Chun-Chuan Chen

Dynamic causal modeling for EEG and MEG

We present a review of dynamic causal modeling (DCM) for magneto‐ and electroencephalography (M/EEG) data. DCM is based on a spatiotemporal model, where the temporal component is formulated in terms of neurobiologically plausible dynamics. Following an intuitive description of the model, we discuss six recent studies, which use DCM to analyze M/EEG and local field potentials. These studies illustrate how DCM can be used to analyze evoked responses (average response in time), induced responses (average response in time‐frequency), and steady‐state responses (average response in frequency). Bayesian model comparison plays a critical role in these analyses, by allowing one to compare equally plausible models in terms of their model evidence. This approach might be very useful in M/EEG research; where correlations among spatial and neuronal model parameter estimates can cause uncertainty about which model best explains the data. Bayesian model comparison resolves these uncertainties in a principled and formal way. We suggest that DCM and Bayesian model comparison provides a useful way to test hypotheses about distributed processing in the brain, using electromagnetic data. Hum Brain Mapp, 2009.

Dynamic causal modelling of induced responses

This paper describes a dynamic causal model (DCM) for induced or spectral responses as measured with the electroencephalogram (EEG) or the magnetoencephalogram (MEG). We model the time-varying power, over a range of frequencies, as the response of a distributed system of coupled electromagnetic sources to a spectral perturbation. The model parameters encode the frequency response to exogenous input and coupling among sources and different frequencies. The Bayesian inversion of this model, given data enables inferences about the parameters of a particular model and allows us to compare different models, or hypotheses. One key aspect of the model is that it differentiates between linear and non-linear coupling; which correspond to within and between-frequency coupling respectively. To establish the face validity of our approach, we generate synthetic data and test the identifiability of various parameters to ensure they can be estimated accurately, under different levels of noise. We then apply our model to EEG data from a face-perception experiment, to ask whether there is evidence for non-linear coupling between early visual cortex and fusiform areas.

Forward and backward connections in the brain: A DCM study of functional asymmetries

In this paper, we provide evidence for functional asymmetries in forward and backward connections that define hierarchical architectures in the brain. We exploit the fact that modulatory or nonlinear influences of one neuronal system on another (i.e., effective connectivity) entail coupling between different frequencies. Functional asymmetry in forward and backward connections was addressed by comparing dynamic causal models of MEG responses induced by visual processing of normal and scrambled faces. We compared models with and without nonlinear (between-frequency) coupling in both forward and backward connections. Bayesian model comparison indicated that the best model had nonlinear forward and backward connections. Using the best model we then quantified frequency-specific causal influences mediating observed spectral responses. We found a striking asymmetry between forward and backward connections; in which high (gamma) frequencies in higher cortical areas suppressed low (alpha) frequencies in lower areas. This suppression was significantly greater than the homologous coupling in the forward connections. Furthermore, exactly the asymmetry was observed when we examined face-selective coupling (i.e., coupling under faces minus scrambled faces). These results highlight the importance of nonlinear coupling among brain regions and point to a functional asymmetry between forward and backward connections in the human brain that is consistent with anatomical and physiological evidence from animal studies. This asymmetry is also consistent with functional architectures implied by theories of perceptual inference in the brain, based on hierarchical generative models.

Nonlinear Coupling in the Human Motor System

The synchronous discharge of neuronal assemblies is thought to facilitate communication between areas within distributed networks in the human brain. This oscillatory activity is especially interesting, given the pathological modulation of specific frequencies in diseases affecting the motor system. Many studies investigating oscillatory activity have focused on same frequency, or linear, coupling between areas of a network. In this study, our aim was to establish a functional architecture in the human motor system responsible for induced responses as measured in normal subjects with magnetoencephalography. Specifically, we looked for evidence for additional nonlinear (between-frequency) coupling among neuronal sources and, in particular, whether nonlinearities were found predominantly in connections within areas (intrinsic), between areas (extrinsic) or both. We modeled the event-related modulation of spectral responses during a simple hand-grip using dynamic casual modeling. We compared models with and without nonlinear connections under conditions of symmetric and asymmetric interhemispheric connectivity. Bayesian model comparison suggested that the task-dependent motor network was asymmetric during right hand movements. Furthermore, it revealed very strong evidence for nonlinear coupling between sources in this distributed network, but interactions among frequencies within a source appeared linear in nature. Our results provide empirical evidence for nonlinear coupling among distributed neuronal sources in the motor system and that these play an important role in modulating spectral responses under normal conditions.

