Poster Sessions

Lesion segmentation is of imperative importance in the acute brain stroke diagnosis. However, the lesion boundaries are not always well defined yet knowing the extension of the penumbra region versus the core lesion is deciding factor. We implement a Fully-Convolutional-Network that employs spline convolution in pre-computed graph representation for stroke core lesions segmentation. This model uses graph operations to excerpt interpixel information from a set of multi-modal input in the ISLES-2018 dataset composed of CT-perfusion parameters, namely CBV, CBF, MTT, and TMAX. Our results prove the feasibility of using geometric deep learning models to solve segmentation problems, and we propose a new model architecture that employs graph-based operations to predict acute stroke brain lesions from CT perfusion parameters. We visualize the activation maps to investigate the feature extraction.

To what extant preverbal infants are agents has been one of the central debates in developmental research and theory. Once infants are able to act, their interaction with the environment becomes more purposeful and rewarding. During last years’ meeting we presented a review of existing studies that have investigated agency in the first year of life. This year, we will present the experimental design, logic and preliminary results of an experiment based on the results of the review. Active contingency paradigms indicate that infants detect multi-sensory contingency and have been interpreted as evidence for the development of a sense of agency. We argued that neither of these measures allows conclusions about the presence of agency because they fail to indicate the infant’s control over the movement. Multi-sensory contingency learning, we argued, does not go beyond reinforcement learning. Consequently, it does not tell us whether infants perceive themselves as agents or have any control over their movements. We suggested that, instead, what will inform us about agency, are infants’ reactions to a break-down of contingency, and the extent to which these reactions show infants’ control over and flexible adaption of their actions. Subsequently we combined previous empirical work from different fields to derive an experimental test for agentic control in a non-verbal population. The experimental design, logic and preliminary results of this experiment will be described on the second poster on this project.

Predictive coding theories argue that deviance detection phenomena, such as mismatch responses and omission responses, are generated by predictive processes with possibly overlapping neural substrates. Molecular imaging and electrophysiology studies of mismatch responses and corollary discharge in the rodent model allowed the development of mechanistic and computational models of these phenomena. These models enable translation between human and non-human animal research and help to uncover fundamental features of change-processing microcircuitry in the neocortex. This microcircuitry is characterized by stimulus-specific adaptation and feedforward inhibition of stimulus-selective populations of pyramidal neurons and interneurons, with specific contributions from different interneuron types. The overlap of the substrates of different types of responses to deviant stimuli remains to be understood. Omission responses, which are observed both in corollary discharge and mismatch response protocols in humans, are underutilized in animal research and may be pivotal in uncovering the substrates of predictive processes. Omission studies comprise a range of methods centered on the withholding of an expected stimulus. This poster aims to provide an overview of omission protocols and showcase their potential to integrate and complement the different models and procedures employed to study prediction and deviance detection. This approach may reveal the biological foundations of core concepts of predictive coding, and allow an empirical test of the framework’s promise to unify theoretical models of attention and perception.

Neuromelanin-sensitive MRI receives interest as potential biomarker in neurodegenerative diseases, such as Parkinson’s and Alzheimer’s disease. It is known that human neuromelanin pigments bind large quantities of toxic metal ions, especially iron but also other transition metals including copper. These melanin-iron complexes are a potential source of paramagnetic relaxation enhancement of water proton. In relaxometry investigations, we found deviations from a simple linear concentration-dependent T1 shortening in synthetic neuromelanins containing different amounts of iron and copper. Knowledge of the occupation of distinct metal binding sites seems crucial for contrast optimization or attempts to quantify metal content by MRI. NOTE: ISMRM Submission 4840

Brain computer interfaces (BCIs) provide an alternative pathway for controlling external devices by means of voluntary modulation of the brain activity, for example, through imagined movements, i.e. so called sensorimotor BCI. It might be especially useful for people with motor disabilities as an assistive device or during the course of rehabilitation. However, users first need to learn to control a BCI, which typically requires completing several training sessions. Moreover, studies report that, on average, around 20 % of participants are not able to achieve adequate control in sensorimotor BCI. A number of studies were conducted in order to extract features from neuronal activity that could explain the key to success in BCI control tasks. In particular, it was shown that the following measures allowed predicting control accuracy in the subsequent sessions: (1) signal-to-noise ratio of the mu-rhythm oscillations, (2) multiscale temporal neuronal dynamics quantified by the Hurst exponent, and (3) functional connectivity between sensorimotor areas. The latter connectivity-based feature might be especially interesting since controlling BCI requires a coordinated involvement of several brain areas, and their interaction could play a key role in BCI learning. In the present study, we aim to extend previous findings by analyzing changes in functional connectivity using a longitudinal dataset of learning to control an EEG-based brain-computer interface, where participants (n = 62) performed 7 or 11 sessions of a 2-D cursor control task based on the imaginary movements of their left and right hand. We will examine whether the observed improvements in BCI performance are accompanied by changes in functional connectivity, thus more extensively testing the potential link between connectivity and success in learning to control BCI.

The distributed nature of the neural substrate, and the difficulty of establishing necessity from correlative data, combine to render the mapping of brain function a far harder task than it seems. Methods capable of combining connective anatomical information with focal disruption of function are needed to disambiguate local from global neural dependence, and critical from merely coincidental activity. Here we elaborate a comprehensive, efficient framework for focal and connective spatial inference based on sparse disruptive data, and demonstrate its application in the context of transient direct electrical stimulation of the human medial frontal wall during the pre-surgical evaluation of patients with focal epilepsy. Our framework reformulates within the statistical parametric mapping (SPM) platform the technique of voxel-wise mass-univariate inference with sparsely sampled data originally introduced by imaging meta-analysis, and extends it to encompass the analysis of distributed maps defined by any criterion of connectivity. Applied to the medial frontal wall, this transient dysconnectome approach reveals marked discrepancies between local and distributed associations of major categories of motor, sensory and speech behaviour, revealing differentiation by remote connectivity to which purely local analysis is blind. Our framework enables disruptive mapping of the human brain based on sparsely sampled data with minimal spatial assumptions, good statistical efficiency, flexible model formulation, and explicit comparison of local and distributed effects.

Humans are remarkably successful at visually recognising objects in their surroundings. Crucial for these capabilities are adequate brain representations of these objects and how they relate to one another. For such a representational structure to be useful, the human brain must compute the overall similarity between object pairs by integrating over a set of representational dimensions on which these objects can be evaluated. However, current methods for exhaustively sampling similarities are very costly, as the resources needed grow non-linearly with the size of the image set. To solve this problem, here, we present an efficient method to generate similarity scores of real-world object image pairs by combining human judgements and deep neural network (DNN) activations. Instead of sampling image similarities directly, our method first predicts values for each image on 49 previously derived representational dimensions. We then use these dimension values to generate similarity scores for any pair of images. Finally, we relate these similarity scores to either behavioral or brain similarities for the same images. We evaluated our approach with dimension predictions derived from DNN activations as well as with dimension ratings generated from human subjects in an online crowdsourcing experiment (n = 25 per dimension), utilising several different sets of images either with known dimension values or with known similarity scores as derived from behavioral, MEG or early visual and late visual fMRI data. We found that humans and DNNs were able to predict dimension values for the test image sets across several datasets, leading to very good predictions of similarity. We further combined dimension predictions across humans and DNNs and found a strong additional increase in performance (R^2 = 84-87%). When applying DNN predictions for different image sets to three existing neuroimaging datasets, we found equally strong or even stronger correlations between brain and DNN - generated similarity scores than between brain and behavioral similarities. Altogether, our method promises a more efficient way of generating similarity values of image pairs and thus opens up the possibility of upscaling research involving object image similarities to very large stimulus sets.

The involvement of large-scale cortical networks such as the default-mode network (DMN) and the multiple-demand network (MDN) in semantic processing is generally considered to boost cognitive control under demanding task effort, although results vary greatly depending on specific task demands. Multiple-demand (MD) areas are identifiable by contrasting high- vs. low-demand task conditions across multiple cognitive domains; in the semantic domain, MDN engagement has been shown for a variety of task designs, albeit with a strong bias for language comprehension tasks, involving lexical ambiguity, referential ambiguity, semantic judgement, and degraded speech comprehension. A popular line of meta-analyses has identified areas of functional dissociation and overlap between MDN and mappings of a language-specific semantic network. It remains very hard to differentiate, however, within MDN activity under behavioural task, between correlates of general task-based executive control and semantic processes: a consensus is still lacking about the boundaries of the network relative to the ventral semantic processing network and the nature of its involvement in support of semantic cognition; domain-general executive resources from the MDN may participate to core aspects of semantic control related to lexical access or semantic composition; alternatively, MD regions could only be involved in supporting extrinsic task-related functions and be unsensitive to the semantic properties of the input. This project exploits a brain stimulation to investigate network interactions between MDN areas and semantic regions by focusing on the contribution of the pre-supplementary motor area (pre-SMA), a region of the MDN showing strong functional associations with controlled semantic processing and the left inferior frontal gyrus as core semantic area. The first experimental project is based on a dual-site off-line transcranial magnetic stimulation (TMS) paradigm, testing the effects of left inferior frontal gyrus (L-IFG) inhibition, pre-SMA inhibition and combined inhibition of both areas on semantic and non-semantic fluency. A follow-up study will combine TMS and fMRI to explore potential short-term plasticity effects on controlled semantic processing under pre-SMA inhibition.

Most movements in daily life are embedded in motion sequences, often involving more than one limb. Previous motor research has, however, focused primarily on motion sequences produced by a single limb or on the simultaneous movements of limbs. For example, in unimanual interference tasks in which reaches to a target have to be adapted to two opposing force fields, it has been shown that if the direction of a prior movement (pre-movement) of the same arm is predictive of the force field’s direction, adaptation to the forces is possible. In contrast other cues (e.g., static visual) do not allow this learning to occur. In this study we investigated whether the facilitative effect of pre-movement generalizes to intermanual motion sequences. In addition, we investigated whether only active movements or just perceptual feedback (visual, proprioceptive) contribute to force field-specific motor adaptation. We measured motor learning by direction-specific adaptation during a reach of the right arm in an interference force field paradigm. We compared performance of five experimental groups: 1) no pre-movement, 2) active same-arm pre-movement, 3) active opposite-arm pre-movement, 4) visual opposite-arm pre-movement, and 5) passive opposite-arm pre-movement. In total, movement kinematics of 64 participants were recorded using an exoskeleton robot. We replicated previous findings showing that active same-arm pre-movements facilitate adaptation to opposing force fields, whereas stationary visual cues, though indicative of force field direction, do not allow the formation of strong separate motor memories. In addition, we found that active pre-movements of the opposite-arm enabled adaptation, whereas visual or passive opposite-arm pre-movements were significantly less effective. Our findings indicate that formation of two separate motor memories in an interference task is not only possible when the opposing force fields are encountered in force field-specific intramanual movements, but also in active intermanual motion sequences. However, only providing sensory feedback (vision or proprioception) of the opposite-arm does not seem to affect the moving arm and thus does not appear to be combined into an intermanual motion sequence. This may reflect the ability of the sensorimotor system to only effectively integrate information relating to active movements while not fully utilizing perceptual cues simply signaling the dynamics of the upcoming movement.

