Preventing antisocial robots: A pathway to artificial empathy.

Given the accelerating powers of artificial intelligence (AI), we must equip artificial agents and robots with empathy to prevent harmful and irreversible decisions. Current approaches to artificial empathy focus on its cognitive or performative processes, overlooking affect, and thus promote sociopathic behaviors. Artificially vulnerable, fully empathic AI is necessary to prevent sociopathic robots and protect human welfare.

Explainable AI: A review of applications to neuroimaging data.

Deep neural networks (DNNs) have transformed the field of computer vision and currently constitute some of the best models for representations learned via hierarchical processing in the human brain. In medical imaging, these models have shown human-level performance and even higher in the early diagnosis of a wide range of diseases. However, the goal is often not only to accurately predict group membership or diagnose but also to provide explanations that support the model decision in a context that a human can readily interpret. The limited transparency has hindered the adoption of DNN algorithms across many domains. Numerous explainable artificial intelligence (XAI) techniques have been developed to peer inside the "black box" and make sense of DNN models, taking somewhat divergent approaches. Here, we suggest that these methods may be considered in light of the interpretation goal, including functional or mechanistic interpretations, developing archetypal class instances, or assessing the relevance of certain features or mappings on a trained model in a post-hoc capacity. We then focus on reviewing recent applications of post-hoc relevance techniques as applied to neuroimaging data. Moreover, this article suggests a method for comparing the reliability of XAI methods, especially in deep neural networks, along with their advantages and pitfalls.

Neuroprotection in late life attention-deficit/hyperactivity disorder: A review of pharmacotherapy and phenotype across the lifespan.

For decades, psychostimulants have been the gold standard pharmaceutical treatment for attention-deficit/hyperactivity disorder (ADHD). In the United States, an astounding 9% of all boys and 4% of girls will be prescribed stimulant drugs at some point during their childhood. Recent meta-analyses have revealed that individuals with ADHD have reduced brain volume loss later in life (>60 y.o.) compared to the normal aging brain, which suggests that either ADHD or its treatment may be neuroprotective. Crucially, these neuroprotective effects were significant in brain regions (e.g., hippocampus, amygdala) where severe volume loss is linked to cognitive impairment and Alzheimer's disease. Historically, the ADHD diagnosis and its pharmacotherapy came about nearly simultaneously, making it difficult to evaluate their effects in isolation. Certain evidence suggests that psychostimulants may normalize structural brain changes typically observed in the ADHD brain. If ADHD itself is neuroprotective, perhaps exercising the brain, then psychostimulants may not be recommended across the lifespan. Alternatively, if stimulant drugs are neuroprotective, then this class of medications may warrant further investigation for their therapeutic effects. Here, we take a bottom-up holistic approach to review the psychopharmacology of ADHD in the context of recent models of attention. We suggest that future studies are greatly needed to better appreciate the interactions amongst an ADHD diagnosis, stimulant treatment across the lifespan, and structure-function alterations in the aging brain.

Decoding Task-Based fMRI Data with Graph Neural Networks, Considering Individual Differences.

Task fMRI provides an opportunity to analyze the working mechanisms of the human brain during specific experimental paradigms. Deep learning models have increasingly been applied for decoding and encoding purposes study to representations in task fMRI data. More recently, graph neural networks, or neural networks models designed to leverage the properties of graph representations, have recently shown promise in task fMRI decoding studies. Here, we propose an end-to-end graph convolutional network (GCN) framework with three convolutional layers to classify task fMRI data from the Human Connectome Project dataset. We compared the predictive performance of our GCN model across four of the most widely used node embedding algorithms-NetMF, RandNE, Node2Vec, and Walklets-to automatically extract the structural properties of the nodes in the functional graph. The empirical results indicated that our GCN framework accurately predicted individual differences (0.978 and 0.976) with the NetMF and RandNE embedding methods, respectively. Furthermore, to assess the effects of individual differences, we tested the classification performance of the model on sub-datasets divided according to gender and fluid intelligence. Experimental results indicated significant differences in the classification predictions of gender, but not high/low fluid intelligence fMRI data. Our experiments yielded promising results and demonstrated the superior ability of our GCN in modeling task fMRI data.

