based on engineered signal features. Although our method still requires at least one epileptic event per patient to calibrate the forecasting model, we conclude that focusing on realistic validation methods allows to more reliably compare different machine learning approaches for seizure prediction, enabling the implementation of robust and effective forecasting systems that can be used in daily healthcare practice.

Altered spreading of neuronal avalanches in temporal lobe epilepsy relates to cognitive performance: A resting-state hdEEG study.

OBJECTIVE: Large aperiodic bursts of activations named neuronal avalanches have been used to characterize whole-brain activity, as their presence typically relates to optimal dynamics. Epilepsy is characterized by alterations in large-scale brain network dynamics. Here we exploited neuronal avalanches to characterize differences in electroencephalography (EEG) basal activity, free from seizures and/or interictal spikes, between patients with temporal lobe epilepsy (TLE) and matched controls. METHOD: We defined neuronal avalanches as starting when the z-scored source-reconstructed EEG signals crossed a specific threshold in any region and ending when all regions returned to baseline. This technique avoids data manipulation or assumptions of signal stationarity, focusing on the aperiodic, scale-free components of the signals. We computed individual avalanche transition matrices to track the probability of avalanche spreading across any two regions, compared them between patients and controls, and related them to memory performance in patients. RESULTS: We observed a robust topography of significant edges clustering in regions functionally and structurally relevant for the TLE, such as the entorhinal cortex, the inferior parietal and fusiform area, the inferior temporal gyrus, and the anterior cingulate cortex. We detected a significant correlation between the centrality of the entorhinal cortex in the transition matrix and the long-term memory performance (delay recall Rey-Osterrieth Complex Figure Test). SIGNIFICANCE: Our results show that the propagation patterns of large-scale neuronal avalanches are altered in TLE during the resting state, suggesting a potential diagnostic application in epilepsy. Furthermore, the relationship between specific patterns of propagation and memory performance support the neurophysiological relevance of neuronal avalanches.

Challenges and new perspectives of developmental cognitive EEG studies.

Despite shared procedures with adults, electroencephalography (EEG) in early development presents many specificities that need to be considered for good quality data collection. In this paper, we provide an overview of the most representative early cognitive developmental EEG studies focusing on the specificities of this neuroimaging technique in young participants, such as attrition and artifacts. We also summarize the most representative results in developmental EEG research obtained in the time and time-frequency domains and use more advanced signal processing methods. Finally, we briefly introduce three recent standardized pipelines that will help promote replicability and comparability across experiments and ages. While this paper does not claim to be exhaustive, it aims to give a sufficiently large overview of the challenges and solutions available to conduct robust cognitive developmental EEG studies.

Behavioral and Socio-Emotional Disorders in Intellectual Giftedness: A Systematic Review.

This work systematically reviewed past literature to investigate the association between intellectual giftedness and socio-emotional and/or behavioral disorders. Nineteen studies met the inclusion criteria, 17 of which have children and/or adolescents as participants, and 12 have a non-gifted control group. Socio-emotional problems, such as withdrawal, were found in 3 out of 8 studies; internalizing disorders, such as anxiety, were found in 5 out of 9; externalizing disorders, such as hyperactivity, were found in 3 out of 5. The most investigated comorbidity was attention-deficit/hyperactivity disorder. A univocal conclusion on the relationship between intellectual giftedness and socio-emotional/behavioral problems cannot be drawn, principally because of the heterogeneity of participants' age, informants, and instruments. The review highlights the need for future studies to use multi-informant and comprehensive assessments, to reach more robust findings, and suggests that age and discrepancy between verbal and non-verbal intellectual abilities should be considered critical factors.

Implicit cognitive flexibility in self-limited focal epilepsy of childhood: An HD-EEG study.