Neural oscillations facilitate speech processing by synchronizing to rhythmic acoustic cues in speech. In particular, delta-band oscillations (< 4 Hertz) synchronize with speech prosody. In a series of behavioral studies, we have observed that rhythmic prosodic contours can trigger downstream effects that persist beyond stimulation, affecting the comprehension of upcoming sentences devoid of prosody. This is in line with the finding that via entrainment, oscillations can inherit a stimulation frequency to persist after stimulus offset. To support the interpretation that our behavioural effects reflect electrophysiological entrainment, we conducted an MEG experiment. We combined an initial prosodic rhythm with a subsequent visual target sentence. Target sentences were either long or short (e.g., “Max sees Tom and Karl laughs” vs. “Max sees Tom and Karl”). In a 2 × 2 design, these were combined with prosodic contours that were either long or short (corresponding to the durations of “Max sees Tom and Karl” and “Max sees Tom”, respectively). In the entrainment part of each experimental trial, a contour was repeated 3 times to induce rhythmic entrainment. In the target part, a visual target sentence was presented word by word; presentation was duration-matched to the rate of the previous stimulus. We first hypothesized that delta-band oscillations would entrain to the rate of the contours. Second, we hypothesized that this frequency would still be detectable in the MEG for the visual target sentence. In the entrainment part, we observed coherence with the prosodic contour at the stimulation rate over all MEG sensors (< 0.001, corrected). Coherence indeed persisted into the target part (p < 0.001, corrected), with an anterior shift of the topography. Critically, when long contours were followed by short sentences, a P300 ERF was observed at the offset of the short sentence — likely indicating an omission response under the expectation of a long sentence. Together with our behavioral results, we conclude that sustained prosodic entrainment affects subsequent sentence comprehension, with the stimulation frequency being conserved by brain areas associated with higher-level linguistic processing. To substantiate the apparent shift from bottom-up (= auditory) to top-down (= predictive) brain regions, we are now conducting source reconstruction.

Language comprehension proceeds at a very fast pace. It is argued that context influences the speed of language comprehension by providing informative cues for the correct processing of the incoming linguistic input. Priming studies investigating the role of context in language processing have shown that humans quickly recognise target words that share orthographic, morphological, or semantic information with their preceding primes. How syntactic information influences the processing of incoming words is however less known. Here we employed a masked syntactic priming paradigm in four behavioural experiments in the German language to test whether masked primes automatically facilitate the categorization of nouns and verbs presented as flashing visual words. Overall, we found robust syntactic priming effects with masked primes—thus suggesting high automaticity of the process—but only when verbs were morpho-syntactically marked (‘‘er kau-t’’; ‘‘he chew-s’’). Furthermore, we found that, compared to baseline, primes slow down target categorisation when the relationship between prime and target is syntactically incorrect, rather than speeding it up when the prime-target relationship is syntactically correct. This argues in favour of an inhibitory nature of syntactic priming. Overall, the data indicate that humans automatically extract abstract syntactic features from word categories as flashing visual words, which has an impact on the speed of successful language processing during language comprehension.

Language acquisition quickly tunes speech processing towards infants’ native language: By the age of 6 months, infants show enhanced processing of the speech sounds—so-called phonemes—of their native language (Kuhl et al., 2007). Critically, this entails the ability to delineate corresponding acoustic segments that are as short as 50 ms (Leong & Goswami, 2015). How is this possible, given that infants’ electroencephalogram (EEG) is dominated by slow electrophysiological activity (Anderson & Perone, 2018), only offering long temporal receptive windows (Hochmann & Kouider, 2022)? Phoneme rate is high, but not all features of phonemes change at a high rate. In natural speech, some phonological features (e.g., voicing, place of articulation) can span sequences of multiple adjacent phonemes—thus spanning temporal intervals that match infants’ temporal receptive windows. We thus hypothesized that feature duration could explain the early acquisition of native phoneme properties. We recorded the electroencephalogram (EEG) from a sample of n=75 children aged 0;3–4;6 years. Children heard translation-equivalent stories in their native language (German) and an unfamiliar language (French). From the EEG, we quantified categorical processing of phonemes through the prediction accuracy of feature-based EEG encoding models (temporal response functions, TRFs). For the native language, TRF analysis revealed an increase of prediction accuracy across age, t(73)=4.14, p<.001. Fitted confidence intervals across the age trajectory detected that native categorical processing significantly deviated from a statistical baseline from age 16 months onwards. The developmental time course of individual features showed a significant correlation between the order in which individual features are acquired and the feature duration in continuous speech, r(16)=.71, p=.002)—that is, the longer a feature extends, the earlier infants are sensitive to it. We are now employing TRFs to compare native vs. non-native phoneme processing. Our findings show that the order of phonological feature acquisition depends on feature duration in continuous speech. This suggests that infants build their native phoneme inventory by tracking features, whose longer durations better match infants’ temporal receptive windows than individual phonemes. Thus, infants’ electrophysiological slowness might enable them to initially focus on slowly changing phonological features before progressing to faster features.

Since the effects of transcutaneous spinal direct current stimulation (tsDCS) on human somatosensory evoked potentials were first reported in 2008, a large number of studies on tsDCS have emerged. This extensive literature shows that tsDCS can induce marked changes in spinal cord excitability and modulate the activity in lemniscal, spinothalamic, and segmental motor systems. Here, we aim to provide a review of the progress that has occurred in using tsDCS to modulate spinal cord function non-invasively and establish a data-base that will provide a structured access-point to the tsDCS literature in order to aid planning of future studies. A systematic literature review was conducted by two independent reviewers according to PRISMA guidelines, using PubMed, Scopus, WebOfScience and GoogleScholar. Database keyword searches included English-language studies and excluded studies that applied different spinal cord stimulation techniques or were not peer-reviewed. We categorized included studies into five groups – healthy volunteers, patients, animals, modelling studies and reviews – as this will allow us to provide a detailed overview of tsDCS’s mechanisms, its efficacy on clinical symptoms, and the electric fields produced by different tsDCS montages. Key information from all studies were entered into a data-base, for the query of which we are developing a graphical user interface. The literature search resulted in 98 articles that have been published since the first use of tsDCS, including 38 on healthy-controls, 25 on patients, 20 on animals, 10 on modelling and 5 reviews. The data-base contains numerous parameters for all studies, such as sample size, study design, tsDCS specifications, study-domain, and outcome variables. While tsDCS studies aimed at modulating a wide variety of behaviours (e.g. from language over sports performance to pain perception), the most prominent topic is motor function and accordingly the most common outcome metric is changes in electromyographic metrics, with thoracic tsDCS being the favoured montage. Overall, our literature search identified nearly 100 tsDCS studies on various topics, which we sorted into different categories for further meta-analytic investigations. We aim for the database and associated graphical user interface to provide a convenient platform for researchers to query the existing tsDCS literature and thus aid in the planning of future studies.

The processing of complex environments is greatly facilitated by the formation and use of concepts. Concepts represent combinations of features shared by similar entities and allow generalization from limited experience to novel situations. Recent research suggests that map-like codes in the hippocampal-entorhinal system can support concept learning by representing the relations between experiences along relevant feature dimensions. Here, we investigate if this map-like representation ('cognitive map') of concepts supports the retrieval of abstracted information to guide inference. In a novel behavioral paradigm, participants first were trained to categorize a set of visual exemplars based on the ratio of their two continuous features. Subsequently, they encountered exemplars that exhibited only one of the features and were instructed to complete the missing feature in a continuous fashion according to the category label. We found that behavioral response patterns during feature inference were attracted more towards the category prototype location than to the nearest experienced exemplar, suggesting the retrieval of an abstract representation. In a functional magnetic resonance imaging (fMRI) experiment, we examine to what degree the hippocampal system, in concert with visual areas, represents prototypes for feature inference. This project can help us to understand the relationship between cognitive maps and abstract category representations.

Everyday decisions require us to predict values of different choice options in the future. Prediction of future values is facilitated by an internal model representing state transitions and associated reward contingencies in a task. Recently, it has been suggested that the hippocampal-entorhinal system represents relationships between states in a map-like format, enabling prediction of future states by facilitating the computation of distances and directions. It is unclear whether and how these spatial coding principles contribute to mapping abstract dimensions such as values. Here, we aim to test the idea that the hippocampal-entorhinal system supports prospective decision making by representing values of choice options in a relational map. To this end, we used a novel decision task requiring participants to predict changing values of two options over a sequence of time steps. Crucially and unbeknownst to participants, such a sequence formed a trajectory through an abstract two-dimensional value space characterized by the reward magnitudes of the two options. Recognizing the direction of the particular trajectory allowed for prediction of future values of the two options. Behavioral results show that participants are able to integrate changes along the two value dimensions to guide decision making, suggesting they form a map of the relationships between options. Preliminary fMRI results indicate prefrontal and hippocampal regions tracking the values of the options characterized by the trajectories in value space. Ultimately, this study can help us elucidate whether the hippocampal-entorhinal system adopts spatial coding principles to support efficient prospective value-based decision making.

Neuroimaging of the spinal cord using fMRI is challenging due to its small diameter, the influence of physiological noise, and strong magnetic field variations. One possible way to mitigate the field variations that arise from air-tissue interfaces at the neck is to employ passive shimming, where a susceptibility-matched material placed around the neck moves the field variations away from the region of interest. Here, we evaluate the effects of a commercially available device for passive shimming on 7T cervical spinal cord imaging. Four healthy volunteers were measured on a 7T MAGNETOM Terra scanner equipped with a custom-built 24-channel neck coil for cervical spinal cord imaging. Passive shimming was carried out by placing pads filled with liquid perfluorocarbon (SatPad Inc., USA) around the volunteers’ necks and image acquisition proceeded with/without the pads. The following image types were acquired in order to evaluate the effect of passive shimming: i) B0 map of the entire cervical spinal cord, ii) T1-weighted image of the entire cervical spinal cord, iii) T2*-weighted image of the cervical cord, and iv) fMRI time-series. We assessed i) the field homogeneity in the spinal cord, ii) the SNR of the T1-weighted and T2*-weighted structural data and iii) the tSNR of the EPI time-series data using tools from the Spinal Cord Toolbox. The use of passive shimming led to a modest improvement of field homogeneity within the cord (B0 map with tune-up shim only: mean SD [ppm]: 0.17 vs 0.22; difference of 31%; B0 map with active shim centered on spinal cord: mean SD [ppm]: 0.19 vs 0.20; difference of 7.3%). While this did not translate into any meaningful improvement for the T1-weighted acquisitions, there was a small improvement for T2*-weighted acquisitions, as demonstrated by enhanced gray matter SNR (mean SNR [a.u.]: 9.8 vs 9.2; difference of 6.4%). Visual inspection of the EPI acquisitions demonstrated an improvement in image quality, but this did not translate into any meaningful improvement in terms of tSNR. Building on previous observations at 3T, the use of susceptibility-matched material around the neck improved field homogeneity, but this only led to marginal gains in signal quality of anatomical and functional images. Future research avenues might include investigating alternative low-cost susceptibility-matched materials as well as applying passive shimming to the thoraco-lumbar spinal cord.

