Convolutional neural networks reveal properties of reach-to-grasp encoding in posterior parietal cortex.

Deep neural networks (DNNs) are widely adopted to decode motor states from both non-invasively and invasively recorded neural signals, e.g., for realizing brain-computer interfaces. However, the neurophysiological interpretation of how DNNs make the decision based on the input neural activity is limitedly addressed, especially when applied to invasively recorded data. This reduces decoder reliability and transparency, and prevents the exploitation of decoders to better comprehend motor neural encoding. Here, we adopted an explainable artificial intelligence approach - based on a convolutional neural network and an explanation technique - to reveal spatial and temporal neural properties of reach-to-grasping from single-neuron recordings of the posterior parietal area V6A. The network was able to accurately decode 5 different grip types, and the explanation technique automatically identified the cells and temporal samples that most influenced the network prediction. Grip encoding in V6A neurons already started at movement preparation, peaking during movement execution. A difference was found within V6A: dorsal V6A neurons progressively encoded more for increasingly advanced grips, while ventral V6A neurons for increasingly rudimentary grips, with both subareas following a linear trend between the amount of grip encoding and the level of grip skills. By revealing the elements of the neural activity most relevant for each grip with no a priori assumptions, our approach supports and advances current knowledge about reach-to-grasp encoding in V6A, and it may represent a general tool able to investigate neural correlates of motor or cognitive tasks (e.g., attention and memory tasks) from single-neuron recordings.

Decoding movement kinematics from EEG using an interpretable convolutional neural network.

Continuous decoding of hand kinematics has been recently explored for the intuitive control of electroencephalography (EEG)-based Brain-Computer Interfaces (BCIs). Deep neural networks (DNNs) are emerging as powerful decoders, for their ability to automatically learn features from lightly pre-processed signals. However, DNNs for kinematics decoding lack in the interpretability of the learned features and are only used to realize within-subject decoders without testing other training approaches potentially beneficial for reducing calibration time, such as transfer learning. Here, we aim to overcome these limitations by using an interpretable convolutional neural network (ICNN) to decode 2-D hand kinematics (position and velocity) from EEG in a pursuit tracking task performed by 13 participants. The ICNN is trained using both within-subject and cross-subject strategies, and also testing the feasibility of transferring the knowledge learned on other subjects on a new one. Moreover, the network eases the interpretation of learned spectral and spatial EEG features. Our ICNN outperformed most of the other state-of-the-art decoders, showing the best trade-off between performance, size, and training time. Furthermore, transfer learning improved kinematics prediction in the low data regime. The network attributed the highest relevance for decoding to the delta-band across all subjects, and to higher frequencies (alpha, beta, low-gamma) for a cluster of them; contralateral central and parieto-occipital sites were the most relevant, reflecting the involvement of sensorimotor, visual and visuo-motor processing. The approach improved the quality of kinematics prediction from the EEG, at the same time allowing interpretation of the most relevant spectral and spatial features.

Changes in brain rhythms and connectivity tracking fear acquisition and reversal.

Fear conditioning is used to investigate the neural bases of threat and anxiety, and to understand their flexible modifications when the environment changes. This study aims to examine the temporal evolution of brain rhythms using electroencephalographic signals recorded in healthy volunteers during a protocol of Pavlovian fear conditioning and reversal. Power changes and Granger connectivity in theta, alpha, and gamma bands are investigated from neuroelectrical activity reconstructed on the cortex. Results show a significant increase in theta power in the left (contralateral to electrical shock) portion of the midcingulate cortex during fear acquisition, and a significant decrease in alpha power in a broad network over the left posterior-frontal and parietal cortex. These changes occur since the initial trials for theta power, but require more trials (3/4) to develop for alpha, and are also present during reversal, despite being less pronounced. In both bands, relevant changes in connectivity are mainly evident in the last block of reversal, just when power differences attenuate. No significant changes in the gamma band were detected. We conclude that the increased theta rhythm in the cingulate cortex subserves fear acquisition and is transmitted to other cortical regions via increased functional connectivity allowing a fast theta synchronization, whereas the decrease in alpha power can represent a partial activation of motor and somatosensory areas contralateral to the shock side in the presence of a dangerous stimulus. In addition, connectivity changes at the end of reversal may reflect long-term alterations in synapses necessary to reverse the previously acquired contingencies.

Motor decoding from the posterior parietal cortex using deep neural networks.

