Recovery of neural dynamics criticality in personalized whole-brain models of stroke.

The critical brain hypothesis states that biological neuronal networks, because of their structural and functional architecture, work near phase transitions for optimal response to internal and external inputs. Criticality thus provides optimal function and behavioral capabilities. We test this hypothesis by examining the influence of brain injury (strokes) on the criticality of neural dynamics estimated at the level of single participants using directly measured individual structural connectomes and whole-brain models. Lesions engender a sub-critical state that recovers over time in parallel with behavior. The improvement of criticality is associated with the re-modeling of specific white-matter connections. We show that personalized whole-brain dynamical models poised at criticality track neural dynamics, alteration post-stroke, and behavior at the level of single participants.

A novel stroke lesion network mapping approach: improved accuracy yet still low deficit prediction.

Lesion network mapping estimates functional network abnormalities caused by a focal brain lesion. The method requires embedding the volume of the lesion into a normative functional connectome and using the average functional magnetic resonance imaging signal from that volume to compute the temporal correlation with all other brain locations. Lesion network mapping yields a map of potentially functionally disconnected regions. Although promising, this approach does not predict behavioural deficits well. We modified lesion network mapping by using the first principal component of the functional magnetic resonance imaging signal computed from the voxels within the lesioned area for temporal correlation. We measured potential improvements in connectivity strength, anatomical specificity of the lesioned network and behavioural prediction in a large cohort of first-time stroke patients at 2-weeks post-injury (n = 123). This principal component functional disconnection approach localized mainly cortical voxels of high signal-to-noise; and it yielded networks with higher anatomical specificity, and stronger behavioural correlation than the standard method. However, when examined with a rigorous leave-one-out machine learning approach, principal component functional disconnection approach did not perform better than the standard lesion network mapping in predicting neurological deficits. In summary, even though our novel method improves the specificity of disconnected networks and correlates with behavioural deficits post-stroke, it does not improve clinical prediction. Further work is needed to capture the complex adjustment of functional networks produced by focal damage in relation to behaviour.

Sensorimotor, Attentional, and Neuroanatomical Predictors of Upper Limb Motor Deficits and Rehabilitation Outcome after Stroke.

The rehabilitation of motor deficits following stroke relies on both sensorimotor and cognitive abilities, thereby involving large-scale brain networks. However, few studies have investigated the integration between motor and cognitive domains, as well as its neuroanatomical basis. In this retrospective study, upper limb motor responsiveness to technology-based rehabilitation was examined in a sample of 29 stroke patients (18 with right and 11 with left brain damage). Pretreatment sensorimotor and attentional abilities were found to influence motor recovery. Training responsiveness increased as a function of the severity of motor deficits, whereas spared attentional abilities, especially visuospatial attention, supported motor improvements. Neuroanatomical analysis of structural lesions and white matter disconnections showed that the poststroke motor performance was associated with putamen, insula, corticospinal tract, and frontoparietal connectivity. Motor rehabilitation outcome was mainly associated with the superior longitudinal fasciculus and partial involvement of the corpus callosum. The latter findings support the hypothesis that motor recovery engages large-scale brain networks that involve cognitive abilities and provides insight into stroke rehabilitation strategies.

A comparison of feature extraction methods for prediction of neuropsychological scores from functional connectivity data of stroke patients.

Multivariate prediction of human behavior from resting state data is gaining increasing popularity in the neuroimaging community, with far-reaching translational implications in neurology and psychiatry. However, the high dimensionality of neuroimaging data increases the risk of overfitting, calling for the use of dimensionality reduction methods to build robust predictive models. In this work, we assess the ability of four well-known dimensionality reduction techniques to extract relevant features from resting state functional connectivity matrices of stroke patients, which are then used to build a predictive model of the associated deficits based on cross-validated regularized regression. In particular, we investigated the prediction ability over different neuropsychological scores referring to language, verbal memory, and spatial memory domains. Principal Component Analysis (PCA) and Independent Component Analysis (ICA) were the two best methods at extracting representative features, followed by Dictionary Learning (DL) and Non-Negative Matrix Factorization (NNMF). Consistent with these results, features extracted by PCA and ICA were found to be the best predictors of the neuropsychological scores across all the considered cognitive domains. For each feature extraction method, we also examined the impact of the regularization method, model complexity (in terms of number of features that entered in the model) and quality of the maps that display predictive edges in the resting state networks. We conclude that PCA-based models, especially when combined with L1 (LASSO) regularization, provide optimal balance between prediction accuracy, model complexity, and interpretability.