Statistical learning supports speech segmentation. Learners are known to learn word boundaries from the co-occurrence probabilities of syllables. For sentence-level comprehension beyond the word level, comprehenders also need to group individual words into multi-word chunks. It has been proposed that electrophysiological processing time windows constrain the duration of such chunks to ~2.7 seconds. We here investigate whether this temporal constraint affects the learning of co-occurrence probabilities that delineate multi-word chunks. We employ a statistical learning paradigm that manipulates the duration of to-be-learnt multi-word chunks. Participants are exposed to an isochronous sequence of bi-syllabic pseudowords that form hidden three-word chunks. We manipulate the duration of these chunks by varying the pause interval between the pseudowords, resulting in chunk durations of 1.95, 2.55 and 3.15 seconds. We test learning by an implicit target detection task and an explicit recognition task. We expect that chunks shorter than 2.7 seconds are learnt best. In the future, we will also use neural frequency tagging on electrophysiological data to provide an electrophysiological—possibly oscillatory—counterpart of the hypothesized behavioral constraint.

Lyrics in popular songs can reflect cultural and psychological changes. Recently, social psychologists and cultural evolutionary researchers began viewing music as a cultural product and used musical lyrics as a window to studying the psychological changes in cultural norms and values. By collecting longitudinal music chart data from 1956 across 50 countries, we make a cross-cultural comparison on song lyrics and investigate the emerging patterns and historical trends. We gather the lyrics of 183,963 songs in 16 different languages and analyse their lexical density (number of unique words to total words ratio), diversity (size of vocabulary in a given year), and lyrical topics (using topic modelling algorithms). Our preliminary results show a constant increase in lexical density in song lyrics around the world, which accelerated from the 2000s. However, the overall size of the vocabulary has gradually shrunk and lyrics have become more repetitive over the years. Using classical and state of the art natural language processing machine learning algorithms (multi-lingual BERT, Top2Vec, and LDA), we observe that universal topics emerge in all languages such as the use of nature and weather as metaphors, love, breakups, sadness, celebrations, financial success, and dance. However, the proportion of up-take of each topic in each language and their popularity trajectory over time appears to be culture-dependent. The observed findings provide valuable insights about the universality and variability of cultural expressions.

Recent models of the hippocampus and entorhinal cortex have suggested that they play a domain-general role in generalising structural information across different situations and sensory stimuli (Whittington et al., 2022). This approach extends the ‘cognitive map’ metaphor to a wider range of tasks, allowing flexible inferences across complex relational structures with no obvious bidimensional projection. Computationally, this can be enabled by a modular structure consisting of a generalisation module which maps onto a relational memory module (Whittington et al., 2020). Nonetheless, it remains an open question how ‘map’ representations lead to different inferences and actions in different situations. One intriguing suggestion is that top-down projections determine the structure of a map, such that it reflects the current task demands (Warren et al., 2022). Computationally, this might indicate a probabilistic vector coding which describes movement through a space. In neural terms, this might explain the discovery of ‘object vector’ cells in the medial entorhinal cortex, which fire when an agent is a certain distance and direction from a target (Høydal et al., 2019). Despite these exciting new proposals, it is unclear to what extent they may underpin inferences in more abstract domains. Two key predictions emerge: firstly, that flexible inference leads to factorised representations of stimulus and structure, and secondly, that transitions within an abstract structure are represented and manipulated by the entorhinal cortex and prefrontal cortex, respectively. To test this, we have designed a series of proposed behavioural and fMRI experiments in which participants learn the structure of a task, where ‘movement’ between states is determined by a series of rules. This space will then be expanded to test participants’ abilities to carry out path integration (or a variant thereof) in abstract state spaces. In the fMRI component, we will aim to test how the brain represents and implements a factorised coding scheme for movement in our abstract state space. In summary, we aim to test (1) whether humans generalise (abstract) structural information and use learned structures for new inferences, and (2) whether these inferences are underpinned by a task-sensitive factorised coding scheme in the medial temporal lobes.

Brain-age has increasingly gained popularity as a neuroimaging marker of health with the promise to be used as a clinical diagnostic tool. However, it is still unclear which specific brain features are of importance for its estimation. To address this issue, Hofmann et al. (2021) used Layer-wise Relevance Propagation (LRP; Lapuschkin et al., 2019) on brain-age predicting multi-level ensembles (MLENS) of 3D-convolutional neural networks. Resulting relevance maps highlighted regions across the whole brain, including white matter regions in which WM lesions (WML) typically appear in elderly subjects. We hypothesized that the MLENS capture WML and use it as an information source to predict higher brain-age. Methods For our analyses we used WML probabilistic maps of 654 participants from a population-based cohort study (LIFE-Adult; Loeffler et al., 2015). These maps were computed based on MRI data using the Lesion segmentation toolbox (Schmidt et al., 2012). Relevance maps were calculated for the FLAIR sub-ensemble of the MLENS. Relevance values were averaged over WML regions and compared to the expected relevance per voxel using a paired sample t-test. Additionally, for each subject we computed the ratio of positive relevance in WML voxels and overall positive relevance. Positive relevance reflects model evidence towards a higher age (above the sample mean). Results The difference between the average relevance in WML voxels and the expected relevance per voxel was statistically significant (Mdiff = 0.001, d = 0.90, t(653) = 22.95, p<.001). WMLs explained on average 1.69% of all positive relevance which was 29 times as much as the expected relevance contribution. Conclusion Our study demonstrates that deep learning models capture WMLs and associate higher brain age with a higher lesion load, underlining that AI-driven brain-age estimation models are capable of learning biologically relevant age-associated structural brain changes. However, we also found that brain-age estimates do not exclusively rely on WML. In future studies further aging-related brain features should be studied with respect to our relevance maps. This would allow us to test how much deep learning models rely on brain features known to be related to aging and how much they might use unstudied features.

In humans, language processing is supported by a fronto-temporal network connecting Broca’s and Wernicke’s area through dorsal and ventral pathways. Even if a homologous network exists in non human-primates, only humans possess the capability of language. To gain insights on the evolution of brain connectivity it is necessary to look at our closest ancestors, the chimpanzees. Most of the current knowledge on primate brain connectivity originates from tracer studies in macaques. Researchers now use diffusion MRI for comparative neuroscience to retrace the evolution of the brain’s structure looking for shared features between primate species. The last decade of research has provided insights on the anatomy of language relevant tracts like the arcuate fascicle (AF) and its homologue in the non-human primates. In chimpanzees, a connection between the inferior frontal gyrus and the superior temporal gyrus has been observed. In macaques this connection is more controversial. Here we study the auditory projections of the AF in a large sample of chimpanzees brains. Using auditory tonotopic informations and diffusion MRI we defined primary auditory fields as well as the auditory belt in chimpanzees. Then, these regions of interest have been used as a starting point for tractography. This resulted in the reconstruction of a dorsal and ventral pathway connecting the temporal gurus with the inferior frontal one, as shown in the literature. We are currently characterising the reconstructed tracts using quantitative measures such as fractional anisotropy and mean diffusivity. Overall this study confirms previous findings in the literature and brings a stronger understanding of the anatomy of the language network homologues in non-human primates. The next step is to see whether or not a connection between the inferior frontal gyrus and the middle and inferior temporal guys exist in chimpanzee and how it compares to the connection found in humans.

Introduction: Different heartbeat-related effects such as the cardiac cycle and heartbeat evoked potentials (HEP) have been shown to shape the processing of sensory information. It remains to be elucidated, however, if these heartbeat-related effects also assert an impact on more complex cognitive functions, such as sustained attention and working memory. Since HEP amplitudes are significantly higher during interoceptive than exteroceptive attention, it is hypothesized that changes in HEP amplitude represent an attentional shift from external task-relevant to internal interoceptive stimuli. Here, we want to test whether task performance on an auditory Novelty-Oddball-Task will be related to HEP amplitudes in the pre-stimulus time window. Methods: We aim to analyze a large dataset of EEG recordings acquired in the LIFE-adult-study (N= N=2308, 60 - 82 years). The subjects performed a 15min Auditory Oddball task. On the single-trial level, event-related responses will be calculated and sorted according to high and low P300 amplitude as well as reaction time (RT). We will identify HEPs in the pre-stimulus time window and calculate their amplitude with respect to high/low P300 amplitude and RT. Results: To follow (ongoing study) Discussion: We hypothesize that the HEP amplitude is modulated along with changing attentional focus from internal interoceptive to external task-related processing. Within this framework, larger HEP amplitudes index a stronger focus on internal processes. Hence, we expect that large HEP amplitudes are followed by small P300 amplitudes and slow reaction times. These results would provide further insights into the interplay between interoceptive and exteroceptive stimulus processing and their interference.

In order to navigate goal-directed behavior in an ever-changing complex environment, a fine balance between maintaining and updating of goal-relevant information is required. A cognitive system crucially involved in these processes is our working memory (WM). While maintenance of WM representations is thought to be implemented in the PFC, updating is accomplished via fronto-striatal go/no-go pathway modulations. Importantly, neurocomputational models, as well as fMRI and PET studies have demonstrated the involvement of dopaminergic signaling in the modulation of these two complementary processes. In obese individuals, the accurate adaptation of WM representations seems to be compromised. For instance, obese participants have been shown to prefer short-term rewards despite of negative long-term outcome or to repeat previously rewarded action, despite of current devaluation. These findings hence suggest that obese individuals might fail to update the negative consequences of their actions, which could contribute to dysfunctional preservation of maladaptive (eating-) behaviors. Additionally, functional changes in the above-mentioned WM-related brain areas have been observed in obesity. Several studies report altered central dopamine levels and dopamine receptor density in striatum, as well as changes in grey matter density in PFC in obese individuals. In the present study, we therefore aim to elucidate on potential impairments in these specific sub-processes of WM functioning in obesity. To this end, we implement a task-based fMRI study with a grouping design (OB vs. lean). We use a modified version of a delayed match-to-sample task originally designed by Fallon and Cools (2014), which compiles specific conditions that probe updating and maintenance of mental representations. Here, I will present preliminary findings of this study.