Self-limited focal epilepsy of childhood (SFEC) is often related to mild impairments in several neuropsychological domains, including cognitive flexibility, which is generally considered a process requiring volition and attention. However, recent evidence showed that it can be implicitly adjusted exploiting simple 'stimulus-response' associations as for example, the probability of the stimulus occurrence. Here, we evaluated the capability to implicitly extract environmental patterns of regularities and use them to flexibly adjust proactive control motor control. We tested 21 children with epilepsy (total IQ > 80; 13 with Childhood epilepsy with centro-temporal spikes, 8 with Panayiotopoulos syndrome (PS); 5-13 years old) compared to a healthy age-matched control group (32 participants). We used the Dynamic Temporal Prediction (DTP) task to investigate how behavioral performance is implicitly shaped by the manipulation of the stimulus occurrence probability over time. We recorded EEG to identify neural markers to differentiate the two groups. SFEC group showed a reduction in accuracy (p = .0013) and response speed (p < .001) as well as an absence of response adjustment (p = .65) in relation to the implicit changes in stimulus probability occurrence, in comparison to the control group. The epilepsy group performance in the DTP showed a significant correlation with the phonemic fluency (r = -0.50) and the Perseverations index of the CPT test (r = 0.53). Finally, children with SFEC did not show the modulation of the contingent negative variation (CNV) evoked potential. Overall, children with SFEC showed poor implicit flexibility compared to a control group. This pattern is individually related to high-level executive function, suggesting to extend neuropsychological assessment to the implicit domain.

What's next? Neural correlates of emotional predictions: A high-density EEG investigation.

Emotions were recently reconsidered as predictions, constructed by the brain (generation stage) to prearrange action (implementation stage), and update internal models according to incoming stimuli (updating stage). However, it is unclear how emotional predictions are shaped by stimuli predictability. This study investigated the role of stimuli predictability on emotional predictions through high-density EEG. Twenty-six undergraduates underwent a S1-S2 paradigm, with emotional faces as S1s and emotional pictures as S2s. Stimuli predictability was manipulated across three blocks, in which S1 valence was predictive of S2 in the 100%, 75%, or 50% of trials. ERPs and brain sources were analyzed for each stage. During prediction generation, a larger N170/superior temporal sulcus activity emerged to fearful faces in blocks with full (100%) and medium (75%) predictive ratios. During implementation, the random block (50%) elicited a valence-independent pre-allocation of resources, reflected by a larger CNV and activation of a wide left network. In the updating stage, emotional pictures always elicited a larger LPP, while a larger P2 to neutral stimuli and a higher activity of the orbitofrontal cortex signaled early valence-dependent and late block-dependent prediction errors. These findings provide the first evidence of how stimuli predictability shape each neurocomputational stage of emotional predictions construction.

The developing predictive brain: How implicit temporal expectancy induced by local and global prediction shapes action preparation across development.

Human behavior is continuously shaped not just as a function of explicitly responding to external world events but also by internal biases implicitly driven by the capacity to extract statistics from complex sensory patterns. Two possible sources of predictability engaged to generate and update temporal expectancy are the implicit extraction of either local or global statistical contingencies in the events' temporal structure. In the context of action preparation the local prediction has been reported to be stable from the age of 6. However, there is no evidence about how the ability to extract and use global statistical patterns to establish temporal expectancy changes across development. Here we used a new, child-friendly reaction time task purposely designed to investigate how local (within-trial expectancy bias) and global (between-block expectancy bias) prediction interplay to generate temporal expectancy and consequently shape action preparation in young (5- to 6-year-old), middle-aged (7- to 8-year-old) and old (9- to 10-year-old) typically developing children. We found that while local temporal prediction showed stable developmental trajectories, the ability to use a global rule to action preparation in terms of both accuracy and speed becomes stable after the age of seven. These findings are discussed by adopting a neuroconstructivist-inspired theoretical account, according to which the developmental constraints on learning from hierarchically nested levels of sensory complexity may constitute a necessary prerequisite for mastering complex domains.

Prompting future events: Effects of temporal cueing and time on task on brain preparation to action.