Objective.Motor decoding is crucial to translate the neural activity for brain-computer interfaces (BCIs) and provides information on how motor states are encoded in the brain. Deep neural networks (DNNs) are emerging as promising neural decoders. Nevertheless, it is still unclear how different DNNs perform in different motor decoding problems and scenarios, and which network could be a good candidate for invasive BCIs.Approach.Fully-connected, convolutional, and recurrent neural networks (FCNNs, CNNs, RNNs) were designed and applied to decode motor states from neurons recorded from V6A area in the posterior parietal cortex (PPC) of macaques. Three motor tasks were considered, involving reaching and reach-to-grasping (the latter under two illumination conditions). DNNs decoded nine reaching endpoints in 3D space or five grip types using a sliding window approach within the trial course. To evaluate decoders simulating a broad variety of scenarios, the performance was also analyzed while artificially reducing the number of recorded neurons and trials, and while performing transfer learning from one task to another. Finally, the accuracy time course was used to analyze V6A motor encoding.Main results.DNNs outperformed a classic Naïve Bayes classifier, and CNNs additionally outperformed XGBoost and Support Vector Machine classifiers across the motor decoding problems. CNNs resulted the top-performing DNNs when using less neurons and trials, and task-to-task transfer learning improved performance especially in the low data regime. Lastly, V6A neurons encoded reaching and reach-to-grasping properties even from action planning, with the encoding of grip properties occurring later, closer to movement execution, and appearing weaker in darkness.Significance.Results suggest that CNNs are effective candidates to realize neural decoders for invasive BCIs in humans from PPC recordings also reducing BCI calibration times (transfer learning), and that a CNN-based data-driven analysis may provide insights about the encoding properties and the functional roles of brain regions.

Deep learning applied to EEG source-data reveals both ventral and dorsal visual stream involvement in holistic processing of social stimuli.

Perception of social stimuli (faces and bodies) relies on "holistic" (i.e., global) mechanisms, as supported by picture-plane inversion: perceiving inverted faces/bodies is harder than perceiving their upright counterpart. Albeit neuroimaging evidence suggested involvement of face-specific brain areas in holistic processing, their spatiotemporal dynamics and selectivity for social stimuli is still debated. Here, we investigate the spatiotemporal dynamics of holistic processing for faces, bodies and houses (adopted as control non-social category), by applying deep learning to high-density electroencephalographic signals (EEG) at source-level. Convolutional neural networks were trained to classify cortical EEG responses to stimulus orientation (upright/inverted), separately for each stimulus type (faces, bodies, houses), resulting to perform well above chance for faces and bodies, and close to chance for houses. By explaining network decision, the 150-200 ms time interval and few visual ventral-stream regions were identified as mostly relevant for discriminating face and body orientation (lateral occipital cortex, and for face only, precuneus cortex, fusiform and lingual gyri), together with two additional dorsal-stream areas (superior and inferior parietal cortices). Overall, the proposed approach is sensitive in detecting cortical activity underlying perceptual phenomena, and by maximally exploiting discriminant information contained in data, may reveal spatiotemporal features previously undisclosed, stimulating novel investigations.

Modulations of Cortical Power and Connectivity in Alpha and Beta Bands during the Preparation of Reaching Movements.

Planning goal-directed movements towards different targets is at the basis of common daily activities (e.g., reaching), involving visual, visuomotor, and sensorimotor brain areas. Alpha (8-13 Hz) and beta (13-30 Hz) oscillations are modulated during movement preparation and are implicated in correct motor functioning. However, how brain regions activate and interact during reaching tasks and how brain rhythms are functionally involved in these interactions is still limitedly explored. Here, alpha and beta brain activity and connectivity during reaching preparation are investigated at EEG-source level, considering a network of task-related cortical areas. Sixty-channel EEG was recorded from 20 healthy participants during a delayed center-out reaching task and projected to the cortex to extract the activity of 8 cortical regions per hemisphere (2 occipital, 2 parietal, 3 peri-central, 1 frontal). Then, we analyzed event-related spectral perturbations and directed connectivity, computed via spectral Granger causality and summarized using graph theory centrality indices (in degree, out degree). Results suggest that alpha and beta oscillations are functionally involved in the preparation of reaching in different ways, with the former mediating the inhibition of the ipsilateral sensorimotor areas and disinhibition of visual areas, and the latter coordinating disinhibition of the contralateral sensorimotor and visuomotor areas.

