End-to-end deep learning approach to mouse behavior classification from cortex-wide calcium imaging.

Deep learning is a powerful tool for neural decoding, broadly applied to systems neuroscience and clinical studies. Interpretable and transparent models that can explain neural decoding for intended behaviors are crucial to identifying essential features of deep learning decoders in brain activity. In this study, we examine the performance of deep learning to classify mouse behavioral states from mesoscopic cortex-wide calcium imaging data. Our convolutional neural network (CNN)-based end-to-end decoder combined with recurrent neural network (RNN) classifies the behavioral states with high accuracy and robustness to individual differences on temporal scales of sub-seconds. Using the CNN-RNN decoder, we identify that the forelimb and hindlimb areas in the somatosensory cortex significantly contribute to behavioral classification. Our findings imply that the end-to-end approach has the potential to be an interpretable deep learning method with unbiased visualization of critical brain regions.

Sensor array design of optically pumped magnetometers for accurately estimating source currents.

An optically pumped magnetometer (OPM) is a new generation of magnetoencephalography (MEG) devices that is small, light, and works at room temperature. Due to these characteristics, OPMs enable flexible and wearable MEG systems. On the other hand, if we have a limited number of OPM sensors, we need to carefully design their sensor arrays depending on our purposes and regions of interests (ROIs). In this study, we propose a method that designs OPM sensor arrays for accurately estimating the cortical currents at the ROIs. Based on the resolution matrix of minimum norm estimate (MNE), our method sequentially determines the position of each sensor to optimize its inverse filter pointing to the ROIs and suppressing the signal leakage from the other areas. We call this method the Sensor array Optimization based on Resolution Matrix (SORM). We conducted simple and realistic simulation tests to evaluate its characteristics and efficacy for real OPM-MEG data. SORM designed the sensor arrays so that their leadfield matrices had high effective ranks as well as high sensitivities to ROIs. Although SORM is based on MNE, the sensor arrays designed by SORM were effective not only when we estimated the cortical currents by MNE but also when we did so by other methods. With real OPM-MEG data we confirmed its validity for real data. These analyses suggest that SORM is especially useful when we want to accurately estimate ROIs' activities with a limited number of OPM sensors, such as brain-machine interfaces and diagnosing brain diseases.

Generalizable brain network markers of major depressive disorder across multiple imaging sites.

Many studies have highlighted the difficulty inherent to the clinical application of fundamental neuroscience knowledge based on machine learning techniques. It is difficult to generalize machine learning brain markers to the data acquired from independent imaging sites, mainly due to large site differences in functional magnetic resonance imaging. We address the difficulty of finding a generalizable marker of major depressive disorder (MDD) that would distinguish patients from healthy controls based on resting-state functional connectivity patterns. For the discovery dataset with 713 participants from 4 imaging sites, we removed site differences using our recently developed harmonization method and developed a machine learning MDD classifier. The classifier achieved an approximately 70% generalization accuracy for an independent validation dataset with 521 participants from 5 different imaging sites. The successful generalization to a perfectly independent dataset acquired from multiple imaging sites is novel and ensures scientific reproducibility and clinical applicability.

Distinctive alterations in the mesocorticolimbic circuits in various psychiatric disorders.