Prediction about event timing plays a leading role in organizing and optimizing behavior. We recorded anticipatory brain activities and evaluated whether temporal orienting processes are reflected by the novel prefrontal negative (pN) component, as already shown for the contingent negative variation (CNV). Fourteen young healthy participants underwent EEG and fMRI recordings in separate sessions; they were asked to perform a Go/No-Go task in which temporal orienting was manipulated: the external condition (a visual display indicating the time of stimulus onset) and the internal condition (time information not provided). In both conditions, the source of the pN was localized in the pars opercularis of the iFg; the source of the CNV was localized in the supplementary motor area and cingulate motor area, as expected. Anticipatory activity was also found in the occipital-parietal cortex. Time on task EEG analysis showed a marked learning effect in the internal condition, while the effect was minor in the external condition. In fMRI, the two conditions had a similar pattern; similarities and differences of results obtained with the two techniques are discussed. Overall, data are consistent with the view that the pN reflects a proactive cognitive control, including temporal orienting.

Cross-frequency Phase-Amplitude Coupling as a Mechanism for Temporal Orienting of Attention in Childhood.

Temporal orienting of attention operates by biasing the allocation of cognitive and motor resources in specific moments in time, resulting in the improved processing of information from expected compared with unexpected targets. Recent findings have shown that temporal orienting operates relatively early across development, suggesting that this attentional mechanism plays a core role for human cognition. However, the exact neurophysiological mechanisms allowing children to attune their attention over time are not well understood. In this study, we presented 8- to 12-year-old children with a temporal cueing task designed to test (1) whether anticipatory oscillatory dynamics predict children's behavioral performance on a trial-by-trial basis and (2) whether anticipatory oscillatory neural activity may be supported by cross-frequency phase-amplitude coupling as previously shown in adults. Crucially, we found that, similar to what has been reported in adults, children's ongoing beta rhythm was strongly coupled with their theta rhythm and that the strength of this coupling distinguished validly cued temporal intervals, relative to neutral cued trials. In addition, in long trials, there was an inverse correlation between oscillatory beta power and children's trial-by-trial reaction, consistent with oscillatory beta power reflecting better response preparation. These findings provide the first experimental evidence that temporal attention in children operates by exploiting oscillatory mechanism.

Subclinical executive function impairment in children with asymptomatic, treated phenylketonuria: A comparison with children with immunodeficiency virus.

In this study we compared the neuropsychological profile of phenylketonuria (PKU) and human immunodeficiency virus (HIV) to examine the specificity of the executive function (EF) impairment reported in these two patologies. A total of 55 age-matched children and adolescents were assessed, including 11 patients with PKU, 16 patients with HIV and 28 healthy controls, underwent a neuropsychological assessment. Although neither the PKU nor the HIV group scored below the normative ranges, both groups showed lower scores in neuropsychological tests engaging EFs than controls. In addition, compared to patients with PKU the HIV group performed significantly worse in the Trail-Making Test A, Corsi Span and Verbal Fluency. These findings suggest that EF impairments in PKU (a) are limited to EFs (i.e., working memory and attentional shifting), (b) are not simply due to generalized processing speed deficits and

Unpleasant stimuli differentially modulate inhibitory processes in an emotional Go/NoGo task: an event-related potential study.

Threat stimuli typically elicit a psychophysiological response pattern supporting the organism's preparation for active defence. Differently, blood stimuli prompt a distinctive autonomic response pattern and sustained processing, which do not call for clear-cut mobilisation for action. However, the contribution of motor disposition in these response patterns remains unclear. One way to address this issue is to investigate whether threat and blood stimuli differentially affect the active suppression of an ongoing motor activity. Thirty-two undergraduates were presented with threat, mutilation, pleasant, and neutral pictures in an emotional Go/NoGo task. The amplitudes of the NoGo-N2 and NoGo-P3 components of the event-related potentials were analysed as indices of conflict monitoring and inhibition of motor response, respectively. Reaction times to Go trials were significantly faster for threat than for mutilations. The NoGo-N2 was significantly larger to threat than to mutilations, whereas the NoGo-P3amplitude did not differ between the two conditions. These findings suggest that threat stimuli facilitated the execution of a prepotent response and enhanced conflict monitoring when action must be withheld. In contrast, blood stimuli did not either promote action in the Go trials or increase conflict in the NoGo condition, suggesting a response pattern compatible with defensive immobility.