Theta and alpha power track the acquisition and reversal of threat predictions and correlate with skin conductance response.

The ability to flexibly adjust one's threat predictions to meet the current environmental contingencies is crucial to survival. Nevertheless, its neural oscillatory correlates remain elusive in humans. Here, we tested whether changes in theta and alpha brain oscillations mark the updating of threat predictions and correlate with response of the peripheral nervous system. To this end, electroencephalogram and electrodermal activity were recorded in a group of healthy adults, who completed a Pavlovian threat conditioning task that included an acquisition and a reversal phase. Both theta and alpha power discriminated between threat and safety, with each frequency band showing unique patterns of modulations during acquisition and reversal. While changes in midcingulate theta power may learn the timing of an upcoming danger, alpha power may reflect the preparation of the somato-motor system. Additionally, ventromedial prefrontal cortex theta may play a role in the inhibition of previously acquired threat responses, when they are no longer appropriate. Finally, theta and alpha power correlated with skin conductance response, establishing a direct relationship between activation of the central and peripheral nervous systems. Taken together these results highlight the existence of multiple oscillatory systems that flexibly regulate their activity for the successful expression of threat responses in an ever-changing environment.

A neurocomputational analysis of visual bias on bimanual tactile spatial perception during a crossmodal exposure.

Vision and touch both support spatial information processing. These sensory systems also exhibit highly specific interactions in spatial perception, which may reflect multisensory representations that are learned through visuo-tactile (VT) experiences. Recently, Wani and colleagues reported that task-irrelevant visual cues bias tactile perception, in a brightness-dependent manner, on a task requiring participants to detect unimanual and bimanual cues. Importantly, tactile performance remained spatially biased after VT exposure, even when no visual cues were presented. These effects on bimanual touch conceivably reflect cross-modal learning, but the neural substrates that are changed by VT experience are unclear. We previously described a neural network capable of simulating VT spatial interactions. Here, we exploited this model to test different hypotheses regarding potential network-level changes that may underlie the VT learning effects. Simulation results indicated that VT learning effects are inconsistent with plasticity restricted to unisensory visual and tactile hand representations. Similarly, VT learning effects were also inconsistent with changes restricted to the strength of inter-hemispheric inhibitory interactions. Instead, we found that both the hand representations and the inter-hemispheric inhibitory interactions need to be plastic to fully recapitulate VT learning effects. Our results imply that crossmodal learning of bimanual spatial perception involves multiple changes distributed over a VT processing cortical network.

Bottom-up vs. top-down connectivity imbalance in individuals with high-autistic traits: An electroencephalographic study.

Brain connectivity is often altered in autism spectrum disorder (ASD). However, there is little consensus on the nature of these alterations, with studies pointing to either increased or decreased connectivity strength across the broad autism spectrum. An important confound in the interpretation of these contradictory results is the lack of information about the directionality of the tested connections. Here, we aimed at disambiguating these confounds by measuring differences in directed connectivity using EEG resting-state recordings in individuals with low and high autistic traits. Brain connectivity was estimated using temporal Granger Causality applied to cortical signals reconstructed from EEG. Between-group differences were summarized using centrality indices taken from graph theory (in degree, out degree, authority, and hubness). Results demonstrate that individuals with higher autistic traits exhibited a significant increase in authority and in degree in frontal regions involved in high-level mechanisms (emotional regulation, decision-making, and social cognition), suggesting that anterior areas mostly receive information from more posterior areas. Moreover, the same individuals exhibited a significant increase in the hubness and out degree over occipital regions (especially the left and right pericalcarine regions, where the primary visual cortex is located), suggesting that these areas mostly send information to more anterior regions. Hubness and authority appeared to be more sensitive indices than the in degree and out degree. The observed brain connectivity differences suggest that, in individual with higher autistic traits, bottom-up signaling overcomes top-down channeled flow. This imbalance may contribute to some behavioral alterations observed in ASD.

Relationship between electroencephalographic data and comfort perception captured in a Virtual Reality design environment of an aircraft cabin.