AIM: Increasing evidence suggests that psychiatric disorders are linked to alterations in the mesocorticolimbic dopamine-related circuits. However, the common and disease-specific alterations remain to be examined in schizophrenia (SCZ), major depressive disorder (MDD), and autism spectrum disorder (ASD). Thus, this study aimed to examine common and disease-specific features related to mesocorticolimbic circuits. METHODS: This study included 555 participants from four institutes with five scanners: 140 individuals with SCZ (45.0% female), 127 individuals with MDD (44.9%), 119 individuals with ASD (15.1%), and 169 healthy controls (HC) (34.9%). All participants underwent resting-state functional magnetic resonance imaging. A parametric empirical Bayes approach was adopted to compare estimated effective connectivity among groups. Intrinsic effective connectivity focusing on the mesocorticolimbic dopamine-related circuits including the ventral tegmental area (VTA), shell and core parts of the nucleus accumbens (NAc), and medial prefrontal cortex (mPFC) were examined using a dynamic causal modeling analysis across these psychiatric disorders. RESULTS: The excitatory shell-to-core connectivity was greater in all patients than in the HC group. The inhibitory shell-to-VTA and shell-to-mPFC connectivities were greater in the ASD group than in the HC, MDD, and SCZ groups. Furthermore, the VTA-to-core and VTA-to-shell connectivities were excitatory in the ASD group, while those connections were inhibitory in the HC, MDD, and SCZ groups. CONCLUSION: Impaired signaling in the mesocorticolimbic dopamine-related circuits could be an underlying neuropathogenesis of various psychiatric disorders. These findings will improve the understanding of unique neural alternations of each disorder and will facilitate identification of effective therapeutic targets.

Predicting behavior through dynamic modes in resting-state fMRI data.

Dynamic properties of resting-state functional connectivity (FC) provide rich information on brain-behavior relationships. Dynamic mode decomposition (DMD) has been used as a method to characterize FC dynamics. However, it remains unclear whether dynamic modes (DMs), spatial-temporal coherent patterns computed by DMD, provide information about individual behavioral differences. This study established a methodological approach to predict individual differences in behavior using DMs. Furthermore, we investigated the contribution of DMs within each of seven specific frequency bands (0-0.1,...,0.6-0.7 Hz) for prediction. To validate our approach, we tested whether each of 59 behavioral measures could be predicted by performing multivariate pattern analysis on a Gram matrix, which was created using subject-specific DMs computed from resting-state functional magnetic resonance imaging (rs-fMRI) data of individuals. DMD successfully predicted behavior and outperformed temporal and spatial independent component analysis, which is the conventional data decomposition method for extracting spatial activity patterns. Most of the behavioral measures that were predicted with significant accuracy in a permutation test were related to cognition. We found that DMs within frequency bands <0.2 Hz primarily contributed to prediction and had spatial structures similar to several common resting-state networks. Our results indicate that DMD is efficient in extracting spatiotemporal features from rs-fMRI data.

Harmonization of resting-state functional MRI data across multiple imaging sites via the separation of site differences into sampling bias and measurement bias.

When collecting large amounts of neuroimaging data associated with psychiatric disorders, images must be acquired from multiple sites because of the limited capacity of a single site. However, site differences represent a barrier when acquiring multisite neuroimaging data. We utilized a traveling-subject dataset in conjunction with a multisite, multidisorder dataset to demonstrate that site differences are composed of biological sampling bias and engineering measurement bias. The effects on resting-state functional MRI connectivity based on pairwise correlations because of both bias types were greater than or equal to psychiatric disorder differences. Furthermore, our findings indicated that each site can sample only from a subpopulation of participants. This result suggests that it is essential to collect large amounts of neuroimaging data from as many sites as possible to appropriately estimate the distribution of the grand population. Finally, we developed a novel harmonization method that removed only the measurement bias by using a traveling-subject dataset and achieved the reduction of the measurement bias by 29% and improvement of the signal-to-noise ratios by 40%. Our results provide fundamental knowledge regarding site effects, which is important for future research using multisite, multidisorder resting-state functional MRI data.

MEG current source reconstruction using a meta-analysis fMRI prior.