Diurnal variations of resting-state fMRI data: A graph-based analysis.

Circadian rhythms (lasting approximately 24 h) control and entrain various physiological processes, ranging from neural activity and hormone secretion to sleep cycles and eating habits. Several studies have shown that time of day (TOD) is associated with human cognition and brain functions. In this study, utilizing a chronotype-based paradigm, we applied a graph theory approach on resting-state functional MRI (rs-fMRI) data to compare whole-brain functional network topology between morning and evening sessions and between morning-type (MT) and evening-type (ET) participants. Sixty-two individuals (31 MT and 31 ET) underwent two fMRI sessions, approximately 1 hour (morning) and 10 h (evening) after their wake-up time, according to their declared habitual sleep-wake pattern on a regular working day. In the global analysis, the findings revealed the effect of TOD on functional connectivity (FC) patterns, including increased small-worldness, assortativity, and synchronization across the day. However, we identified no significant differences based on chronotype categories. The study of the modular structure of the brain at mesoscale showed that functional networks tended to be more integrated with one another in the evening session than in the morning session. Local/regional changes were affected by both factors (i.e., TOD and chronotype), mostly in areas associated with somatomotor, attention, frontoparietal, and default networks. Furthermore, connectivity and hub analyses revealed that the somatomotor, ventral attention, and visual networks covered the most highly connected areas in the morning and evening sessions: the latter two were more active in the morning sessions, and the first was identified as being more active in the evening. Finally, we performed a correlation analysis to determine whether global and nodal measures were associated with subjective assessments across participants. Collectively, these findings contribute to an increased understanding of diurnal fluctuations in resting brain activity and highlight the role of TOD in future studies on brain function and the design of fMRI experiments.

Temporal embedding and spatiotemporal feature selection boost multi-voxel pattern analysis decoding accuracy.

BACKGROUND: In fMRI decoding, temporal embedding of spatial features of the brain allows the incorporation of brain activity dynamics into the multivariate pattern classification process, and provides enriched information about stimulus-specific response patterns and potentially improved prediction accuracy. NEW METHOD: This study investigates the possibility of enhancing the classification performance by exploring temporal embedding, to identify the optimum combination of spatiotemporal features based on their classification performance. We investigated the importance of spatiotemporal feature selection using a slow event-related design adapted from the classic Haxby study (Haxby et al., 2001). Data were collected using a multiband fMRI sequence with temporal resolution of 0.568 s. COMPARISON WITH EXISTING METHODS: A wide range of spatiotemporal observations were created as various combinations of spatiotemporal features. Using both random forest, and support vector machine, classifiers prediction accuracies for these combinations were then compared with the single spatial multivariate pattern approach that uses only a single temporal observation. RESULTS: Our findings showed that, on average, spatiotemporal feature selection improved prediction accuracy. Moreover, the random forest algorithm outperformed the support vector machine and benefitted from temporal information to a greater extent. CONCLUSIONS: As expected, the most influential temporal durations were found to be around the peak of the hemodynamic response function, a few seconds after the stimuli onset until -4 s after the peak of the hemodynamic response function. The superiority of spatiotemporal feature selection over single time-point spatial approaches invites future work to design optimal approaches that incorporate spatiotemporal dependencies into feature selection for decoding.

Hemodynamic latency is associated with reduced intelligence across the lifespan: an fMRI DCM study of aging, cerebrovascular integrity, and cognitive ability.