Spatiotemporal neurodynamics underlying internally and externally driven temporal prediction: a high spatial resolution ERP study.

Temporal prediction (TP) is a flexible and dynamic cognitive ability. Depending on the internal or external nature of information exploited to generate TP, distinct cognitive and brain mechanisms are engaged with the same final goal of reducing uncertainty about the future. In this study, we investigated the specific brain mechanisms involved in internally and externally driven TP. To this end, we employed an experimental paradigm purposely designed to elicit and compare externally and internally driven TP and a combined approach based on the application of a distributed source reconstruction modeling on a high spatial resolution electrophysiological data array. Specific spatiotemporal ERP signatures were identified, with significant modulation of contingent negative variation and frontal late sustained positivity in external and internal TP contexts, respectively. These different electrophysiological patterns were supported by the engagement of distinct neural networks, including a left sensorimotor and a prefrontal circuit for externally and internally driven TP, respectively.

Female gender doubles executive dysfunction risk in ALS: a case-control study in 165 patients.

BACKGROUND: Cognitive impairment, mainly characterised by executive dysfunction, occurs in about half of cases in amyotrophic lateral sclerosis (ALS). There is evidence that gender influences some clinical features of the disease, but its influence on the cognitive spectrum is unknown. Our objective was to investigate the impact of gender on cognitive profiles of patients with ALS. METHODS: A retrospective study based on an exhaustive neuropsychological battery was performed in a group of 165 (70 females, 95 males) sporadic, non-demented patients with ALS compared with 134 healthy control participants. This assessment primarily focused on executive, memory and language functions. RESULTS: 47 (29%) patients revealed impairment in executive function and 30 (18%) patients revealed cognitive non-executive impairment. Independent from mood tone and clinical variables, a significantly greater executive impairment was determined in female patients than in male patients and control participants. The relative risk for ALS females having impairment in executive function compared with male patients was 2.6 (95% CI 1.6 to 4.4; p=0.0003). ALS females scored lower in Phonemic Fluency, Trial Making, and Wisconsin Card Sorting test. CONCLUSIONS: Results highlight a significant vulnerability of ALS female patients to develop cognitive dysfunctions peculiar to the disease, independently of bulbar onset. The explicative hypotheses of the data are focused on two interpretative lines not mutually exclusive: the role of gonadal hormones and gender-related brain asymmetry pre-existing to the disease. These findings, never reported before in the literature, can have important implications for models of ALS pathogenesis and for future clinical trial designs.

Neural correlates of attention to emotional facial expressions in dysphoria.

The present study investigated whether dysphoric individuals have a difficulty in disengaging attention from negative stimuli and/or reduced attention to positive information. Sad, neutral and happy facial stimuli were presented in an attention-shifting task to 18 dysphoric and 18 control participants. Reaction times to neutral shapes (squares and diamonds) and the event-related potentials to emotional faces were recorded. Dysphoric individuals did not show impaired attentional disengagement from sad faces or facilitated disengagement from happy faces. Right occipital lateralisation of P100 was absent in dysphoric individuals, possibly indicating reduced attention-related sensory facilitation for faces. Frontal P200 was largest for sad faces among dysphoric individuals, whereas controls showed larger amplitude to both sad and happy as compared with neutral expressions, suggesting that dysphoric individuals deployed early attention to sad, but not happy, expressions. Importantly, the results were obtained controlling for the participants' trait anxiety. We conclude that at least under some circumstances the presence of depressive symptoms can modulate early, automatic stages of emotional processing.

