Okito Yamashita

Visual Image Reconstruction from Human Brain Activity using a Combination of Multiscale Local Image Decoders

Perceptual experience consists of an enormous number of possible states. Previous fMRI studies have predicted a perceptual state by classifying brain activity into prespecified categories. Constraint-free visual image reconstruction is more challenging, as it is impractical to specify brain activity for all possible images. In this study, we reconstructed visual images by combining local image bases of multiple scales, whose contrasts were independently decoded from fMRI activity by automatically selecting relevant voxels and exploiting their correlated patterns. Binary-contrast, 10 x 10-patch images (2(100) possible states) were accurately reconstructed without any image prior on a single trial or volume basis by measuring brain activity only for several hundred random images. Reconstruction was also used to identify the presented image among millions of candidates. The results suggest that our approach provides an effective means to read out complex perceptual states from brain activity while discovering information representation in multivoxel patterns.

Sparse estimation automatically selects voxels relevant for the decoding of fMRI activity patterns

Recent studies have used pattern classification algorithms to predict or decode task parameters from individual fMRI activity patterns. For fMRI decoding, it is important to choose an appropriate set of voxels (or features) as inputs to the decoder, since the presence of many irrelevant voxels could lead to poor generalization performance, a problem known as overfitting. Although individual voxels could be chosen based on univariate statistics, the resulting set of voxels could be suboptimal if correlations among voxels carry important information. Here, we propose a novel linear classification algorithm, called sparse logistic regression (SLR), that automatically selects relevant voxels while estimating their weight parameters for classification. Using simulation data, we confirmed that SLR can automatically remove irrelevant voxels and thereby attain higher classification performance than other methods in the presence of many irrelevant voxels. SLR also proved effective with real fMRI data obtained from two visual experiments, successfully identifying voxels in corresponding locations of visual cortex. SLR-selected voxels often led to better performance than those selected based on univariate statistics, by exploiting correlated noise among voxels to allow for better pattern separation. We conclude that SLR provides a robust method for fMRI decoding and can also serve as a stand-alone tool for voxel selection.

A solution to the dynamical inverse problem of EEG generation using spatiotemporal Kalman filtering

We present a new approach for estimating solutions of the dynamical inverse problem of EEG generation. In contrast to previous approaches, we reinterpret this problem as a filtering problem in a state space framework; for the purpose of its solution, we propose a new extension of Kalman filtering to the case of spatiotemporal dynamics. The temporal evolution of the distributed generators of the EEG can be reconstructed at each voxel of a discretisation of the gray matter of brain. By fitting linear autoregressive models with neighbourhood interactions to EEG time series, new classes of inverse solutions with improved resolution and localisation ability can be explored. For the purposes of model comparison and parameter estimation from given data, we employ a likelihood maximisation approach. Both for instantaneous and dynamical inverse solutions, we derive estimators of the time-dependent estimation error at each voxel. The performance of the algorithm is demonstrated by application to simulated and clinical EEG recordings. It is shown that by choosing appropriate dynamical models, it becomes possible to obtain inverse solutions of considerably improved quality, as compared to the usual instantaneous inverse solutions.

Recursive penalized least squares solution for dynamical inverse problems of EEG generation

In the dynamical inverse problem of electroencephalogram (EEG) generation where a specific dynamics for the electrical current distribution is assumed, we can impose general spatiotemporal constraints onto the solution by casting the problem into a state space representation and assuming a specific class of parametric models for the dynamics. The Akaike Bayesian Information Criterion (ABIC), which is based on the Type II likelihood, was used to estimate the parameters and evaluate the model. In addition, dynamic low‐resolution brain electromagnetic tomography (LORETA), a new approach for estimating the current distribution is introduced. A recursive penalized least squares (RPLS) step forms the main element of our implementation. To obtain improved inverse solutions, dynamic LORETA exploits both spatial and temporal information, whereas LORETA uses only spatial information. A considerable improvement in performance compared to LORETA was found when dynamic LORETA was applied to simulated EEG data, and the new method was applied also to clinical EEG data. Hum. Brain Mapp. 21:221–235, 2004.