Changes in neurovascular coupling are associated with both Alzheimer's disease and vascular dementia in later life, but this may be confounded by cerebrovascular risk. We hypothesized that hemodynamic latency would be associated with reduced cognitive functioning across the lifespan, holding constant demographic and cerebrovascular risk. In 387 adults aged 18-85 (mean = 48.82), dynamic causal modeling was used to estimate the hemodynamic response function in the left and right V1 and V3-ventral regions of the visual cortex in response to a simple checkerboard block design stimulus with minimal cognitive demands. The hemodynamic latency (transit time) in the visual cortex was used to predict general cognitive ability (Full-Scale IQ), controlling for demographic variables (age, race, education, socioeconomic status) and cerebrovascular risk factors (hypertension, alcohol use, smoking, high cholesterol, BMI, type 2 diabetes, cardiac disorders). Increased hemodynamic latency in the visual cortex predicted reduced cognitive function (p < 0.05), holding constant demographic and cerebrovascular risk. Increased alcohol use was associated with reduced overall cognitive function (Full Scale IQ 2.8 pts, p < 0.05), while cardiac disorders (Full Scale IQ 3.3 IQ pts; p < 0.05), high cholesterol (Full Scale IQ 3.9 pts; p < 0.05), and years of education (2 IQ pts/year; p < 0.001) were associated with higher general cognitive ability. Increased hemodynamic latency was associated with reduced executive functioning (p < 0.05) as well as reductions in verbal concept formation (p < 0.05) and the ability to synthesize and analyze abstract visual information (p < 0.01). Hemodynamic latency is associated with reduced cognitive ability across the lifespan, independently of other demographic and cerebrovascular risk factors. Vascular health may predict cognitive ability long before the onset of dementias.

Predicting Empathy From Resting State Brain Connectivity: A Multivariate Approach.

Recent task fMRI studies suggest that individual differences in trait empathy and empathic concern are mediated by patterns of connectivity between self-other resonance and top-down control networks that are stable across task demands. An untested implication of this hypothesis is that these stable patterns of connectivity should be visible even in the absence of empathy tasks. Using machine learning, we demonstrate that patterns of resting state fMRI connectivity (i.e. the degree of synchronous BOLD activity across multiple cortical areas in the absence of explicit task demands) of resonance and control networks predict trait empathic concern (n = 58). Empathic concern was also predicted by connectivity patterns within the somatomotor network. These findings further support the role of resonance-control network interactions and of somatomotor function in our vicariously driven concern for others. Furthermore, a practical implication of these results is that it is possible to assess empathic predispositions in individuals without needing to perform conventional empathy assessments.

Effects of Chronic Sleep Restriction on the Brain Functional Network, as Revealed by Graph Theory.

Sleep is a complex and dynamic process for maintaining homeostasis, and a lack of sleep can disrupt whole-body functioning. No organ is as vulnerable to the loss of sleep as the brain. Accordingly, we examined a set of task-based functional magnetic resonance imaging (fMRI) data by using graph theory to assess brain topological changes in subjects in a state of chronic sleep restriction, and then identified diurnal variability in the graph-theoretic measures. Task-based fMRI data were collected in a 1.5T MR scanner from the same participants on two days: after a week of fully restorative sleep and after a week with 35% sleep curtailment. Each day included four scanning sessions throughout the day (at approximately 10:00 AM, 2:00 PM, 6:00 PM, and 10:00 PM). A modified spatial cueing task was applied to evaluate sustained attention. After sleep restriction, the characteristic path length significantly increased at all measurement times, and small-worldness significantly decreased. Assortativity, a measure of network fault tolerance, diminished over the course of the day in both conditions. Local graph measures were altered primarily across the limbic system (particularly in the hippocampus, parahippocampal gyrus, and amygdala), default mode network, and visual network.

Interpreting encoding and decoding models.