How previous experience shapes future affective subjective ratings: A follow-up study investigating implicit learning and cue ambiguity.

People use their previous experience to predict future affective events. Since we live in ever-changing environments, affective predictions must generalize from past contexts (from which they may be implicitly learned) to new, potentially ambiguous contexts. This study investigated how past (un)certain relationships influence subjective experience following new ambiguous cues, and whether past relationships can be learned implicitly. Two S1-S2 paradigms were employed as learning and test phases in two experiments. S1s were colored circles, S2s negative or neutral affective pictures. Participants (Experiment 1 N = 121, Experiment 2 N = 116) were assigned to the certain (CG) or uncertain group (UG), and they were presented with 100% (CG) or 50% (UG) S1-S2 congruency during an uninstructed (Experiment 1) or implicit (Experiment 2) learning phase. During the test phase both groups were presented with a new 75% S1-S2 paradigm, and ambiguous (Experiment 1) or unambiguous (Experiment 2) S1s. Participants were asked to rate the expected valence of upcoming S2s (expectancy ratings), or their experienced valence and arousal (valence and arousal ratings). In Experiment 1 ambiguous cues elicited less negative expectancy ratings, and less unpleasant valence ratings, independently of prior experience. In Experiment 2, both groups showed similar expectancies, predicting upcoming pictures' valence according to the 75% contingencies of the test phase. Overall, we found that in the presence of ambiguous cues subjective affective experience is dampened, and that implicit previous experience does not emerge at the subjective level by significantly shaping reported affective experience.

Faces do not guide attention in an object-based facilitation manner.

Numerous studies on face processing have revealed their special ability to affect attention, but relatively little research has been done on how faces guide spatial attention allocation. To enrich this field, this study resorted to the object-based attention (OBA) effect in a modified double-rectangle paradigm where the rectangles were replaced with human faces and mosaic patterns (non-face objects). Experiment 1 replicated the typical OBA effect in the non-face objects, but this effect was absent in Asian and Caucasian faces. Experiment 2 removed the eye region from Asian faces, but still found no object-based facilitation in the faces without eyes. In Experiment 3, the OBA effect was also observed for faces when the faces disappear a short period before the responses. Overall, these results revealed that when two faces are presented together, they do not exert object-based facilitation regardless of their facial features such as race and the presence of eyes. We argue that the lack of a typical OBA effect is due to the filtering cost induced by the entire face content. This cost slows down the response when attention shifts within a face and results in the absence of object-based facilitation.

Little fast, little slow, should I stay or should I go? Adapting cognitive control to local-global temporal prediction across typical development.

Adaptive cognitive control (CC), the ability to adjust goal-directed behavior according to changing environmental demand, can be instantiated bottom-up by implicit knowledge, including temporal predictability of task-relevant events. In S1-S2 tasks, either local (trial-by-trial hazard expectation) or global (block-by-block expectation) temporal information can induce prediction, allowing for proactive action control. Recent developmental evidence showed that adaptive CC based on global temporal prediction emerges earlier than when it is based on the local one only. However, very little is known about how children learn to dynamically adjust behavior on the fly according to changing global predictive information. Addressing this issue is nevertheless crucial to unravel the mechanisms underlying adaptive CC flexibility. Here we used a modified version of the Dynamic Temporal Prediction task to investigate how typically developing younger (6-8 years) and older children (9-11 years), adolescents (12-15 years) and adults (21-31 years) use global prediction to shape adaptive CC over time. Specifically, the short-long percentage of S2 preparatory intervals was manipulated list-wide to create a slow-fast-slow-fast fixed block sequence and test how efficiently the response speed adapted accordingly. Overall, results revealed that in all groups behavioral performance is successfully adjusted as a function of global prediction in the late phase of the task (block 3 to 4). Remarkably, only adolescents and adults exhibit an early adaptation of adaptive CC (block 1 to 2), while children younger than 11 show sluggish ability in inferring implicit changes in global predictive rules. This age-related dissociation suggests that, although being present from an early age, adaptive CC based on global predictive information needs more developmental space to become flexible in an efficient way. In the light of a neuroconstructivist approach, we suggest that bottom-up driven implicit flexibility may represent a key prerequisite for the development of efficient explicit cognitive control.