Human interaction relies on our ability to infer and interpret the mental states (e.g., intentions, desires, or beliefs) of others. This ability, referred to as Theory of Mind, is central to human executive functioning, as it is crucial for our adequate adaptation to our social surroundings, and therefore survival. So far, research on the neural correlates of Theory of Mind has focused extensively on the cerebral cortex, whereas the role of the cerebellum has been largely overlooked, despite clinical evidence that suggests a widespread involvement of the cerebellum in Theory of Mind processing. In light of this, the present project aims to comprehensively examine the contribution of the cerebellum to Theory of Mind from childhood to adulthood. The core aim is to identify regions of the cerebellum involved in Theory of Mind, and how these regions support Theory of Mind processing in the rest of the brain through anatomical and effective connections with the cerebral cortex. For this purpose, this project will leverage large-scale multimodal neuroimaging data across two studies. Study 1 will identify white matter pathways and effective connections between the cerebellum and cerebral regions involved in Theory of Mind in a large sample of adults. In Study 2, a similar approach will be used in a sample of children aged between 3 and 12 years, in order to investigate the contribution of early-life maturation of the cerebellum to the emergence of Theory of Mind. This project will contribute towards a novel, whole-brain view of Theory of Mind, and also identify cerebellar regions that are crucial for its development, thus paving the way for further understanding and mediation of Theory of Mind impairments.

Oscillatory brain activity is thought to support language processing by both tracking and predicting speech. Neural oscillations can track acoustic and abstract linguistic units, such as syllables and phrases. However, it is less understood how they help to predict upcoming acoustic and abstract information. In the first study, we investigate how the brain tracks acoustic and abstract cues, such as prosody and transitional probabilities (TPs). In the second study, we focus on how this tracking engages when and what predictions of upcoming linguistic information. We hope to unravel (i) which neural circuits are involved in the tracking of prosody and TPs, and (ii) how neural oscillations contribute to predicting when and what in language processing. Our experiments combine a statistical learning paradigm with magnetoencephalography (MEG). In the first experiment, we use a 2-by-2 frequency-tagging paradigm and expose participants to continuous streams of syllables, in which prosody (flat/rhythmic) and/or TPs (uniform/rhythmic) could delineate chunks of three syllables. When participants are exposed to rhythmic streams, we expect to find neural tracking at the chunk rate in distinct MEG sources depending on the nature of the cues (i.e., prosody/TPs). In addition, we expect tracking of TPs to predict chunk learning both neurally and behaviorally during a subsequent test phase. In the second experiment, we expose participants to structured or random streams of syllables. In the structured condition—but not in the random condition—chunks of three syllables can be learned as words based on TPs. In a subsequent testing phase, we first present a series of syllables to entrain neural oscillations. Then, we assess whether entrainment associates with when and what predictions of a target that is presented downstream. Here, we expect facilitated processing of a target that is aligned to a predicted time point (based on entrainment) and with the knowledge of the lexicon (based on TPs). These experiments have the potential to show that (i) rhythmic processing of TPs and prosodic cues is neurally dissociable and jointly impacts chunk learning, and that (ii) neural oscillations provide a mechanism for predicting when and what to optimize speech processing.

Finding locations in a familiar environment requires accurate representations of space (‘cognitive maps’). In many species, the neural basis of cognitive maps are a set of specialized neurons in the hippocampal formation. Previous research has found that stretching or squashing environmental dimensions alters the firing pattern of hippocampal place cells and entorhinal grid cells in freely moving rodents as well as human spatial memory. This Indicates that boundaries defining the geometry of space play an important role in determining the nature of cognitive maps. Here, we examine how behavioral changes to environmental deformations relate to those on a neural level as measured with functional magnetic resonance imaging (fMRI) in humans, and how these effects can be explained by models of cellular firing. In an ongoing two-day study, we first train participants to learn the location of objects inside a virtual arena presented in desktop-based virtual reality. During subsequent scanning, we ask them to re-visit each location inside the arena as well as actively imagine them. Critically, the arena deforms from square to rectangular shape across days. We hypothesize that during imagination, location-related similarity patterns in the hippocampus correlate with those observed in spatial memory and differ across environments as predicted by the boundary vector cell model of place cell firing. Furthermore, during virtual navigation, we predict geometry-dependent changes of the direction-modulated fMRI signal in the entorhinal cortex. Our study can contribute to a better understanding of how coding principles in the hippocampal formation flexibly adapt to deformations of the environment.

In social contexts, humans frequently infer what other people think and believe. This ability to reason about other minds is called Theory of Mind (ToM). In pre-school aged children, a recent study identified distinct and independent cortical brain regions associated with the early development of a non-verbal form of this ability (implicit ToM) in contrast to the classic, verbal form (explicit ToM) (Grosse Wiesmann et al., 2020). Based on this study, we asked for the white matter networks associated with implicit and explicit ToM. We hypothesize a dual pathway model of ToM: a dorsal connection including the medial prefrontal cortex, the precuneus and the tempoparietal junction supporting explicit ToM and a ventral connection including the anterior insula and supramarginal gyrus supporting implicit ToM development. To test this hypotheses, we analysed an existing dataset of behavioral and sMRI/dMRI data collected from a sample of 3-to-4-year-old children (Grosse Wiesmann et al., 2017; Grosse Wiesmann et al., 2020). Data analysis is ongoing, and we will present preliminary results.

The human neocortex consists of long- and short-range association fibres connecting distinct specialised but synchronised networks. All these connections, their microstructural properties and the thereby determined transmission strength and efficiency can be understood as the microstructural connectome. The most essential microstructural characteristics are axon diameter, myelination, g-ratio and iron content. The majority of the present human structural connectomes were estimated by Diffusion-Weighted Magnetic Resonance Imaging (DWI) given its unique potential to measure structural brain connectivity in the living human brain. DWI-based connectomes are well suited for mapping large range white matter (WM) fibres. 90% of all cortico-cortical WM connections however are composed of short association fibres which are a vital factor in brain development, plasticity and ageing. Short association fibres connecting adjacent gyri and running right below the cortex are short in length and thus, have short conduction times. It has been suggested that these so-called U-fibres show a late and weak myelination as well as a small axonal diameter, but quantitative data reliably showing these aspects are still missing to date. Taken together, current estimates of the connectome only comprise about 10% of all WM fibres and are not only constrained by incompleteness but also method-related biases. Quantitative magnetic resonance imaging (qMRI) and advanced DWI are optimally suited to study microstructural properties of the WM. To obtain a comprehensive microstructural human connectome, we will acquire a large dataset on 100 participants with ages distributed across adult lifespan and containing ultra-high resolution qMRI at 7T, ultrahigh resolution DWI for U-fibre mapping and ultra-high b-value DWI acquisition for effective axonal diameter and g-ratio imaging. These large cohort comprehensive microstructural connectome data, will contribute to a better understanding of brain maturation, plasticity and ageing and, for the first time, integrate both short and long-range connections, together with microstructural information from cortex and WM tracts across the adult life span.

Ample work offers insights into the neural correlates of parenthood. Yet few studies consider brain activation during real-time parent-offspring interaction. In this fMRI study, we therefore sampled brain activation of preschool children’s parents during a novel event-related adaptation of the virtual ball-tossing Cyberball paradigm (Williams, Cheung & Choi, 2000), for the first time including both fathers (n=48) and mothers (n=40). Parents supposedly participated with their own and an unknown child. The event-related design allowed us to compare three main conditions (inclusion, exclusion, and [new] re-inclusion) and up to three additional task sub-conditions (catch, observe, throw). Following the fMRI scan (3T Siemens Skyra, 32-channel head coil, T2*-weighted GE-EPI sequence with multiband acceleration factor 3), parents completed several self-reports, including an adapted version of the Need Threat Questionnaire (NTQ; van Beest & Williams, 2006). In accordance with recent meta-analytic evidence, we found that social exclusion in parents primarily activated brain areas involved in emotion processing and regulation as well as mentalizing (e.g., posterior cingulate cortex, posterior and anterior insula, ventrolateral prefrontal cortex), rather than dorsal anterior cingulate cortex originally linked to “social pain” (Vijayakumar et al., 2017; Mwilambwe-Tshilobo & Spreng, 2021). Furthermore, our analyses revealed a selective decrease in reward-related activation (e.g., putamen) for catch (versus observe) trials during re-inclusion (versus inclusion). We are currently extending these analyses by examining the influence of child familiarity (own versus unknown child), parental and child biological sex, as well as associations with self-reports (including NTQ scores, perceived paradigm genuineness, and parental caregiving beliefs).

Background: Biophysical modelling of the induced electric field (E-field) helps to identify the neural structures that are effectively stimulated by Transcranial magnetic stimulation (TMS). To date, modelling frameworks have been exclusively used to map TMS effects in the primary motor cortex. The current project aims at transferring this knowledge to cognitive functions in the healthy human brain, and ultimately develop more effective stimulation approaches. Methods: We will combine functional magnetic resonance imaging (fMRI), TMS, and Electroencephalography (EEG) to study attentional processes, i.e., attention orientation (associated with the dorsal attention network, DAN) and reorientation (regulated by the ventral attention network, VAN). For modeling of the individual E-field, T1- and T2-weighted images and diffusion MR images will be collected in all subjects. We aim to (1) elucidate the functional interaction of the DAN and VAN during attentional orientation and reorientation by means of effective connectivity analyses of fMRI data. (2) identify optimal TMS targets for attentional processes in a TMS localization study and verifying these targets in a validation study. (3) map the immediate consequences of a TMS-induced perturbation through dynamic causal modeling (DCM) in an online TMS-EEG study. (4) unravel plastic after-effects of continuous theta burst stimulation (cTBS) on task/resting-related activity and connectivity patterns through offline TMS-EEG. Expected Results and Impact: We expect that the results of this project will substantially advance the current knowledge about the neurophysiological effects of TMS on cognitive functions.