Successful aircraft cabin design depends on how the different stakeholders are involved since the first phases of product development. To predict passenger satisfaction prior to the manufacturing phase, human response was investigated in a Virtual Reality (VR) environment simulating a cabin aircraft. Subjective assessments of virtual designs have been collected via questionnaires, while the underlying neural mechanisms have been captured through electroencephalographic (EEG) data. In particular, we focused on the modulation of EEG alpha rhythm as a valuable marker of the brain's internal state and investigated which changes in alpha power and connectivity can be related to a different visual comfort perception by comparing groups with higher and lower comfort rates. Results show that alpha-band power decreased in occipital regions during subjects' immersion in the virtual cabin compared with the relaxation state, reflecting attention to the environment. Moreover, alpha-band power was modulated by comfort perception: lower comfort was associated with a lower alpha power compared to higher comfort. Further, alpha-band Granger connectivity shows top-down mechanisms in higher comfort participants, modulating attention and restoring partial relaxation. Present results contribute to understanding the role of alpha rhythm in visual comfort perception and demonstrate that VR and EEG represent promising tools to quantify human-environment interactions.

A Bayesian-optimized design for an interpretable convolutional neural network to decode and analyze the P300 response in autism.

Objective.P300 can be analyzed in autism spectrum disorder (ASD) to derive biomarkers and can be decoded in brain-computer interfaces to reinforce ASD impaired skills. Convolutional neural networks (CNNs) have been proposed for P300 decoding, outperforming traditional algorithms but they (a) do not investigate optimal designs in different training conditions; (b) lack in interpretability. To overcome these limitations, an interpretable CNN (ICNN), that we recently proposed for motor decoding, has been modified and adopted here, with its optimal design searched via Bayesian optimization.Approach.The ICNN provides a straightforward interpretation of spectral and spatial features learned to decode P300. The Bayesian-optimized (BO) ICNN design was investigated separately for different training strategies (within-subject, within-session, and cross-subject) and BO models were used for the subsequent analyses. Specifically, transfer learning (TL) potentialities were investigated by assessing how pretrained cross-subject BO models performed on a new subject vs. random-initialized models. Furthermore, within-subject BO-derived models were combined with an explanation technique (ICNN + ET) to analyze P300 spectral and spatial features.Main results.The ICNN resulted comparable or even outperformed existing CNNs, at the same time being lighter. BO ICNN designs differed depending on the training strategy, needing more capacity as the training set variability increased. Furthermore, TL provided higher performance than networks trained from scratch. The ICNN + ET analysis suggested the frequency range [2, 5.8] Hz as the most relevant, and spatial features showed a right-hemispheric parietal asymmetry. The ICNN + ET-derived features, but not ERP-derived features, resulted significantly and highly correlated to autism diagnostic observation schedule clinical scores.Significance.This study substantiates the idea that a CNN can be designed both accurate and interpretable for P300 decoding, with an optimized design depending on the training condition. The novel ICNN-based analysis tool was able to better capture ASD neural signatures than traditional event-related potential analysis, possibly paving the way for identifying novel biomarkers.

Decoding sensorimotor information from superior parietal lobule of macaque via Convolutional Neural Networks.

Despite the well-recognized role of the posterior parietal cortex (PPC) in processing sensory information to guide action, the differential encoding properties of this dynamic processing, as operated by different PPC brain areas, are scarcely known. Within the monkey's PPC, the superior parietal lobule hosts areas V6A, PEc, and PE included in the dorso-medial visual stream that is specialized in planning and guiding reaching movements. Here, a Convolutional Neural Network (CNN) approach is used to investigate how the information is processed in these areas. We trained two macaque monkeys to perform a delayed reaching task towards 9 positions (distributed on 3 different depth and direction levels) in the 3D peripersonal space. The activity of single cells was recorded from V6A, PEc, PE and fed to convolutional neural networks that were designed and trained to exploit the temporal structure of neuronal activation patterns, to decode the target positions reached by the monkey. Bayesian Optimization was used to define the main CNN hyper-parameters. In addition to discrete positions in space, we used the same network architecture to decode plausible reaching trajectories. We found that data from the most caudal V6A and PEc areas outperformed PE area in the spatial position decoding. In all areas, decoding accuracies started to increase at the time the target to reach was instructed to the monkey, and reached a plateau at movement onset. The results support a dynamic encoding of the different phases and properties of the reaching movement differentially distributed over a network of interconnected areas. This study highlights the usefulness of neurons' firing rate decoding via CNNs to improve our understanding of how sensorimotor information is encoded in PPC to perform reaching movements. The obtained results may have implications in the perspective of novel neuroprosthetic devices based on the decoding of these rich signals for faithfully carrying out patient's intentions.