Magnetoencephalography (MEG) offers a unique way to noninvasively investigate millisecond-order cortical activities by mapping sensor signals (magnetic fields outside the head) to cortical current sources using current source reconstruction methods. Current source reconstruction is defined as an ill-posed inverse problem, since the number of sensors is less than the number of current sources. One powerful approach to solving this problem is to use functional MRI (fMRI) data as a spatial constraint, although it boosts the cost of measurement and the burden on subjects. Here, we show how to use the meta-analysis fMRI data synthesized from thousands of papers instead of the individually recorded fMRI data. To mitigate the differences between the meta-analysis and individual data, the former are imported as prior information of the hierarchical Bayesian estimation. Using realistic simulations, we found out the performance of current source reconstruction using meta-analysis fMRI data to be better than that using low-quality individual fMRI data and conventional methods. By applying experimental data of a face recognition task, we qualitatively confirmed that group analysis results using the meta-analysis fMRI data showed a tendency similar to the results using the individual fMRI data. Our results indicate that the use of meta-analysis fMRI data improves current source reconstruction without additional measurement costs. We assume the proposed method would have greater effect for modalities with lower measurement costs, such as optically pumped magnetometers.

Whole-brain propagating patterns in human resting-state brain activities.

Repetitive propagating activities in resting-state brain activities have been widely observed in various species and regions. Because they resemble the preceding brain activities during tasks, they are assumed to reflect past experiences embedded in neuronal circuits. "Whole-brain" propagating activities may also reflect a process that integrates information distributed over the entire brain, such as visual and motor information. Here we reveal whole-brain propagating activities from human resting-state magnetoencephalography (MEG) and electroencephalography (EEG) data. We simultaneously recorded the MEGs and EEGs and estimated the source currents from both measurements. Then using our recently proposed algorithm, we extracted repetitive spatiotemporal patterns from the source currents. The estimated patterns consisted of multiple frequency components, each of which transiently exhibited the frequency-specific resting-state networks (RSNs) of functional MRIs (fMRIs), such as the default mode and sensorimotor networks. A simulation test suggested that the spatiotemporal patterns reflected the phase alignment of the multiple frequency oscillators induced by the propagating activities along the anatomical connectivity. These results argue that whole-brain propagating activities transiently exhibited multiple RSNs in their multiple frequency components, suggesting that they reflected a process to integrate the information distributed over the frequencies and networks.

Comparison of traveling-subject and ComBat harmonization methods for assessing structural brain characteristics.

Multisite magnetic resonance imaging (MRI) is increasingly used in clinical research and development. Measurement biases-caused by site differences in scanner/image-acquisition protocols-negatively influence the reliability and reproducibility of image-analysis methods. Harmonization can reduce bias and improve the reproducibility of multisite datasets. Herein, a traveling-subject (TS) dataset including 56 T1-weighted MRI scans of 20 healthy participants in three different MRI procedures-20, 19, and 17 subjects in Procedures 1, 2, and 3, respectively-was considered to compare the reproducibility of TS-GLM, ComBat, and TS-ComBat harmonization methods. The minimum participant count required for harmonization was determined, and the Cohen's d between different MRI procedures was evaluated as a measurement-bias indicator. The measurement-bias reduction realized with different methods was evaluated by comparing test-retest scans for 20 healthy participants. Moreover, the minimum subject count for harmonization was determined by comparing test-retest datasets. The results revealed that TS-GLM and TS-ComBat reduced measurement bias by up to 85 and 81.3%, respectively. Meanwhile, ComBat showed a reduction of only 59.0%. At least 6 TSs were required to harmonize data obtained from different MRI scanners, complying with the imaging protocol predetermined for multisite investigations and operated with similar scan parameters. The results indicate that TS-based harmonization outperforms ComBat for measurement-bias reduction and is optimal for MRI data in well-prepared multisite investigations. One drawback is the small sample size used, potentially limiting the applicability of ComBat. Investigation on the number of subjects needed for a large-scale study is an interesting future problem.

Expansion coding and computation in the cerebellum: 50 years after the Marr-Albus codon theory.

Fifty years ago, David Marr and James Albus proposed a computational model of cerebellar cortical function based on the pioneering circuit models described by John Eccles, Masao Ito and Janos Szentagothai. The Marr-Albus model remains one of the most enduring and influential models in computational neuroscience, despite apparent falsification of some of the original predictions. We re-examine the Marr-Albus model in the context of the modern theory of computational neural networks and in the context of expanded interpretations of the connectivity and function of cerebellar cortex within the full motor system. By doing so, we show that the original elements of the codon theory continue to make important predictions for cerebellar mechanism, and we show that evidence appearing to contradict the original model is based on an artificially narrow interpretation of cerebellar structure and motor function.