Encoding and decoding models are widely used in systems, cognitive, and computational neuroscience to make sense of brain-activity data. However, the interpretation of their results requires care. Decoding models can help reveal whether particular information is present in a brain region in a format the decoder can exploit. Encoding models make comprehensive predictions about representational spaces. In the context of sensory experiments, where stimuli are experimentally controlled, encoding models enable us to test and compare brain-computational theories. Encoding and decoding models typically include fitted linear-model components. Sometimes the weights of the fitted linear combinations are interpreted as reflecting, in an encoding model, the contribution of different sensory features to the representation or, in a decoding model, the contribution of different measured brain responses to a decoded feature. Such interpretations can be problematic when the predictor variables or their noise components are correlated and when priors (or penalties) are used to regularize the fit. Encoding and decoding models are evaluated in terms of their generalization performance. The correct interpretation depends on the level of generalization a model achieves (e.g. to new response measurements for the same stimuli, to new stimuli from the same population, or to stimuli from a different population). Significant decoding or encoding performance of a single model (at whatever level of generality) does not provide strong constraints for theory. Many models must be tested and inferentially compared for analyses to drive theoretical progress.

Inter-hemispheric Asymmetry Patterns in the ADHD Brain: A Neuroimaging Replication Study.

Attention-deficit/ hyperactivity disorder (ADHD) is the most common neurodevelopment disorder in children, and many genetic markers have been linked to the behavioral phenotypes of this highly heritable disease. The neuroimaging correlates are similarly complex, with multiple combinations of structural and functional alterations associated with the disease presentations of hyperactivity and inattentiveness. Thus far, neuroimaging studies have provided mixed results in ADHD patients, particularly with respect to the laterality of findings. It is possible that hemispheric asymmetry differences may help reconcile the variability of these findings. We recently reported that inter-hemispheric asymmetry differences were more sensitive descriptors for identifying differences between ADHD and typically developing (TD) brains (n=849) across volumetric, morphometric, and white matter neuroimaging metrics. Here, we examined the replicability of these findings across a new data set (n=202) of TD and ADHD subjects at the time of diagnosis (medication naive) and after a six week course of either stimulant drugs, non-stimulant medications, or combination therapy. Our findings replicated our earlier work across a number of volumetric and white matter measures confirming that asymmetry is more robust at detecting differences between TD and ADHD brains. However, the effects of medication failed to produce significant alterations across either lateralized or symmetry measures. We suggest that the delay in brain volume maturation observed in ADHD youths may be hemisphere dependent. Future work may investigate the extent to which these inter-hemispheric asymmetry differences are causal or compensatory in nature.

Neuroimaging of Traumatic Brain Injury.

The purpose of this article is to review conventional and advanced neuroimaging techniques performed in the setting of traumatic brain injury (TBI). The primary goal for the treatment of patients with suspected TBI is to prevent secondary injury. In the setting of a moderate to severe TBI, the most appropriate initial neuroimaging examination is a noncontrast head computed tomography (CT), which can reveal life-threatening injuries and direct emergent neurosurgical intervention. We will focus much of the article on advanced neuroimaging techniques including perfusion imaging and diffusion tensor imaging and discuss their potentials and challenges. We believe that advanced neuroimaging techniques may improve the accuracy of diagnosis of TBI and improve management of TBI.

Cognitive computational neuroscience.

To learn how cognition is implemented in the brain, we must build computational models that can perform cognitive tasks, and test such models with brain and behavioral experiments. Cognitive science has developed computational models that decompose cognition into functional components. Computational neuroscience has modeled how interacting neurons can implement elementary components of cognition. It is time to assemble the pieces of the puzzle of brain computation and to better integrate these separate disciplines. Modern technologies enable us to measure and manipulate brain activity in unprecedentedly rich ways in animals and humans. However, experiments will yield theoretical insight only when employed to test brain-computational models. Here we review recent work in the intersection of cognitive science, computational neuroscience and artificial intelligence. Computational models that mimic brain information processing during perceptual, cognitive and control tasks are beginning to be developed and tested with brain and behavioral data.

Perfusion Imaging in Acute Traumatic Brain Injury.

Traumatic brain injury (TBI) is a significant problem worldwide and neuroimaging plays a critical role in diagnosis and management. Recently, perfusion neuroimaging techniques have been explored in TBI to determine and characterize potential perfusion neuroimaging biomarkers to aid in diagnosis, treatment, and prognosis. In this article, computed tomography (CT) bolus perfusion, MR imaging bolus perfusion, MR imaging arterial spin labeling perfusion, and xenon CT are reviewed with a focus on their applications in acute TBI. Future research directions are also discussed.