The gut microbiota is a key modulator of gut-brain-behavior signaling in obesity. This means variations in microbiota composition; functional genes and their metabolites may directly or indirectly affect the brain via vagal stimulation or immune-neuroendocrine pathways. Contrariwise, brain signals through the autonomic nervous system (ANS) and the hypothalamic-pituitary-adrenal (HPA) axis affect gastrointestinal activities such as gut microbial abundances, patterns of gene expression, and intestinal permeability. In this study, we aim to test the hypothesis that a microbiome-changing intervention improves food decision-making and to determine the underlying microbiota and metabolic mechanisms. To this end, 90 overweight/obese adults will be enrolled in a randomized controlled trial to test the effects of a pre-biotic dietary intervention (supplementary intake of soluble fiber) or a behavioral lifestyle intervention (weekly educational program) vs. control condition (supplementary intake of isocaloric starch) over a period of 26 weeks. Before and after the intervention/control period, participants will undergo task-based functional and structural MRI and cognitive testing. The gut microbiota will be assessed using 16S rDNA next-generation sequencing (V3/V4 region) in stool samples. Diet, anthropometrics, and lifestyle-related factors will be monitored by questionnaires and metabolomics will be assayed in peripheral blood and stool (e.g., SCFA). In brief, using a modulation of gut-brain communication through a prebiotic diet and lifestyle intervention, respectively, will enable us to discover microbiota communities that play a key role for eating behavior. These mechanistic insights could help to develop novel preventive and therapeutic options to combat unhealthy weight gain in our obesogenic society.

Neurofeedback (NF) aims to normalize reduced food-specific prefrontal cortex (PFC) signals in binge-eating disorder (BED) and overweight by providing individuals real-time feedback on brain signals and asking for active modulation. Building on improved food-specific self-regulation in BED following electroencephalography-NF, the first randomized-controlled trial currently examines the efficacy of food-specific functional near-infrared spectroscopy (fNIRS)-NF for BED (NIRSBED, DRKS00014752). Based on this trial, this preregistered study examined neural and psychological mechanisms of fNIRS-NF over the PFC, including real-time, offline, and subjective regulation success in each fNIRS-NF-session and subjective loss of control over eating before and after each fNIRS-NF-session. We assumed a positive association between real-time and subjective fNIRS-NF regulation success in each fNIRS-NF-session and decreased subjective loss of control over eating after vs. before fNIRS-NF-sessions, which should correlate with higher subjective regulation success. During each of 12 60-min fNIRS-NF-sessions, n=22 adults with BED were asked to minimize individually appetitive food pictures varying with the PFC signal. Real-time regulation success was measured via the picture size, offline regulation success via the PFC signal during active regulation versus passive viewing of foods. Subjective loss of control over eating before and after fNIRS-NF-sessions and subjective success after fNIRS-NF-sessions were assessed via questionnaire. Results showed that real-time and subjective regulation success in each fNIRS-NF-session correlated positively. Subjective loss of control over eating decreased after vs. before fNIRS-NF-sessions, but individual decreases in loss of control over eating after vs. before fNIRS-NF-sessions were not associated with subjective or offline regulation success in fNIRS-NF-sessions. Overall, the association between subjective and real-time regulation success confirmed the assumed fNIRS-NF-mechanism of operant conditioning. The potential of food-specific fNIRS-NF for BED was suggested by the modulation of loss of control over eating as the key diagnostic feature of BED during fNIRS-NF-sessions. However, this modulation was not consciously perceived by participants nor reflected in changes in underlying PFC networks, indicating an implicit mechanism underlying fNIRS-NF. These results may guide the optimization of future NF studies in larger BED samples.

The capability of biological neural networks to codify information in spike patterns is strongly influenced by the regulation of the network energy resources. This capability has inspired different techniques in Computer Science to mimic biological networks in order to solve specific tasks. One of these techniques is Spiking Neural Networks (SNN). The basic spiking neuron models the temporal dynamic of membrane potential of a neuron to generate action potentials or spikes. SNN can be applied in the field of Neuromorphic Olfaction to emulate the behavior of the olfactory system in order to build technological devices to recognize odors, for example, electronic noses. E-noses are electrochemical sensors that can transduce odor molecules into electrical signals. These sensors suffer from a phenomenon called drifting (small non-deterministic temporal variations of the sensor response, when it is exposed to the same chemical under similar conditions). Considering plasticity can help to approach the drifting phenomenon. Therefore, synaptic dynamics consider the dynamic of release and capture of neurotransmitters between a presynaptic neuron, the synaptic cleft and the postsynaptic neuron. Integrating synaptic dynamics in some architectures of SNN can be suitable to approach the drifting phenomenon of e-noses. Additionally, metrics of energy and information are proposed to study the behavior of the proposed architectures. For energy metrics, we propose the study of dynamical system properties of SNN and relate these metrics with the Free-energy principle. For information metrics, we propose the application of information measurements based on coding strategies, entropy and similarity. Additionally, we propose the study of neural manifolds and transformation from high-dimensional outputs of models (network space) into possible low-dimensional spaces (neural space). The aim of this project is to model components of the Olfactory System through SNN with synaptic dynamics using metrics in terms of energy and information, to try to solve the challenge of drifting.

Introduction CPSP is a frequent consequence of stroke developing within 3 months after a stroke affecting the somatosensory system. Among stroke patients the prevalence is 8% and it is often refractory to medical treatment impairing patients’ quality of life dramatically. Aim Investigate potential clinical and neuroimaging features of CPSP patients before the pain occurred and with the development of pain. General Setting Prospective clinical study in patients with acute somatosensory stroke. After 9 months, patients are grouped into those who developed CPSP and those who did not (NPSS: non pain stroke patients). Methods 61 somatosensory stroke patients were recruited at Charité. After neurological examination, resting-state fMRI and structural MRI (MPRAGE) were performed at several time points until 9 months. We performed functional connectivity analysis and connectivity gradients analysis using CONN, Brainspace and Surfstat. For seed-based connectivity analysis, seeds were chosen from areas of the pain network and from areas with structural difference between CPSP and NPSS patients in our study. Hypotheses 1. CPSP patients differ from NPSS patients clinically already before the pain occurs. 2. Resting state connectivity of specific brain areas (“seeds”) of patients with CPSP differs from NPSS patients already before the pain. We postulate this to occur in areas which are part of the pain network. 3. In a whole brain exploratory analysis, connectivity gradients differ between CPSP and NPSS patients. Results 1. Clinically, CPSP patients showed more severe neurological impairment (P=0.003) and poorer prognosis (P< 0.001) than NPSS patients. CPSP patients showed poorer abilities of daily life(P< 0.001) and poorer quality of life (P=0.011) than NPSS patients. 2. CPSP compared to NPSS patients showed stronger functional connectivity between the ipsilesional angular gyrus and bilateral sensorimotor superior areas as well as in the contralesional postcentral gyrus. 3. For CPSP patients on gradient 1, one parcel with the ipsilesional angular gyrus was closer to the somatosensory and pain (Puncorrected=0.098). Conclusion Our study demonstrates for the first time that functional connectivity of several brain areas differs between CPSP patients and NPSS patients. Especially ipsilesional angular gyrus might be an indicator for the subsequent development of pain. These findings might be the basis for prediction of pain in acute stroke patients.

Unilateral spatial neglect is a disabling neurocognitive disorder characterized by a lack of attention to stimuli in the contralesional, often left hemispace (Heilman et al., 2000). Conventional pencil-and-paper tests, such as cancellation or line bisection tasks, unreliably and insufficiently assess discrete symptoms of neglect, which can impair activities of daily living. To sensitively detect and quantify subtle neglect symptoms, we developed iVRoad: In this intersection scenario of two parallel, two-lane, heavily trafficked roads, participants are to cross the virtual street, drop a letter into a mailbox on the left or right side and return. We tested iVRoad using the HTC Vive Pro Eye in 60 subjects (right hemisphere stroke with left neglect (N = 20), without neglect (N = 20) and matched healthy controls (N = 20)). Performance (e.g., reaction time, errors) and movement parameters (e.g., head rotation, eye movement) were analyzed to identify group-specific behavioral patterns. Given the lateral orientation in neglect, we compared performance between groups with respect to alternating traffic directions (i.e., cars coming from the left, right, or both sides). Overall, the task was well-tolerated by all participants. The following parameters were most suitable to distinguish neglect from no-neglect and control subjects: reaction time (F(2, 50) = 16.04, p < .001, ηp2 = 0.25), left-sided errors (F(2, 50) = 12.34, p < .001, ηp2 = 0.33), and lateral head movements for cars approaching from the left side (F(2, 50) = 12.73, p < .001, ηp2 = 0.34). IVRoad is an immersive, naturalistic task, which can measure clinically relevant behavioral variance and detect discrete neglect symptoms.

There is growing interest in the investigation of somatosensory processing in the human spinal cord via non-invasively recorded somatosensory evoked potentials (SEPs). However, these efforts are hampered by the impact of physiological noise on electrospinographic (ESG) recordings. Particularly damaging is the impact of the cardiac artefact, the effect of which can be as large as ~1mV, while the SEPs of interest are generally smaller than ~1µV. Current approaches for studying SEPs using ESG thus involve either large-scale averaging (~2000 trials) or extensive high pass filtering (at ~30Hz). However, both approaches have significant drawbacks: firstly, in many cognitive neuroscience paradigms, it is not possible to record 2000 trials per condition, and secondly, several electrophysiological signals of interest have frequency content that would be removed by high pass filtering. Thus, there is an obvious need to develop alternative approaches to remove the cardiac artefact from ESG recordings.
We have previously developed a principal component analysis (PCA) based approach for cardiac denoising of ESG data (adapted from simultaneous EEG-fMRI), but it is unclear how this algorithm compares against other approaches. Therefore, we systematically evaluated the performance of this algorithm on ESG data against alternative approaches including independent component analysis (ICA) and signal space projection (SSP). The algorithms were evaluated based on both their ability to reduce the presence of the cardiac artefact and to improve the signal-to-noise ratio (SNR=[peak amplitude/std in baseline period]) of the SEPs. We first evaluated each algorithm on a dataset known to include SEPs with a relatively high SNR (due to mixed-nerve stimulation) and found conclusive evidence that SSP with a low number of projectors (n<6) is the best approach. The application of SSP resulted in reduced impact of the cardiac artefact compared to PCA and improvements of up to 100% in the SNR of the SEPs. While ICA effectively reduced the cardiac artefact, it failed to produce SEPs of high SNR. To validate these results, the algorithms will be tested on a new dataset including SEPs with a more limited SNR (due to sensory-nerve stimulation only). Finally, we aim to determine whether successful denoising allows for the analysis of spinal cord SEPs on a single trial level, which is of increasing interest in cognitive neuroscience.