A dynamic causal model for evoked and induced responses

Neuronal responses exhibit two stimulus or task-related components: evoked and induced. The functional role of induced responses has been ascribed to ‘top-down’ modulation through backward connections and lateral interactions; as opposed to the bottom-up driving processes that may predominate in evoked components. The implication is that evoked and induced components may reflect different neuronal processes. The conventional way of separating evoked and induced responses assumes that they can be decomposed linearly; in that induced responses are the average of the power minus the power of the average (the evoked component). However, this decomposition may not hold if both components are generated by nonlinear processes. In this work, we propose a Dynamic Causal Model that models evoked and induced responses at the same time. This allows us to explain both components in terms of shared mechanisms (coupling) and changes in coupling that are necessary to explain any induced components. To establish the face validity of our approach, we used Bayesian Model Selection to show that the scheme can disambiguate between models of synthetic data that did and did not contain induced components. We then repeated the analysis using MEG data during a hand grip task to ask whether induced responses in motor control circuits are mediated by ‘top-down’ or backward connections. Our result provides empirical evidence that induced responses are more likely to reflect backward message passing in the brain, while evoked and induced components share certain characteristics and mechanisms.

Neuronal correlates of a virtual-reality-based passive sensory P300 network.

P300, a positive event-related potential (ERP) evoked at around 300 ms after stimulus, can be elicited using an active or passive oddball paradigm. Active P300 requires a person’s intentional response, whereas passive P300 does not require an intentional response. Passive P300 has been used in incommunicative patients for consciousness detection and brain computer interface. Active and passive P300 differ in amplitude, but not in latency or scalp distribution. However, no study has addressed the mechanism underlying the production of passive P300. In particular, it remains unclear whether the passive P300 shares an identical active P300 generating network architecture when no response is required. This study aims to explore the hierarchical network of passive sensory P300 production using dynamic causal modelling (DCM) for ERP and a novel virtual reality (VR)-based passive oddball paradigm. Moreover, we investigated the causal relationship of this passive P300 network and the changes in connection strength to address the possible functional roles. A classical ERP analysis was performed to verify that the proposed VR-based game can reliably elicit passive P300. The DCM results suggested that the passive and active P300 share the same parietal-frontal neural network for attentional control and, underlying the passive network, the feed-forward modulation is stronger than the feed-back one. The functional role of this forward modulation may indicate the delivery of sensory information, automatic detection of differences, and stimulus-driven attentional processes involved in performing this passive task. To our best knowledge, this is the first study to address the passive P300 network. The results of this study may provide a reference for future clinical studies on addressing the network alternations under pathological states of incommunicative patients. However, caution is required when comparing patients’ analytic results with this study. For example, the task presented here is not applicable to incommunicative patients.

Accelerating Computation of DCM for ERP in MATLAB by External Function Calls to the GPU

This study aims to improve the performance of Dynamic Causal Modelling for Event Related Potentials (DCM for ERP) in MATLAB by using external function calls to a graphics processing unit (GPU). DCM for ERP is an advanced method for studying neuronal effective connectivity. DCM utilizes an iterative procedure, the expectation maximization (EM) algorithm, to find the optimal parameters given a set of observations and the underlying probability model. As the EM algorithm is computationally demanding and the analysis faces possible combinatorial explosion of models to be tested, we propose a parallel computing scheme using the GPU to achieve a fast estimation of DCM for ERP. The computation of DCM for ERP is dynamically partitioned and distributed to threads for parallel processing, according to the DCM model complexity and the hardware constraints. The performance efficiency of this hardware-dependent thread arrangement strategy was evaluated using the synthetic data. The experimental data were used to validate the accuracy of the proposed computing scheme and quantify the time saving in practice. The simulation results show that the proposed scheme can accelerate the computation by a factor of 155 for the parallel part. For experimental data, the speedup factor is about 7 per model on average, depending on the model complexity and the data. This GPU-based implementation of DCM for ERP gives qualitatively the same results as the original MATLAB implementation does at the group level analysis. In conclusion, we believe that the proposed GPU-based implementation is very useful for users as a fast screen tool to select the most likely model and may provide implementation guidance for possible future clinical applications such as online diagnosis.