Decoding the encoding of functional brain networks: An fMRI classification comparison of non-negative matrix factorization (NMF), independent component analysis (ICA), and sparse coding algorithms.

BACKGROUND: Brain networks in fMRI are typically identified using spatial independent component analysis (ICA), yet other mathematical constraints provide alternate biologically-plausible frameworks for generating brain networks. Non-negative matrix factorization (NMF) would suppress negative BOLD signal by enforcing positivity. Spatial sparse coding algorithms (L1 Regularized Learning and K-SVD) would impose local specialization and a discouragement of multitasking, where the total observed activity in a single voxel originates from a restricted number of possible brain networks. NEW METHOD: The assumptions of independence, positivity, and sparsity to encode task-related brain networks are compared; the resulting brain networks within scan for different constraints are used as basis functions to encode observed functional activity. These encodings are then decoded using machine learning, by using the time series weights to predict within scan whether a subject is viewing a video, listening to an audio cue, or at rest, in 304 fMRI scans from 51 subjects. RESULTS AND COMPARISON WITH EXISTING METHOD: The sparse coding algorithm of L1 Regularized Learning outperformed 4 variations of ICA (p<0.001) for predicting the task being performed within each scan using artifact-cleaned components. The NMF algorithms, which suppressed negative BOLD signal, had the poorest accuracy compared to the ICA and sparse coding algorithms. Holding constant the effect of the extraction algorithm, encodings using sparser spatial networks (containing more zero-valued voxels) had higher classification accuracy (p<0.001). Lower classification accuracy occurred when the extracted spatial maps contained more CSF regions (p<0.001). CONCLUSION: The success of sparse coding algorithms suggests that algorithms which enforce sparsity, discourage multitasking, and promote local specialization may capture better the underlying source processes than those which allow inexhaustible local processes such as ICA. Negative BOLD signal may capture task-related activations.

Towards a Neuronal Gauge Theory.

Given the amount of knowledge and data accruing in the neurosciences, is it time to formulate a general principle for neuronal dynamics that holds at evolutionary, developmental, and perceptual timescales? In this paper, we propose that the brain (and other self-organised biological systems) can be characterised via the mathematical apparatus of a gauge theory. The picture that emerges from this approach suggests that any biological system (from a neuron to an organism) can be cast as resolving uncertainty about its external milieu, either by changing its internal states or its relationship to the environment. Using formal arguments, we show that a gauge theory for neuronal dynamics--based on approximate Bayesian inference--has the potential to shed new light on phenomena that have thus far eluded a formal description, such as attention and the link between action and perception.

Dynamic causal modelling of electrographic seizure activity using Bayesian belief updating.

Seizure activity in EEG recordings can persist for hours with seizure dynamics changing rapidly over time and space. To characterise the spatiotemporal evolution of seizure activity, large data sets often need to be analysed. Dynamic causal modelling (DCM) can be used to estimate the synaptic drivers of cortical dynamics during a seizure; however, the requisite (Bayesian) inversion procedure is computationally expensive. In this note, we describe a straightforward procedure, within the DCM framework, that provides efficient inversion of seizure activity measured with non-invasive and invasive physiological recordings; namely, EEG/ECoG. We describe the theoretical background behind a Bayesian belief updating scheme for DCM. The scheme is tested on simulated and empirical seizure activity (recorded both invasively and non-invasively) and compared with standard Bayesian inversion. We show that the Bayesian belief updating scheme provides similar estimates of time-varying synaptic parameters, compared to standard schemes, indicating no significant qualitative change in accuracy. The difference in variance explained was small (less than 5%). The updating method was substantially more efficient, taking approximately 5-10min compared to approximately 1-2h. Moreover, the setup of the model under the updating scheme allows for a clear specification of how neuronal variables fluctuate over separable timescales. This method now allows us to investigate the effect of fast (neuronal) activity on slow fluctuations in (synaptic) parameters, paving a way forward to understand how seizure activity is generated.