Post-stroke deficit prediction from lesion and indirect structural and functional disconnection.

Behavioural deficits in stroke reflect both structural damage at the site of injury, and widespread network dysfunction caused by structural, functional, and metabolic disconnection. Two recent methods allow for the estimation of structural and functional disconnection from clinical structural imaging. This is achieved by embedding a patient's lesion into an atlas of functional and structural connections in healthy subjects, and deriving the ensemble of structural and functional connections that pass through the lesion, thus indirectly estimating its impact on the whole brain connectome. This indirect assessment of network dysfunction is more readily available than direct measures of functional and structural connectivity obtained with functional and diffusion MRI, respectively, and it is in theory applicable to a wide variety of disorders. To validate the clinical relevance of these methods, we quantified the prediction of behavioural deficits in a prospective cohort of 132 first-time stroke patients studied at 2 weeks post-injury (mean age 52.8 years, range 22-77; 63 females; 64 right hemispheres). Specifically, we used multivariate ridge regression to relate deficits in multiple functional domains (left and right visual, left and right motor, language, spatial attention, spatial and verbal memory) with the pattern of lesion and indirect structural or functional disconnection. In a subgroup of patients, we also measured direct alterations of functional connectivity with resting-state functional MRI. Both lesion and indirect structural disconnection maps were predictive of behavioural impairment in all domains (0.16 < R2 < 0.58) except for verbal memory (0.05 < R2 < 0.06). Prediction from indirect functional disconnection was scarce or negligible (0.01 < R2 < 0.18) except for the right visual field deficits (R2 = 0.38), even though multivariate maps were anatomically plausible in all domains. Prediction from direct measures of functional MRI functional connectivity in a subset of patients was clearly superior to indirect functional disconnection. In conclusion, the indirect estimation of structural connectivity damage successfully predicted behavioural deficits post-stroke to a level comparable to lesion information. However, indirect estimation of functional disconnection did not predict behavioural deficits, nor was a substitute for direct functional connectivity measurements, especially for cognitive disorders.

A Comparison of Shallow and Deep Learning Methods for Predicting Cognitive Performance of Stroke Patients From MRI Lesion Images.

Stroke causes behavioral deficits in multiple cognitive domains and there is a growing interest in predicting patient performance from neuroimaging data using machine learning techniques. Here, we investigated a deep learning approach based on convolutional neural networks (CNNs) for predicting the severity of language disorder from 3D lesion images from magnetic resonance imaging (MRI) in a heterogeneous sample of stroke patients. CNN performance was compared to that of conventional (shallow) machine learning methods, including ridge regression (RR) on the images' principal components and support vector regression. We also devised a hybrid method based on re-using CNN's high-level features as additional input to the RR model. Predictive accuracy of the four different methods was further investigated in relation to the size of the training set and the level of redundancy across lesion images in the dataset, which was evaluated in terms of location and topological properties of the lesions. The Hybrid model achieved the best performance in most cases, thereby suggesting that the high-level features extracted by CNNs are complementary to principal component analysis features and improve the model's predictive accuracy. Moreover, our analyses indicate that both the size of training data and image redundancy are critical factors in determining the accuracy of a computational model in predicting behavioral outcome from the structural brain imaging data of stroke patients.

The Role of Architectural and Learning Constraints in Neural Network Models: A Case Study on Visual Space Coding.