EEG-based motor network biomarkers for identifying target patients with stroke for upper limb rehabilitation and its construct validity

Rehabilitation is the main therapeutic approach for reducing poststroke functional deficits in the affected upper limb; however, significant between-patient variability in rehabilitation efficacy indicates the need to target patients who are likely to have clinically significant improvement after treatment. Many studies have determined robust predictors of recovery and treatment gains and yielded many great results using linear approachs. Evidence has emerged that the nonlinearity is a crucial aspect to study the inter-areal communication in human brains and abnormality of oscillatory activities in the motor system is linked to the pathological states. In this study, we hypothesized that combinations of linear and nonlinear (cross-frequency) network connectivity parameters are favourable biomarkers for stratifying patients for upper limb rehabilitation with increased accuracy. We identified the biomarkers by using 37 prerehabilitation electroencephalogram (EEG) datasets during a movement task through effective connectivity and logistic regression analyses. The predictive power of these biomarkers was then tested by using 16 independent datasets (i.e. construct validation). In addition, 14 right handed healthy subjects were also enrolled for comparisons. The result shows that the beta plus gamma or theta network features provided the best classification accuracy of 92%. The predictive value and the sensitivity of these biomarkers were 81.3% and 90.9%, respectively. Subcortical lesion, the time poststroke and initial Wolf Motor Function Test (WMFT) score were identified as the most significant clinical variables affecting the classification accuracy of this predictive model. Moreover, 12 of 14 normal controls were classified as having favourable recovery. In conclusion, EEG-based linear and nonlinear motor network biomarkers are robust and can help clinical decision making.

The changes of improvement-related motor kinetics after virtual reality based rehabilitation

Stroke is a major cause of adult disability and up to two thirds of stroke survivors have left with motor deficits of the affected upper limb (UL). Rehabilitation is the main therapeutic approach for reducing poststroke functional deficits but significant variability exists between patients regarding rehabilitation efficacy. Because the recovery mechanisms induced by rehabilitation are not fully understood, it is still not clear which treatment approaches and techniques are most beneficial. Virtual reality (VR) is a computer-based environment that provides the users an immersive experience of a synthetic world and allows a systemic testing of human functions under the simulated environment of controllable parameters. A large body of evidence has demonstrated the efficacy of VR based rehabilitation for improving the upper limb functional recovery. However, the question which components in the VR game are most valuable for promoting the recovery has not been examined. This study aims to identify the motion kinetics extracted from the VR based rehabilitation that are significantly correlated with the functional improvement. Twenty-one stroke patients were recruited and received 1 hour rehabilitation with the intensity of 3 times per week over 8 weeks, in total 24 training sessions, using a home-made VR program. All patients underwent the evaluation of their function before and after rehabilitation by Fugl-Meyer assessment (FMA). The parameters of human kinetics, such as speed/max speed, velocity and trajectory during VR-based rehabilitation were recorded digitally and the changes of the kinetics after finishing 24 training sessions were then correlated with the post-rehabilitation changes of FMA. The statistic result showed that the increase of efficiency, speed stability and straightness of trajectory leads to a better functional improvement, in addition to the increase of palm strength. In conclusion, the changes of the improvement-related motor kinetics could serve as guidance when designing the individualized treatment strategy.

Accelerating Computation of DCM for ERP with GPU-Based Parallel Strategy

This paper presents the use of graphic processing unit (GPU) to accelerate a brain-activity analytical tool, the Dynamic Causal Modelling for Event Related Potential (DCM for ERP) in MATLAB. DCM for ERP is a recently developed advanced method for studying neuronal effective connectivity and making inference about the brain functions. DCM utilizes an iterative procedure, the expectation maximization (EM) algorithm, to find the optimal parameters given a set of observed events (data) and the underlying probability model, such that the likelihood function is maximized. As the EM algorithm is computationally demanding, time consuming and largely data dependent, we propose a parallel computing scheme using GPUs to achieve a fast estimation of neural effective connectivity in DCM. The computational loading of EM was partitioned and dynamically distributed to either the threads or blocks according to the DCM model complex (i.e. the number of parameters to be estimated). The performance of this dynamic loading arrangement in terms of execution time and accuracy loss were evaluated using synthetic data. The results show that our method can accelerate a computation task by about 30 times as fast as the MATLAB version.