Diffusion Tensor Imaging of TBI: Potentials and Challenges.

Neuroimaging plays a critical role in the setting in traumatic brain injury (TBI). Diffusion tensor imaging (DTI) is an advanced magnetic resonance imaging technique that is capable of providing rich information on the brain's neuroanatomic connectome. The purpose of this article is to systematically review the role of DTI and advanced diffusion techniques in the setting of TBI, including diffusion kurtosis imaging (DKI), neurite orientation dispersion and density imaging, diffusion spectrum imaging, and q-ball imaging. We discuss clinical applications of DTI and review the DTI literature as it pertains to TBI. Despite the continued advancements in DTI and related diffusion techniques over the past 20 years, DTI techniques are sensitive for TBI at the group level only and there is insufficient evidence that DTI plays a role at the individual level. We conclude by discussing future directions in DTI research in TBI including the role of machine learning in the pattern classification of TBI.

Method for simultaneous fMRI/EEG data collection during a focused attention suggestion for differential thermal sensation.

In the present work, we demonstrate a method for concurrent collection of EEG/fMRI data. In our setup, EEG data are collected using a high-density 256-channel sensor net. The EEG amplifier itself is contained in a field isolation containment system (FICS), and MRI clock signals are synchronized with EEG data collection for subsequent MR artifact characterization and removal. We demonstrate this method first for resting state data collection. Thereafter, we demonstrate a protocol for EEG/fMRI data recording, while subjects listen to a tape asking them to visualize that their left hand is immersed in a cold-water bath and referred to, here, as the cold glove paradigm. Thermal differentials between each hand are measured throughout EEG/fMRI data collection using an MR compatible temperature sensor that we developed for this purpose. We collect cold glove EEG/fMRI data along with simultaneous differential hand temperature measurements both before and after hypnotic induction. Between pre and post sessions, single modality EEG data are collected during the hypnotic induction and depth assessment process. Our representative results demonstrate that significant changes in the EEG power spectrum can be measured during hypnotic induction, and that hand temperature changes during the cold glove paradigm can be detected rapidly using our MR compatible differential thermometry device.

Non-negative matrix factorization of multimodal MRI, fMRI and phenotypic data reveals differential changes in default mode subnetworks in ADHD.

In the multimodal neuroimaging framework, data on a single subject are collected from inherently different sources such as functional MRI, structural MRI, behavioral and/or phenotypic information. The information each source provides is not independent; a subset of features from each modality maps to one or more common latent dimensions, which can be interpreted using generative models. These latent dimensions, or "topics," provide a sparse summary of the generative process behind the features for each individual. Topic modeling, an unsupervised generative model, has been used to map seemingly disparate features to a common domain. We use Non-Negative Matrix Factorization (NMF) to infer the latent structure of multimodal ADHD data containing fMRI, MRI, phenotypic and behavioral measurements. We compare four different NMF algorithms and find that the sparsest decomposition is also the most differentiating between ADHD and healthy patients. We identify dimensions that map to interpretable, recognizable dimensions such as motion, default mode network activity, and other such features of the input data. For example, structural and functional graph theory features related to default mode subnetworks clustered with the ADHD-Inattentive diagnosis. Structural measurements of the default mode network (DMN) regions such as the posterior cingulate, precuneus, and parahippocampal regions were all related to the ADHD-Inattentive diagnosis. Ventral DMN subnetworks may have more functional connections in ADHD-I, while dorsal DMN may have less. ADHD topics are dependent upon diagnostic site, suggesting diagnostic differences across geographic locations. We assess our findings in light of the ADHD-200 classification competition, and contrast our unsupervised, nominated topics with previously published supervised learning methods. Finally, we demonstrate the validity of these latent variables as biomarkers by using them for classification of ADHD in 730 patients. Cumulatively, this manuscript addresses how multimodal data in ADHD can be interpreted by latent dimensions.