To support language comprehension, contextual information can be used to predict upcoming input (Kuperberg and Jaeger, 2016). If bottom-up input matches these top-down predictions, processing is facilitated. If input instead violates predictions, additional processing demands are needed to fluidly adapt to the unexpected input. Here, two event-related potential (ERP) studies probed whether the brain engages distinct neural mechanisms in response to output that fulfills versus violates strong predictions. Previously, distinct ERPs have been linked to different contextual manipulations. The N400 response reflects the benefits of confirmed predictions and is smaller for semantically expected vs. unexpected or anomalous words. In contrast, two late positive-going ERPs (600-1000 ms) index prediction violations. During reading, the late posterior positivity is evoked by unexpected but semantically implausible words, whereas the late frontal positivity is elicited by unexpected but semantically plausible words (Kuperberg et al., 2020). While the N400 is similar for both visual and auditory words, little is known about the modality-independence of the late positivities. Given the need for replication of phenomena that are new or for which evidence is scarce (see Nieuwland, 2017), the present experiments aim to conceptually replicate and extend these earlier findings. To this end, participants read (Exp. 1) or listened (Exp. 2) to German three-sentence scenarios in which the third sentence constrained for a broad event structure (e.g., Agent cautioned animate-Patient). High constraint contexts additionally constrained for a specific word (e.g., a two-sentence context about a beach, lifeguards, and sharks constrained for “swimmers”). Low constraint context did not constrain for any specific word. We measured ERPs on critical nouns that fulfilled and/or violated each of these constraints (i.e., expected, unexpected, anomalous). Preliminary findings indicate clear, dissociable effects to fulfilled semantic predictions (a reduced N400), to lexical prediction violations (an increased late frontal positivity), and to animacy violations (an increased late posterior positivity) both during reading and listening. These findings show that confirmed and violated predictions manifest as distinct spatiotemporal neural signatures independently from language and modality.

Psychosocial stress is omnipresent in modern western societies and associated with detrimental health effects. In addition to our first-hand stress experiences, we also resonate with the stress of the people around us. In fact, one can be stressed solely by observing the stress of another. Stress resonates both on the emotional level and at the level of sympathetic and hypothalamic-pituitary-adrenal activation. At the current time-point, our understanding of the underlying mechanisms of such stress resonance is limited, however. What we do know is that higher levels of self-reported empathy are linked to higher endocrine stress resonance. Given recent evidence suggesting an association between attachment style and brain-to-brain synchrony, another potential influencing factor of physiological stress resonance could be attachment style. Therefore, the current study aims to provide insight into the associations between psychosocial stress, attachment style and empathic processing on the bio-behavioral level. Subjecting romantic partners to a standardized empathic stress test (the Empathic Trier Social Stress Test), we investigate stress resonance on the behavioral, cardiovascular, endocrine and brain-to-brain level, and combine these measures with assessments of attachment and behavioral empathy.

The hippocampal-entorhinal system is remarkably efficient in organizing relations between sensory stimuli, such as state transition probabilities, in a cognitive map. Such knowledge representation is assumed to enable fast learning of novel relations and the generalization of reward, likely facilitating goal-directed behavior. However, in addition to experienced transition probabilities, objects may simultaneously share other types of relational information, like reward contingencies. This fMRI study investigates how the representation of relational knowledge is influenced by a subsequently learned latent reward structure, whether the structural relations between objects still are represented veridically. Participants first acquire knowledge about object relations based on object transitions that follow a hidden graph structure. In a subsequent decision-making task, each object gets associated with fluctuating reward values. Critically, two parts of the graph structure share the same reward contingencies. Behavioral data suggest that participants successfully acquire structural knowledge and can utilize it to find shortcuts when navigating the graph. Model-based analysis reveals that participants can extract the additional underlying latent reward structure and generalize over states sharing the same reward contingencies. Furthermore, they can apply their acquired structural knowledge to correctly infer current reward values of objects whose values they never directly experienced. Suggesting participants represent both the experienced transitions and the shared reward contingencies between objects; moreover, they can combine both types of information for generalization and inference. Ongoing fMRI analyses will examine how object relations and shared reward contingencies are represented in the hippocampal-entorhinal system and interact with other brain regions to guide choice behavior.

Cortical functional organization shows inter-hemispheric differences, linked to higher-order cognitive functions such as language and attention. This functional brain asymmetry has been related to its underlying structure, such as white matter connectivity of the corpus callosum, cortical thickness, and surface area. Here, we studied asymmetry of intra-cortical microstructural variation across the entire cortical mantle and compared histological asymmetry axes with asymmetry of functional organization. We assessed asymmetry of cell-body staining intensity profiles in an ultra-high–resolution 3D histological reconstruction of an ex vivo brain (male donor, age = 65 years). Finally, we assessed whether histologically defined asymmetry of microstructural profiles showed spatial correspondence to asymmetry in resting-state functional connectome organization (UKB, n = 34,604, age = 56 ± 8 years). Microstructure in the human brain showed posterior-anterior variation in intensity profiles with particular primary visual (t = 23.09, P FDR < 0.001), somatomotor (t = 4.05, P FDR < 0.001), language (t = 14.84, P FDR < 0.001), posterior-multimodal (t = 15.11, P FDR < 0.001), ventral-multimodal (t = 3.15, P FDR = 0.003), and orbito-affective (t = 3.17, P FDR = 0.003) networks showing higher microstructural intensity values in the left hemisphere. Conversely, secondary visual (t = -8.79, P FDR < 0.001), cingulo-opercular (t = -3.41, P FDR = 0.002), dorsal attention (t = -7.28, P FDR < 0.001), frontoparietal (t = -5.20, P FDR < 0.001), and auditory (t = -15.19, P FDR < 0.001) networks had particularly strong intensity profiles in the right hemisphere. Correlations between functional organization asymmetry and laminar intensity asymmetry were mainly observed in layer III (intra-hemisphere: r = 0.207, P_spin = 0.025; inter-hemisphere: r = 0.187, P_spin = 0.053) and IV (intra-hemisphere: r = 0.252, P_spin = 0.010; inter-hemisphere: r = 0.224, P_spin = 0.024). Here we observed marked asymmetry of ex-vivo microstructure in language, posterior-multimodal, visual, and dorsal attention networks. These patterns of asymmetry were related to asymmetry of functional organization observed in UKB. Our findings provide novel evidence on the microstructural basis of inter-hemispheric differences in the human cortex, suggesting a key role of microstructural variation along cortical column depth.

The ability to process syntax is a key component for a child to acquire during language development. In the adult brain, two white matter fiber tracts are crucial for language processing, namely the arcuate fascicle (AF) and inferior fronto-occipital fasciculus (IFOF). Distinct functional roles have been proposed for these fiber tracts, with the left AF being crucial for complex syntax. During the preschool period, children acquire major syntactic concepts, and their productivity of complex sentence structures increases. In this ongoing study, we investigate the maturation of the two major language fiber tracts in relation to children’s syntactic comprehension and production abilities during this critical developmental period. To this end, we analyze magnetic resonance imaging (MRI) and behavioral data in a cross-sectional sample of 90 3- to 6-year-old children. To obtain an estimate of children’s syntactic comprehension ability, we use data from a standardized sentence comprehension test. Further, we recode children’s available sentence production data to assess their syntactic proficiency on the production level. Individual language fiber tracts are reconstructed from children’s diffusion-weighted MRI data, and indices of white matter maturation are retrieved. We aim to disentangle the AF that terminates in the inferior frontal gyrus (IFG) from a second dorsal tract that terminates in the premotor cortex. Thus, we segment the following tracts: the IFOF, the dorsal tract targeting the premotor cortex, the AF targeting the IFG and the corticospinal tract as a control. We relate maturation indices with children’s syntax scores at each node of the fiber tracts in separate general linear models. With this approach, we will be able to identify locations along the fiber tracts that show a relation between structural maturation and children’s syntactic performance. Results from this study will enhance our understanding of the neural basis underlying preschooler’s syntactic abilities.

Advances in Deep Learning models have surpassed our ability to understand the inner “thinking process” that led to their results, leaving us with the same interrogations we ask to understand their biological counterpart, the brain. Here, the geometry of collective neural representations seems to be a crucial characteristic of information processing in both artificial and natural neural networks. However, the dynamical aspect of these neural maps and the cognitive processes that operate on them are still largely unexplored. With this PhD project, we aim to understand the structural and functional patterns underlying the “thinking process” in Artificial Neural Networks (ANNs), taking inspiration from neuroscience. We want to study ANNs as we would study any model organism solving a sequential task with the experimental and analytical tools of cognitive neuroscience. We will analyze the structure and the dynamics of ANN’s neural manifolds when learning to play Atari games. As a next step, we will study what characteristics of neural manifold learning can support processing and integration of different types of information in ANNs. We will compare these characteristics with the findings of a parallel study in humans where we present both the AI models and the human subjects with spatial and conceptual tasks. By developing our neuroscientific approach to study the representational geometry of AI models we hope to gain insights into the general mechanisms of neural computation in both artificial and biological systems. We expect that we can translate the conceptual findings and computational tools from this project into practical applications for explainable AI, multimodal learning or neuroimaging data analysis.

Infants' ability to track objects develops quickly within the first six months of life. From the second year of life on, they also seem to track what other people can see, referred to as visual perspective taking. Notably, both infants and adults do not only take into account the perspective of others when actively reasoning about them. The perspective of others also seems to shape our own representation of the environment when it is irrelevant to what we are currently doing, referred to as altercentric cognition. Here we ask how others’ visual perspective modulates adults’ and infants’ neural object processing. To test this, we make use of rhythmically entrained brain oscillations: Viewing an object that flickers at a specific frequency results in brain oscillations at exactly the same rhythm; these oscillations thus provide a specific neural signature of our object representation. Participants were presented with an agent observing an object flickering at 4 Hz. The object either disappeared into a tunnel (blocking the participant’s as well as the agent’s view) or behind an occluder (blocking only the participant’s but not the agent’s view). We hypothesized that adults and infants (aged 12-14 months) also show entrained oscillations in reaction to someone else seeing the object, even when they no longer see it themselves. Indeed, adults (N=40) showed a higher response amplitude when the agent continued to see the object (occluder condition) compared to when she could no longer see it (tunnel condition). This was the case while the object disappeared as well as after the object had been fully occluded. Infants (N=56) also showed higher entrained 4 Hz signal in the occluder compared to the tunnel condition but only after the object had been fully occluded. These findings indicate that infants’ and adults’ neural object processing and memory is altercentrically modulated by the perspective of other agents.