Neural Networks and Connectivity among Brain Regions.

As is widely understood, brain functioning depends on the interaction among several neural populations, which are linked via complex connectivity circuits and work together (in antagonistic or synergistic ways) to exchange information, synchronize their activity, adapt plastically to external stimuli or internal requirements, and more generally to participate in solving multifaceted cognitive tasks [...].

The Directionality of Fronto-Posterior Brain Connectivity Is Associated with the Degree of Individual Autistic Traits.

Altered patterns of brain connectivity have been found in autism spectrum disorder (ASD) and associated with specific symptoms and behavioral features. Growing evidence suggests that the autistic peculiarities are not confined to the clinical population but extend along a continuum between healthy and maladaptive conditions. The aim of this study was to investigate whether a differentiated connectivity pattern could also be tracked along the continuum of autistic traits in a non-clinical population. A Granger causality analysis conducted on a resting-state EEG recording showed that connectivity along the posterior-frontal gradient is sensitive to the magnitude of individual autistic traits and mostly conveyed through fast oscillatory activity. Specifically, participants with higher autistic traits were characterized by a prevalence of ascending connections starting from posterior regions ramping the cortical hierarchy. These findings point to the presence of a tendency within the neural mapping of individuals with higher autistic features in conveying proportionally more bottom-up information. This pattern of findings mimics those found in clinical forms of autism, supporting the idea of a neurobiological continuum between autistic traits and ASD.

A Novel Method to Assess Motor Cortex Connectivity and Event Related Desynchronization Based on Mass Models.

Knowledge of motor cortex connectivity is of great value in cognitive neuroscience, in order to provide a better understanding of motor organization and its alterations in pathological conditions. Traditional methods provide connectivity estimations which may vary depending on the task. This work aims to propose a new method for motor connectivity assessment based on the hypothesis of a task-independent connectivity network, assuming nonlinear behavior. The model considers six cortical regions of interest (ROIs) involved in hand movement. The dynamics of each region is simulated using a neural mass model, which reproduces the oscillatory activity through the interaction among four neural populations. Parameters of the model have been assigned to simulate both power spectral densities and coherences of a patient with left-hemisphere stroke during resting condition, movement of the affected, and movement of the unaffected hand. The presented model can simulate the three conditions using a single set of connectivity parameters, assuming that only inputs to the ROIs change from one condition to the other. The proposed procedure represents an innovative method to assess a brain circuit, which does not rely on a task-dependent connectivity network and allows brain rhythms and desynchronization to be assessed on a quantitative basis.

A Lightweight Multi-Scale Convolutional Neural Network for P300 Decoding: Analysis of Training Strategies and Uncovering of Network Decision.

Convolutional neural networks (CNNs), which automatically learn features from raw data to approximate functions, are being increasingly applied to the end-to-end analysis of electroencephalographic (EEG) signals, especially for decoding brain states in brain-computer interfaces (BCIs). Nevertheless, CNNs introduce a large number of trainable parameters, may require long training times, and lack in interpretability of learned features. The aim of this study is to propose a CNN design for P300 decoding with emphasis on its lightweight design while guaranteeing high performance, on the effects of different training strategies, and on the use of post-hoc techniques to explain network decisions. The proposed design, named MS-EEGNet, learned temporal features in two different timescales (i.e., multi-scale, MS) in an efficient and optimized (in terms of trainable parameters) way, and was validated on three P300 datasets. The CNN was trained using different strategies (within-participant and within-session, within-participant and cross-session, leave-one-subject-out, transfer learning) and was compared with several state-of-the-art (SOA) algorithms. Furthermore, variants of the baseline MS-EEGNet were analyzed to evaluate the impact of different hyper-parameters on performance. Lastly, saliency maps were used to derive representations of the relevant spatio-temporal features that drove CNN decisions. MS-EEGNet was the lightest CNN compared with the tested SOA CNNs, despite its multiple timescales, and significantly outperformed the SOA algorithms. Post-hoc hyper-parameter analysis confirmed the benefits of the innovative aspects of MS-EEGNet. Furthermore, MS-EEGNet did benefit from transfer learning, especially using a low number of training examples, suggesting that the proposed approach could be used in BCIs to accurately decode the P300 event while reducing calibration times. Representations derived from the saliency maps matched the P300 spatio-temporal distribution, further validating the proposed decoding approach. This study, by specifically addressing the aspects of lightweight design, transfer learning, and interpretability, can contribute to advance the development of deep learning algorithms for P300-based BCIs.