Estimating repetitive spatiotemporal patterns from many subjects' resting-state fMRIs.

Recently, we proposed a method to estimate repetitive spatiotemporal patterns from resting-state brain activity data (SpatioTemporal Pattern estimation, STeP) (Takeda et al., 2016). From such resting-state data as functional MRI (fMRI), STeP can estimate several spatiotemporal patterns and their onsets even if they are overlapping. Nowadays, a growing number of resting-state data are publicly available from such databases as the Autism Brain Imaging Data Exchange (ABIDE), which promote a better understanding of resting-state brain activities. In this study, we extend STeP to make it applicable to such big databases, thus proposing the method we call BigSTeP. From many subjects' resting-state data, BigSTeP estimates spatiotemporal patterns that are common across subjects (common spatiotemporal patterns) as well as the corresponding spatiotemporal patterns in each subject (subject-specific spatiotemporal patterns). After verifying the performance of BigSTeP by simulation tests, we applied it to over 1,000 subjects' resting-state fMRIs (rsfMRIs) obtained from ABIDE I. This revealed two common spatiotemporal patterns and the corresponding subject-specific spatiotemporal patterns. The common spatiotemporal patterns included spatial patterns resembling the default mode (DMN), sensorimotor, auditory, and visual networks, suggesting that these networks are time-locked with each other. We compared the subject-specific spatiotemporal patterns between autism spectrum disorder (ASD) and typically developed (TD) groups. As a result, significant differences were concentrated at a specific time in a pattern, when the DMN exhibited large positive activity. This suggests that the differences are context-dependent, that is, the differences in fMRI activities between ASDs and TDs do not always occur during the resting state but tend to occur when the DMN exhibits large positive activity. All of these results demonstrate the usefulness of BigSTeP in extracting inspiring hypotheses from big databases in a data-driven way.

Estimating repetitive spatiotemporal patterns from resting-state brain activity data.

Repetitive spatiotemporal patterns in spontaneous brain activities have been widely examined in non-human studies. These studies have reported that such patterns reflect past experiences embedded in neural circuits. In human magnetoencephalography (MEG) and electroencephalography (EEG) studies, however, spatiotemporal patterns in resting-state brain activities have not been extensively examined. This is because estimating spatiotemporal patterns from resting-state MEG/EEG data is difficult due to their unknown onsets. Here, we propose a method to estimate repetitive spatiotemporal patterns from resting-state brain activity data, including MEG/EEG. Without the information of onsets, the proposed method can estimate several spatiotemporal patterns, even if they are overlapping. We verified the performance of the method by detailed simulation tests. Furthermore, we examined whether the proposed method could estimate the visual evoked magnetic fields (VEFs) without using stimulus onset information. The proposed method successfully detected the stimulus onsets and estimated the VEFs, implying the applicability of this method to real MEG data. The proposed method was applied to resting-state functional magnetic resonance imaging (fMRI) data and MEG data. The results revealed informative spatiotemporal patterns representing consecutive brain activities that dynamically change with time. Using this method, it is possible to reveal discrete events spontaneously occurring in our brains, such as memory retrieval.

Multi-subject and multi-task experimental validation of the hierarchical Bayesian diffuse optical tomography algorithm.