Stroke often severely affects language function. A better understanding of post-stroke language recovery is crucial to identify reorganisation mechanisms. Previous neuroimaging work demonstrated distinct mechanisms in different phases of language reorganisation after stroke: global network disturbance in the acute phase, upregulation of the bilateral domain-general multiple-demand network in the subacute phase, and reintegration of left temporal language areas in the chronic phase. These previous studies further suggest that phase-specific mechanisms depend on individual lesion sites. However, changes in the interactions within and between the language and multiple-demand network during language recovery have not been explored yet. To fill this gap, the present study examined changes in the effective connectivity in the language and the multiple-demand network after temporo-parietal and frontal stroke in the left hemisphere. In a sample of 34 patients, we investigated the directed functional connections and their modulation by speech and reversed speech in the acute, subacute and chronic phases using Dynamic Causal Modelling of fMRI data. Preliminary results suggest that, in general, multiple-demand regions exerted a facilitatory influence onto language areas. This facilitatory influence was further increased by speech. We also found phase-specific differences in the effective connectivity. In the acute phase, the left posterior temporal cortex (PTC) exerted an inhibitory influence on both left inferior frontal gyrus (IFG) and multiple-demand areas, however, speech turned the inhibitory influence excitatory. In the subacute and chronic phases, left IFG showed a facilitatory influence on left PTC, but speech changed the facilitatory influence into inhibition. Speech did not significantly modulate the interaction between the left and right IFG in the acute phase, but turned the non-significant intrinsic connection from the left to right IFG to inhibition in the subacute phase, and turned the inhibition from the left to right IFG to excitation in the chronic phase. These preliminary results support the notion that domain-general areas are crucial for language recovery after stroke. Our findings further suggest changes in the interaction between language and domain-general areas across the time course of recovery. In the next step, we will investigate cross-sectional changes separately for patients with lesions in the left frontal and temporo-parietal cortex.

Learning through repeated exposure plays a crucial role in how human listeners acquire memories of sounds in everyday life. Numerous studies have underlined a remarkable capacity for auditory perceptual learning even for random and meaningless acoustic patterns. What remains less explored is whether and how perceptual learning is modulated by different aspects of the learning context. The current EEG study sets out to explore the influence of two factors, as well as their interaction, on perceptual learning of random acoustic patterns: temporal regularity and listeners’ attention. Using an established implicit learning paradigm, we present participants with 3.5-s acoustic sequences, half of which contain repetitions of a 200-ms pattern. Unbeknownst to the listeners, in each experimental block one 200-ms “reference” pattern is repeated across trials, while all other repeated patterns occur only in one trial. In accordance with earlier studies, learning is measured indirectly as the contrast between the repeatedly presented reference patterns, assumed to be learnt through repeated exposure, and the patterns that occur only once. Temporal regularity of pattern recurrence is varied between blocks, such that in half of the blocks the pattern onsets within a sequence are regular, while in the other half of the blocks they are jittered. Listeners’ attention is manipulated between sessions, such that in a first session attention is directed away from the auditory stimulation by means of a demanding visual distractor task, and in a second session attention is directed towards the acoustic pattern repetitions through a repetition detection task. Consistent with previous findings, we expect to observe a negativity in the event-related potential relative to pattern onset, which differs in amplitude between reference patterns and patterns that occur only once. With respect to the manipulated aspects of the learning context, we assume that temporal regularity and attention facilitate learning, such that the learning effect is larger in blocks with regular compared to blocks with jittered sequences, and in the session with attention to the pattern repetitions compared to the session without attention. Critically, a significant learning effect in the jittered blocks in the session without attention to the auditory stimulation would indicate that random acoustic patterns can be memorised incidentally even in the absence of periodicity as a cue for pattern onset.

Conflicting patterns of hyper- and hypo-cooperative behavioral strategies with peers have emerged among abused and neglected youth using game-theoretical tasks (Keil et al., 2019; Pitula et al., 2017). Physiological concomitants of cooperation may offer a means to disentangle differential neurocognitive pathways through which early life adversity gives rise to these behavioral strategies. High-frequency heart rate variability (hf-HRV) is considered a correlate of self-regulation and positive social engagement, as well as a mediator between early life adversity and psychopathology (Beauchaine & Thayer, 2015; Sigrist et al., 2021). Thus, hf-HRV offers a promising candidate to further elucidate such neurocognitive mechanisms. We therefore aimed to track hf-HRV while participants interact with cooperative and selfish peers. For this study a subset of approximately 240 12 to 22-year-olds, were drawn from a prospective longitudinal study (AMIS), oversampling for risk of adversity to investigate the sequelae of childhood maltreatment (White et al., 2015). While caregivers completed an interview to assess maltreatment experiences, participants were ostensibly connected with online co-players, to play a computerised public goods game (Keil et al., 2017). Meanwhile, ECG was collected to estimate local power (LP), an established index of parasympathetic activity with high temporal resolution (Bornemann et al., 2016). We predict that at-risk adolescents will show hyper-cooperative behavior coupled with reduced LP (i.e., greater vagal withdrawal; Young-Southward et al., 2020).

When we remember personally experienced events from the past, we mostly rely on our episodic memory, which declines with age. Older people begin to be less confident in their memory, preceding decline of memory accuracy in standardized tests [1]. Our research group examines the factors leading to the lower confidence and subjective experience of declining memory. Our aim is to investigate the change in awareness and understanding of episodic memory accuracy from younger to older age. This study extends previous work [2] within the Experimental And Social PsYchology (EASY) research group at the IoPPN examining changes in memory awareness, referred to as metacognition. The experiment includes encoding of a new video stimulus, simulating a day around London, providing a more naturalistic and contextual approach to test episodic memory in a controlled experimental environment. [1] Wong, J. T., Cramer, S. J., & Gallo, D. A. (2012). Age-related reduction of the confidence–accuracy relationship in episodic memory: Effects of recollection quality and retrieval monitoring. Psychology and Aging, 27(4), 1053–1065.  [2] Kapsetaki, M. E., Militaru, I. E., Sanguino, I., Boccanera, M., Zaara, N., Zaman, A., Loreto, F., Malhotra, P. A., & Russell, C. (Accepted/In press). Type of encoded material and age modulate the relationship between episodic recall of visual perspective and autobiographical memory. Journal Of Cognitive Psychology.

People who have experienced COVID-19 have reported a number of psychophysiological effects following their recovery from the illness. Some of the psychological complications seem to stem from a phenomenon named brain fog which might interfere with the psychological processes that underlie the process of memory encoding and recall (Hellmuth et al., 2021). Memory is one of the most fundamental cognitive neuropsychological processes that aid human function. This paper presents data from a sample of 80 participants aged 18-26, which includes 40 participants who have been exposed to COVID-19 and 40 unaffected individuals, in order to examine the differences in their performance in the Deese-Roediger-McDermott paradigm between the participants from the two groups, testing the extent of false memory in them. The data from the two groups would be subjected to tests of mean difference after testing for normality and SPSS-20 would be used for the statistical analysis for this study. Results would be discussed in the full length paper and the findings will help us understand whether there is a contribution of COVID-19 infection on the formation of false memories and the respective direction. Knowing and understanding this phenomenon might make healthcare providers and other people more understanding of the long-term complications post-recovery. Keywords: false memory, DRM paradigm, COVID-19, cognitive psychology, brain-fog

The dual-stream architecture of core linguistic processes in the left hemisphere is well established (Friederici, 2011; Hickok & Poeppel, 2007). Previous fMRI research in our lab demonstrated a similar dual-stream organization in the right hemisphere for the processing of linguistic prosody (Sammler et al., 2015). It comprised an auditory ventral pathway connecting posterior (pSTS) and anterior superior temporal sulcus (aSTS), and auditory-motor dorsal pathways connecting pSTS and inferior frontal gyrus (IFG) and premotor cortex (PMC). However, it remains to be shown what information is represented and exchanged within this network and at which points in time. To this end, magnetoencephalography data were collected from 34 native German listeners while they listened to single words (“Bar” [bar], “Paar” [pair]) that gradually varied in prosody (statement – question) and word-initial phoneme (/b/ – /p/) along orthogonal continua generated by audio morphing. Participants categorized these words in terms of either prosody or phoneme in alternating blocks. We will analyze the data in three major steps. First, we will use minimum-norm estimation to localize and compare neural activity evoked during the two tasks. This is to replicate the dual-stream architecture of prosody perception and to identify the temporal dynamics of the contributing regions. Searchlight-based multivariate pattern analysis (MVPA) will complement this comparison to identify task-related regions of interest for the following steps. Second, a representational similarity analysis (RSA) will be conducted to identify and dissociate regions and time points that represent the pure acoustics of the prosodic contours or the abstract categories of statement or question. Therefore, the neural activity patterns across stimuli will be correlated with modeled activity patterns based on either the acoustic or categorical (behavioural) dissimilarity of the stimuli. We expect to find evidence for acoustic representations in early rather than late time windows and predominantly in pSTS, while activity patterns in later time windows and aSTS, IFG, and PMC may be more strongly correlated with the categorical model. Finally, we will analyze the transfer entropy of the neural activity patterns between regions to explore the directed information transfer within the network and across time. Overall, this study will illuminate the representational dynamics of speech prosody in the brain.

Communicative cues such as eye contact have been shown to increase infants’ brain activation in response to visual stimuli (Hoehl et al., 2014; Hutman et al., 2015) and promote shared attention in early development (Hoehl & Bertenthal, 2021; Siposova & Carpenter, 2019). In this study, we assessed whether communicative cues (i.e. eye-contact, infant-directed speech [IDS] and pointing) between infant and caregiver enhance dyads’ mutual neural processes. For this we applied rhythmic visual stimulation in a dual electroencephalography (EEG) paradigm measuring simultaneously the brain activity of 11-12-month-old infants and their mothers (N = 49). To track mutual visual processing, we presented images flickered at 4 Hz, depicting natural objects in front of a natural background (Cichy et al., 2016). Flickering images elicited neural responses (steady-state visually evoked potentials [SSVEPs]) at 4 Hz that were recorded with EEG (Köster et al., 2017) in order to assess dynamic changes in dyads’ shared attention. Dyads observed the images in two conditions: in a joint attention (JA) condition mothers communicatively showed the images to their infants by engaging in eye-contact, pointing and making a comment in IDS; whereas in a joint watching (JW) condition dyads watched the images without communicative engagement. We hypothesised that communicative cues during joint attention will increase infants’ and caregivers’ attention to objects (Robertson et al., 2012) and facilitate a greater similarity between their mutual neural processes (Yoshioka et al., 2021). To test attention increases, we will compare infants’ (and mothers’) visual processing (SSVEPs at 4 Hz) between JA and JW. Next, to assess mutual neural processes, we will compare amplitude envelope correlations between infants’ and mothers’ SSVEPs between conditions. This study will enable new insights into how communicative cues during joint attention modulate both infants’ and caregivers’ brain activities and shapes shared visual experience. It will also pave the way for applying rhythmic visual stimulation in social interactional studies with developmental samples.