The Relationship between Oscillations in Brain Regions and Functional Connectivity: A Critical Analysis with the Aid of Neural Mass Models.

Propagation of brain rhythms among cortical regions is a relevant aspect of cognitive neuroscience, which is often investigated using functional connectivity (FC) estimation techniques. The aim of this work is to assess the relationship between rhythm propagation, FC and brain functioning using data generated from neural mass models of connected Regions of Interest (ROIs). We simulated networks of four interconnected ROIs, each with a different intrinsic rhythm (in θ, α, ß and γ ranges). Connectivity was estimated using eight estimators and the relationship between structural connectivity and FC was assessed as a function of the connectivity strength and of the inputs to the ROIs. Results show that the Granger estimation provides the best accuracy, with a good capacity to evaluate the connectivity strength. However, the estimated values strongly depend on the input to the ROIs and hence on nonlinear phenomena. When a population works in the linear region, its capacity to transmit a rhythm increases drastically. Conversely, when it saturates, oscillatory activity becomes strongly affected by rhythms incoming from other regions. Changes in functional connectivity do not always reflect a physical change in the synapses. A unique connectivity network can propagate rhythms in very different ways depending on the specific working conditions.

Alpha and theta mechanisms operating in internal-external attention competition.

Attention is the ability to prioritize a set of information at expense of others and can be internally- or externally-oriented. Alpha and theta oscillations have been extensively implicated in attention. However, it is unclear how these oscillations operate when sensory distractors are presented continuously during task-relevant internal processes, in close-to-real-life conditions. Here, EEG signals from healthy participants were obtained at rest and in three attentional conditions, characterized by the execution of a mental math task (internal attention), presentation of pictures on a monitor (external attention), and task execution under the distracting action of picture presentation (internal-external competition). Alpha and theta power were investigated at scalp level and at some cortical regions of interest (ROIs); moreover, functional directed connectivity was estimated via spectral Granger Causality. Results show that frontal midline theta was distinctive of mental task execution and was more prominent during competition compared to internal attention alone, possibly reflecting higher executive control; anterior cingulate cortex appeared as mainly involved and causally connected to distant (temporal/occipital) regions. Alpha power in visual ROIs strongly decreased in external attention alone, while it assumed values close to rest during competition, reflecting reduced visual engagement against distractors; connectivity results suggested that bidirectional alpha influences between frontal and visual regions could contribute to reduce visual interference in internal attention. This study can help to understand how our brain copes with internal-external attention competition, a condition intrinsic in the human sensory-cognitive interplay, and to elucidate the relationships between brain oscillations and attentional functions/dysfunctions in daily tasks.

From statistical regularities in multisensory inputs to peripersonal space representation and body ownership: Insights from a neural network model.

Peripersonal space (PPS), the interface between the self and the environment, is represented by a network of multisensory neurons with visual (or auditory) receptive fields anchored to specific body parts, and tactile receptive fields covering the same body parts. Neurophysiological and behavioural features of hand PPS representation have been previously modelled through a neural network constituted by one multisensory population integrating tactile inputs with visual/auditory external stimuli. Reference frame transformations were not explicitly modelled, as stimuli were encoded in pre-computed hand-centred coordinates. Here we present a novel model, aiming to overcome this limitation by including a proprioceptive population encoding hand position. We confirmed behaviourally the plausibility of the proposed architecture, showing that visuo-proprioceptive information is integrated to enhance tactile processing on the hand. Moreover, the network's connectivity was spontaneously tuned through a Hebbian-like mechanism, under two minimal assumptions. First, the plasticity rule was designed to learn the statistical regularities of visual, proprioceptive and tactile inputs. Second, such statistical regularities were simply those imposed by the body structure. The network learned to integrate proprioceptive and visual stimuli, and to compute their hand-centred coordinates to predict tactile stimulation. Through the same mechanism, the network reproduced behavioural correlates of manipulations implicated in subjective body ownership: the invisible and the rubber hand illusion. We thus propose that PPS representation and body ownership may emerge through a unified neurocomputational process; the integration of multisensory information consistently with a model of the body in the environment, learned from the natural statistics of sensory inputs.