Can faces affect object-based attention? Evidence from online experiments.

This study tested how human faces affect object-based attention (OBA) through two online experiments in a modified double-rectangle paradigm. The results of Experiment 1 revealed that faces did not elicit the OBA effect as non-face objects, which was caused by a longer response time (RT) when attention is focused on faces relative to non-face objects. In addition, by observing faster RTs when attention was engaged horizontally rather than vertically, we found a significant horizontal attention bias, which might override the OBA effect if vertical rectangles were the only items presented; these results were replicated in Experiment 2 (using only vertical rectangles) after directly measuring horizontal bias and excluding its influence on the OBA effect. This study suggested that faces cannot elicit the same-object advantage in the double-rectangle paradigm and provided a method to measure the OBA effect free from horizontal bias.

Face specific neural anticipatory activity in infants 4 and 9 months old.

The possibility of predicting the specific features of forthcoming environmental events is fundamental for our survival since it allows us to proactively regulate our behaviour, enhancing our chance of survival. This is particularly crucial for stimuli providing socially relevant information for communication and interaction, such as faces. While it has been consistently demonstrated that the human brain shows preferential and ontogenetically early face-evoked activity, it is unknown whether specialized neural routes are engaged by face-predictive activity early in life. In this study, we recorded high-density electrophysiological (ERP) activity in adults and 9- and 4-month-old infants undergoing an audio-visual paradigm purposely designed to predict the appearance of faces or objects starting from congruent auditory cues (i.e., human voice vs nonhuman sounds). Contingent negative variation or CNV was measured to investigate anticipatory activity as a reliable marker of stimulus expectancy even in the absence of explicit motor demand. The results suggest that CNV can also be reliably elicited in the youngest group of 4-month-old infants, providing further evidence that expectation-related anticipatory activity is an intrinsic, early property of the human cortex. Crucially, the findings also indicate that the predictive information provided by the cue (i.e., human voice vs nonhuman sounds) turns into the recruitment of different anticipatory neural dynamics for faces and objects.

Emotion regulation strategies differentially modulate neural activity across affective prediction stages: An HD-EEG investigation.

Emotion regulation (ER) strategies can influence how affective predictions are constructed by the brain (generation stage) to prearrange action (implementation stage) and update internal models according to incoming stimuli (updating stage). However, neurocomputational mechanisms by which this is achieved are unclear. We investigated through high-density EEG if different ER strategies (expressive suppression vs. cognitive reappraisal) predicted event-related potentials (ERPs) and brain source activity across affective prediction stages, as a function of contextual uncertainty. An S1-S2 paradigm with emotional faces and pictures as S1s and S2s was presented to 36 undergraduates. Contextual uncertainty was manipulated across three blocks with 100, 75, or 50% S1-S2 affective congruency. The effects of ER strategies, as assessed through the Emotion Regulation Questionnaire, on ERP and brain source activity were tested for each prediction stage through linear mixed-effects models. No ER strategy affected prediction generation. During implementation, in the 75% block, a higher tendency to suppress emotions predicted higher activity in the left supplementary motor area at 1,500-2,000 ms post-stimulus, and smaller amplitude of the Contingent Negative Variation at 2,000-2,500 ms. During updating, in the 75% block, a higher tendency to cognitively reappraise emotions predicted larger P2, Late Positive Potential, and right orbitofrontal cortex activity. These results suggest that both ER strategies interact with the levels of contextual uncertainty by differently modulating ERPs and source activity, and that different strategies are deployed in a moderately predictive context, supporting the efficient updating of affective predictive models only in the context in which model updating occurs.