Regulation of Motor Representation by Phase–Amplitude Coupling in the Sensorimotor Cortex

High-γ amplitude (80–150 Hz) represents motor information, such as movement types, on the sensorimotor cortex. In several cortical areas, high-γ amplitudes are coupled with low-frequency phases, e.g., α and θ (phase–amplitude coupling, PAC). However, such coupling has not been studied in the sensorimotor cortex; thus, its potential functional role has yet to be explored. We investigated PAC of high-γ amplitude in the sensorimotor cortex during waiting for and the execution of movements using electrocorticographic (ECoG) recordings in humans. ECoG signals were recorded from the sensorimotor cortices of 4 epilepsy patients while they performed three different hand movements. A subset of electrodes showed high-γ activity selective to movement type around the timing of motor execution, while the same electrodes showed nonselective high-γ activity during the waiting period (>2 s before execution). Cross frequency coupling analysis revealed that the high-γ amplitude during waiting was strongly coupled with the α phase (10–14 Hz) at the electrodes with movement-selective high-γ amplitudes during execution. This coupling constituted the high-γ amplitude peaking around the trough of the α oscillation, and its strength and phase were not predictive of movement type. As the coupling attenuated toward the timing of motor execution, the high-γ amplitude appeared to be released from the α phase to build a motor representation with phase-independent activity. Our results suggest that PAC modulates motor representation in the sensorimotor cortex by holding and releasing high-γ activity in movement-selective cortical regions.

Evaluation of hierarchical Bayesian method through retinotopic brain activities reconstruction from fMRI and MEG signals

A hierarchical Bayesian method estimated current sources from MEG data, incorporating an fMRI constraint as a hierarchical prior whose strength is controlled by hyperparameters. A previous study [Sato, M., Yoshioka, T., Kajihara, S., Toyama, K., Goda, N., Doya, K., Kawato, M., 2004. Hierarchical Bayesian estimation for MEG inverse problem. Neuroimage 23, 806-826] demonstrated that fMRI information improves the localization accuracy for simulated data. The goal of the present study is to confirm the usefulness of the hierarchical Bayesian method by the real MEG and fMRI experiments using visual stimuli with a fan-shaped checkerboard pattern presented in four visual quadrants. The proper range of hyperparameters was systematically analyzed using goodness of estimate measures for the estimated currents. The robustness with respect to false-positive activities in the fMRI information was also evaluated by using noisy priors constructed by adding artificial noises to real fMRI signals. It was shown that with appropriate hyperparameter values, the retinotopic organization and temporal dynamics in the early visual area were reconstructed, which were in a close correspondence with the known brain imaging and electrophysiology of the humans and monkeys. The false-positive effects of the noisy priors were suppressed by using appropriate hyperparameter values. The hierarchical Bayesian method also was capable of reconstructing retinotopic sequential activation in V1 with fine spatiotemporal resolution, from MEG data elicited by sequential stimulation of the four visual quadrants with the fan-shaped checker board pattern at much shorter intervals (150 and 400 ms) than the temporal resolution of fMRI. These results indicate the potential capability for the hierarchical Bayesian method combining MEG with fMRI to improve the spatiotemporal resolution of noninvasive brain activity measurement.