Language and vision, as two fundamental capacities of human cognition, are dynamically interdependent. The research in the past decades has provided an increasingly detailed picture of human language and visual processing separately, yet more research is needed regarding their multimodal interaction. Most existing computational models cannot fully account for how real-world stimuli are represented in our minds and based on what cognitive mechanisms the knowledge is acquired to perform different tasks. Artificial neural networks which have made significant advances in computer vision and natural language processing, have the potential to address these limitations by providing a novel unified computational framework through which human language and vision can be both studied. On the other hand, the existing multimodal applications in language and vision can be enhanced with cognitive signals or theories, and the robustness, efficiency, and interpretability of these artificial intelligence systems can be further improved. The proposed work can be seen as a step toward human-level artificial intelligence, which helps machines learn and use natural language and vision as humans do, and at the same time further our understanding of human cognitive processing.

Several recent studies have demonstrated similarities between activation patterns in large neural language models and human brain responses to the same text. A recurring pattern is that model activations become more brain-like with layer depth: representations in emerging in middle and higher model layers show stronger similarity to human brain activity patterns. However, it is less investigated what content makes these representations more brain-like, and what model-internal components or operations contribute to this. In our research, the aim is to get closer to answering these questions by combining model-brain similarity analyses with interpretability techniques on NLP models themselves. In this study, we use Representational Similarity Analysis (RSA) to compare activity patterns in transformer language models to human fMRI signals during story reading. Next to using the hidden state embeddings generated at each model layer, we also extract individual activation vectors for each of the models' attention heads. We find that there are big differences in brain-similarity between heads, even within model layers. Moreover, we find that some heads reach higher similarity scores than the highest scoring layer embeddings. Given the wide range of analysis methods for interpreting attention patterns in transformer models, this is a promising step towards our goal of better understanding what learned model behaviour gives rise to more brain-like language representations. We add some first explorations of the models' attention behaviour across layers and heads, aiming to see if any particular pattern aligns consistently with higher brain-similarity.

Motivation: Enhancing trust to a higher level is known from the literature to result in effectiveness increases (from monitoring activities acceleration to higher propensity for whistle-blowing and higher tax compliance on a voluntary basis), in better-educated individuals more concerned about public goods with a greater level of vaccination and less myopic decisions. The higher is trust the stronger is the legal system, the smoother are adaptation changes and the faster are rates of economic growth and eventually, the greater is level of welfare. Fairness -- which varies significantly across individuals and play a crucial role in decision-making -- has also been thoroughly studied. Nevertheless, to the best of our knowledge, there is no explicit research on how individuals' fairness satisfaction affect interpersonal trust. Research question: How does the (dis)satisfaction of subjective fairness preferences promote building trust? Potential contribution: If documented, a new channel of interpersonal trust-building will be revealed. The major implications are that under a high level of interpersonal trust (i) inequality mitigates and (ii) fairness rules converge to strict egalitarianism. Regarding policy implications, it can be called that greater plurality/dimensionality of fairness is needed: in social inclusivity, in the taxation system. In case the effect is observed, this study can contribute to several pieces of literature: on trust and the social context, on beliefs-dependence, and on trust dimensionality and heterogeneity.

Until recently, Theory of Mind (ToM) – the ability to ascribe mental states to others – was assumed to develop around the age of 4 years. In the past, this capacity was tested using verbal false belief tasks. The last 1.5 decades provided data suggesting ToM abilities already before 2 years of age using non-verbal false belief tasks. Recent data indicates that verbal and non-verbal ToM abilities may rely on different processing systems, often referred to as implicit and explicit ToM. The system underlying implicit ToM, however, requires further investigation and a direct comparison to explicit ToM is missing. We adapted an existing false belief task – the Sandbox task (Bernstein, 2011) – which yields a continuous measure for explicit ToM by measuring an egocentric bias in one’s judgement of another person’s false belief about an object location. In addition, we developed an implicit version of this task, which measures an altercentric bias in one’s own judgement of an object’s location. These two versions allow for explicit and implicit ToM to be measured within the same task. To establish the task, we tested it in an online study with 96 adults, using Bayesian sequential hypothesis testing. We found evidence for an altercentric bias that depended on the order in which the implicit/explicit tasks were conducted. The altercentric bias was only present when participants conducted the explicit task first (BF = 7.33). This indicates that altercentric biases that have been suggested to constitute automatic implicit ToM processes may in fact not occur automatically independently of the relevance of the other’s perspective for the task.

Purpose
Accurate quantification of relaxation times is essential for studying the structural composition of the human brain (Edwards et al. 2018). However, estimating T2 at high field-strength is hampered by potential experimental bias and long scan times. When using 2D Multi-Echo Spin-Echo (MESE) sequences (Carr et al. 1954; Meiboom et al. 1958), stimulated echoes caused by inhomogeneities of the B1 transmit field and slice profile degradation violate the assumption of a simple exponential signal decay (Van De Moortele et al. 2005; Uddin et al. 2013). The problem can be solved by simulating the signal response to a given MESE sequence using the Bloch equations and searching for the best matching curve to the acquired data (Ben-Eliezer et al. 2015). We investigated the accuracy and robustness of this method applied to data acquired with sub-millimetre resolution at 7 T.
Material & Methods
All measurements were conducted on a Siemens Magnetom 7 T scanner using a 32-channel RF receive head coil. A standard single spin-echo (SSE) technique (Hahn 1950) was set up for reference (TE = 12 − 150ms,TR = 4500ms), which took ∼ 33min, at 1 mm isotropic resolution. An aligned MESE sequence was used with nominal refocusing pulses of 180°. SAR limitations caused a reduction in echo-train length to ETL = 15 (echo-spacing = 12.1 ms) and the tailored VERSE gradients (Hargreaves et al. 2004). Total scan time was ∼ 7min for a 1 mm and ∼ 14min for a 0.6 mm isotropic resolution scan.
Results
T2 maps were estimated for both SSE and MESE measurments. Conventionally modelled MESE T2 was overestimated ∼2-fold for all tissue compartments. Quantitative agreement was achieved when using the simulated signal predictions for fitting. The values are in good agreement with previously reported 7 T data (Zhu et al. 2014). High resolution of 0.6 mm isotropic was possible, although, imperfect bias corrections were still apparent in temporal brain regions, where B1 inhomogeneities at 7 T are typically pronounced and signal dropouts can get too severe for proper correction.
Conclusion
We demonstrate T2 mapping with sub-millimetre resolution at 7 T for in vivo using a generalizable analysis pipeline. At high field strength, bias transmit field effects lead to an overestimation of the T2 values when modelling MESE signal with a conventinal signal description. The proper simulation of the signal evolution can solve those problems and is key to accurate quantitative T2.

Repetition suppression (RS), the reduced neural response to a repeated presentation of a stimulus, is a phenomenon consistently reported in many sensory domains (Grill-Spector et al., 2006). Different explanations have been proposed for RS, disagreeing on the influence of expectations. Predictive processing models take expectations and their violations (errors) into account, proposing that repetitions are predictable and therefore elicit smaller error responses (Auksztulewicz & Friston, 2016). While studies in the auditory and visual domain support such a predictive processing account, results in the domain of pain are sparse and remain inconclusive (Valentini et al., 2011, Wang et al., 2010). Here, we thus aim to elucidate the influence of expectations on RS of nociceptive stimuli. Therefore, we adapted an established paradigm from the auditory domain (Todorovic et al., 2011). Participants (N = 12) received short CO2 laser stimuli (125 ms). Each trial started with one stimulus, which could be repeated after 1s or not. Per block, these repetitions were either frequent (and therefore expected) or rare (and therefore surprising). Participants were informed about the nature of the upcoming block to elicit specific expectations. We acquired EEG data and measures of autonomic nervous system (ANS) activation (skin conductance and pupil dilation responses) to investigate different response levels. First, we confirmed that participants were able to distinguish between single and repeated stimuli (81% correct responses) and to form the desired expectations regarding the stimulus repetitions (expected 73% vs. 20% stimulus repetitions in each condition). As our main measure of interest, we analysed the laser-evoked vertex potential (N2P2 complex). We were able to observe a clear RS effect, which however was not modified by the induced expectations (28% vs. 31% peak-to-peak amplitude reduction of the N2P2 for expected and unexpected repetitions respectively). These initial results clearly speak against a predictive processing account of RS in nociception. To test for the consistency of these findings across different processing levels, we will further analyse the earlier laser-evoked potential component N1 as well as ANS responses. Since predictions and violations thereof have been associated with different patterns of oscillatory activity (Ploner et al., 2017), time-frequency analyses will complement our approach.

Pain research in humans has mostly focused on the brain, even though the spinal cord is an initial processing site within the nociceptive pathway of the central nervous system and a key target of descending modulation (Todd, 2010; Fields, 2004). The application of functional magnetic resonance imaging (fMRI) to the human spinal cord is still a relatively young field of research and faces many challenges (Summer, Brooks, & Cohen-Adad, 2014). Here we aim to probe the limitations of task-based spinal fMRI by investigating the reliability of spinal cord BOLD responses to identical nociceptive stimulation across two consecutive days. In two fMRI sessions, 40 participants received phasic painful heat stimuli on their left forearm . 3T fMRI data were acquired using a z-shimmed gradient-echo EPI sequence (Finsterbusch, Eippert, & Buechel, 2012). Data were motion-corrected, high-pass filtered and subjected to rigorous denoising during first-level model estimation. Group-level analyses were carried out in a standardized space (De Leener et al., 2018) using multiple comparison correction via voxel-wise non-parametric permutation testing (p < 0.05). Test-retest reliability was quantified using the intraclass correlation coefficient (ICC(3,1)) (Shrout & Fleiss, 1979). At the group level, we found spatially specific BOLD responses at the expected location (ipsilateral dorsal part of segment C6) that demonstrated spatial overlap across both days. Based on this, participant-wise GLM parameter estimates as well as CNR estimates were extracted, which however demonstrated poor reliability across days (i.e. all ICCs < 0.3). We found that a heat pain stimulus as short as 1s is able to evoke a BOLD response in the ipsilateral dorsal horn in spinal cord segment C6, extending previous studies that used more powerful stimulation parameters (Sprenger et al., 2012). We observed spatially overlapping BOLD responses across two days, which were however strongly varying within individuals, resulting in poor reliability. Poor test-retest reliability of task-based spinal fMRI data based on heat pain has previously been reported (Weber et al., 2016), where reliability was assessed between runs within one day. Currently, the reliability of state-of-the-art task-based spinal fMRI in the context of pain processing is thus considerably lower than for brain fMRI (Quiton et al., 2014; Upadhyay, 2015).