The recent "deep learning revolution" in artificial neural networks had strong impact and widespread deployment for engineering applications, but the use of deep learning for neurocomputational modeling has been so far limited. In this article we argue that unsupervised deep learning represents an important step forward for improving neurocomputational models of perception and cognition, because it emphasizes the role of generative learning as opposed to discriminative (supervised) learning. As a case study, we present a series of simulations investigating the emergence of neural coding of visual space for sensorimotor transformations. We compare different network architectures commonly used as building blocks for unsupervised deep learning by systematically testing the type of receptive fields and gain modulation developed by the hidden neurons. In particular, we compare Restricted Boltzmann Machines (RBMs), which are stochastic, generative networks with bidirectional connections trained using contrastive divergence, with autoencoders, which are deterministic networks trained using error backpropagation. For both learning architectures we also explore the role of sparse coding, which has been identified as a fundamental principle of neural computation. The unsupervised models are then compared with supervised, feed-forward networks that learn an explicit mapping between different spatial reference frames. Our simulations show that both architectural and learning constraints strongly influenced the emergent coding of visual space in terms of distribution of tuning functions at the level of single neurons. Unsupervised models, and particularly RBMs, were found to more closely adhere to neurophysiological data from single-cell recordings in the primate parietal cortex. These results provide new insights into how basic properties of artificial neural networks might be relevant for modeling neural information processing in biological systems.

A new adaptive videogame for training attention and executive functions: design principles and initial validation.

A growing body of evidence suggests that action videogames could enhance a variety of cognitive skills and more specifically attention skills. The aim of this study was to develop a novel adaptive videogame to support the rehabilitation of the most common consequences of traumatic brain injury (TBI), that is the impairment of attention and executive functions. TBI patients can be affected by psychomotor slowness and by difficulties in dealing with distraction, maintain a cognitive set for a long time, processing different simultaneously presented stimuli, and planning purposeful behavior. Accordingly, we designed a videogame that was specifically conceived to activate those functions. Playing involves visuospatial planning and selective attention, active maintenance of the cognitive set representing the goal, and error monitoring. Moreover, different game trials require to alternate between two tasks (i.e., task switching) or to perform the two tasks simultaneously (i.e., divided attention/dual-tasking). The videogame is controlled by a multidimensional adaptive algorithm that calibrates task difficulty on-line based on a model of user performance that is updated on a trial-by-trial basis. We report simulations of user performance designed to test the adaptive game as well as a validation study with healthy participants engaged in a training protocol. The results confirmed the involvement of the cognitive abilities that the game is supposed to enhance and suggested that training improved attentional control during play.

Deep Unsupervised Learning on a Desktop PC: A Primer for Cognitive Scientists.

Deep belief networks hold great promise for the simulation of human cognition because they show how structured and abstract representations may emerge from probabilistic unsupervised learning. These networks build a hierarchy of progressively more complex distributed representations of the sensory data by fitting a hierarchical generative model. However, learning in deep networks typically requires big datasets and it can involve millions of connection weights, which implies that simulations on standard computers are unfeasible. Developing realistic, medium-to-large-scale learning models of cognition would therefore seem to require expertise in programing parallel-computing hardware, and this might explain why the use of this promising approach is still largely confined to the machine learning community. Here we show how simulations of deep unsupervised learning can be easily performed on a desktop PC by exploiting the processors of low cost graphic cards (graphic processor units) without any specific programing effort, thanks to the use of high-level programming routines (available in MATLAB or Python). We also show that even an entry-level graphic card can outperform a small high-performance computing cluster in terms of learning time and with no loss of learning quality. We therefore conclude that graphic card implementations pave the way for a widespread use of deep learning among cognitive scientists for modeling cognition and behavior.

Space coding for sensorimotor transformations can emerge through unsupervised learning.

The posterior parietal cortex (PPC) is fundamental for sensorimotor transformations because it combines multiple sensory inputs and posture signals into different spatial reference frames that drive motor programming. Here, we present a computational model mimicking the sensorimotor transformations occurring in the PPC. A recurrent neural network with one layer of hidden neurons (restricted Boltzmann machine) learned a stochastic generative model of the sensory data without supervision. After the unsupervised learning phase, the activity of the hidden neurons was used to compute a motor program (a population code on a bidimensional map) through a simple linear projection and delta rule learning. The average motor error, calculated as the difference between the expected and the computed output, was less than 3°. Importantly, analyses of the hidden neurons revealed gain-modulated visual receptive fields, thereby showing that space coding for sensorimotor transformations similar to that observed in the PPC can emerge through unsupervised learning. These results suggest that gain modulation is an efficient coding strategy to integrate visual and postural information toward the generation of motor commands.