fMRI activation maps based on the NN-ARx model

The most significant progresses in the understanding of human brain functions have been possible due to the use of functional magnetic resonance imaging (fMRI), which when used in combination with other standard neuroimaging techniques (i.e., EEG) provides researchers with a potential tool to elucidate many biophysical principles, established previously by animal comparative studies. However, to date, most of the methods proposed in the literature seeking fMRI signs have been limited to the use of a top-down data analysis approach, thus ignoring a pool of physiological facts. In spite of the important contributions achieved by applying these methods to actual data, there is a disproportionate gap between theoretical models and data-analysis strategies while trying to focus on several new prospects, like for example fMRI/EEG data fusion, causality/connectivity patterns, and nonlinear BOLD signal dynamics. In this paper, we propose a new approach which will allow many of the abovementioned hot topics to be addressed in the near future with an underlying interpretability based on bottom-up modeling. In particular, the theta-MAP presented in the paper to test brain activation corresponds very well with the standardized t test of the SPM99 toolbox. Additionally, a new Impulse Response Function (IRF) has been formulated, directly related to the well-established concept of the hemodynamics response function (HRF). The model uses not only the information contained in the signal but also that in the structure of the background noise to simultaneously estimate the IRF and the autocorrelation function (ACF) by using an autoregressive (AR) model with a filtered Poisson process driving the dynamics. The short-range contributions of voxels within the near-neighborhood are also included, and the potential drift was characterized by a polynomial series. Since our model originated from an immediate extension of the hemodynamics approach [Friston, K.J., Mechelli, A., Turner, R., Price C.J. (2000a). Nonlinear responses in fMRI: the balloon model, volterra kernels, and other hemodynamics. NeuroImage 12, 466-477.], a natural interpretability of the results is feasible.

Evaluating frequency-wise directed connectivity of BOLD signals applying relative power contribution with the linear multivariate time-series models

In this article, we propose a statistical method to evaluate directed interactions of functional magnetic-resonance imaging (fMRI) data. The multivariate autoregressive (MAR) model was combined with the relative power contribution (RPC) in this analysis. The MAR model was fitted to the data to specify the direction of connections, and the RPC quantifies the strength of connections. As the RPC is computed in the frequency domain, we can evaluate the connectivity for each frequency component. From this, we can establish whether the specified connections represent low- or high-frequency connectivity, which cannot be examined solely using the estimated MAR coefficients. We applied this analysis method to fMRI data obtained during visual motion tasks, confirming previous reports of bottom-up connectivity around the frequency corresponding to the block experimental design. Furthermore, we used the MAR model with exogenous variables (MARX) to extend our understanding of these data, and to show how the input to V1 transfers to higher cortical areas.

Hierarchical Bayesian estimation improves depth accuracy and spatial resolution of diffuse optical tomography

High-density diffuse optical tomography (HD-DOT) is an emerging technique for visualizing the internal state of biological tissues. The large number of overlapping measurement channels due to the use of high-density probe arrays permits the reconstruction of the internal optical properties, even with a reflectance-only measurement. However, accurate three-dimensional reconstruction is still a challenging problem. First, the exponentially decaying sensitivity causes a systematic depth-localization error. Second, the nature of diffusive light makes the image blurred. In this paper, we propose a three-dimensional reconstruction method that overcomes these two problems by introducing sensitivity-normalized regularization and sparsity into the hierarchical Bayesian method. Phantom experiments were performed to validate the proposed method under three conditions of probe interval: 26 mm, 18.4 mm, and 13 mm. We found that two absorbers with distances shorter than the probe interval could be discriminated under the high-density conditions of 18.4-mm and 13-mm intervals. This discrimination ability was possible even if the depths of the two absorbers were different from each other. These results show the high spatial resolution of the proposed method in both depth and horizontal directions.

A State-Space Modeling Approach for Localization of Focal Current Sources From MEG

State-space modeling is a promising approach for current source reconstruction from magnetoencephalography (MEG) because it constrains the spatiotemporal behavior of inverse solutions in a flexible manner. However, state-space model-based source localization research remains underdeveloped; extraction of spatially focal current sources and handling of the high dimensionality of the distributed source model remain problematic. In this study, we propose a novel state-space model-based method that resolves these problems, extending our previous source localization method to include a temporal constraint by state-space modeling. To enable focal current reconstruction, we account for spatially inhomogeneous temporal dynamics by introducing dynamics model parameters that differ for each cortical position. The model parameters and the intensity of the current sources are jointly estimated according to a Bayesian framework. We circumvent the high dimensionality of the problem by assuming prior distributions of the model parameters to reduce the sensitivity to unmodeled components, and by adopting variational Bayesian inference to reduce the computational cost. Through simulation experiments and application to real MEG data, we have confirmed that our proposed method successfully reconstructs focal current activities, which evolve with their temporal dynamics.