Diffuse optical tomography (DOT) is an emerging technology for improving the spatial resolution and spatial specificity of conventional multi-channel near-infrared spectroscopy (NIRS) by the use of high-density measurements and an image reconstruction algorithm. We recently proposed a hierarchical Bayesian DOT algorithm that allows for accurate simultaneous reconstruction of scalp and cortical hemodynamic changes, and verified its performance with a phantom experiment, a computer simulation, and experimental data from one human subject. We extend our previous human case study to a multi-subject, multi-task study, to demonstrate the validity of the algorithm on a wider population and varied task conditions. We measured brain activity during three graded tasks (hand movement, index finger movement, and no-movement), in 12 subjects, using high-density NIRS and functional magnetic resonance imaging (fMRI), acquired in different sessions. The reconstruction performance of our algorithm, and the current gold-standard method for DOT image reconstruction, were evaluated using the blood-oxygenation-level-dependent (BOLD) signals of the fMRI as a reference. In comparison with the BOLD signals, our method achieved a median localization error of 6 and 8mm, and a spatial-pattern similarity of 0.6 and 0.4 for the hand and finger tasks, respectively. It also did not reconstruct any activity in the no-movement task. Compared with the current gold-standard method, the new method showed fewer false positives, which resulted in improved spatial-pattern similarity, although the localization errors of the main activity clusters were comparable.

Reduction of global interference of scalp-hemodynamics in functional near-infrared spectroscopy using short distance probes.

Functional near-infrared spectroscopy (fNIRS) is used to measure cerebral activity because it is simple and portable. However, scalp-hemodynamics often contaminates fNIRS signals, leading to detection of cortical activity in regions that are actually inactive. Methods for removing these artifacts using standard source-detector distance channels (Long-channel) tend to over-estimate the artifacts, while methods using additional short source-detector distance channels (Short-channel) require numerous probes to cover broad cortical areas, which leads to a high cost and prolonged experimental time. Here, we propose a new method that effectively combines the existing techniques, preserving the accuracy of estimating cerebral activity and avoiding the disadvantages inherent when applying the techniques individually. Our new method accomplishes this by estimating a global scalp-hemodynamic component from a small number of Short-channels, and removing its influence from the Long-channels using a general linear model (GLM). To demonstrate the feasibility of this method, we collected fNIRS and functional magnetic resonance imaging (fMRI) measurements during a motor task. First, we measured changes in oxygenated hemoglobin concentration (∆Oxy-Hb) from 18 Short-channels placed over motor-related areas, and confirmed that the majority of scalp-hemodynamics was globally consistent and could be estimated from as few as four Short-channels using principal component analysis. We then measured ∆Oxy-Hb from 4 Short- and 43 Long-channels. The GLM identified cerebral activity comparable to that measured separately by fMRI, even when scalp-hemodynamics exhibited substantial task-related modulation. These results suggest that combining measurements from four Short-channels with a GLM provides robust estimation of cerebral activity at a low cost.

MEG source reconstruction based on identification of directed source interactions on whole-brain anatomical networks.

We present an MEG source reconstruction method that simultaneously reconstructs source amplitudes and identifies source interactions across the whole brain. In the proposed method, a full multivariate autoregressive (MAR) model formulates directed interactions (i.e., effective connectivity) between sources. The MAR coefficients (the entries of the MAR matrix) are constrained by the prior knowledge of whole-brain anatomical networks inferred from diffusion MRI. Moreover, to increase the accuracy and robustness of our method, we apply an fMRI prior on the spatial activity patterns and a sparse prior on the MAR coefficients. The observation process of MEG data, the source dynamics, and a series of the priors are combined into a Bayesian framework using a state-space representation. The parameters, such as the source amplitudes and the MAR coefficients, are jointly estimated from a variational Bayesian learning algorithm. By formulating the source dynamics in the context of MEG source reconstruction, and unifying the estimations of source amplitudes and interactions, we can identify the effective connectivity without requiring the selection of regions of interest. Our method is quantitatively and qualitatively evaluated on simulated and experimental data, respectively. Compared with non-dynamic methods, in which the interactions are estimated after source reconstruction with no dynamic constraints, the proposed dynamic method improves most of the performance measures in simulations, and provides better physiological interpretation and inter-subject consistency in real data applications.

Manifold alteration between major depressive disorder and healthy control subjects using dynamic mode decomposition in resting-state fMRI data.

Background: The World Health Organization has reported that approximately 300 million individuals suffer from the mood disorder known as MDD. Non-invasive measurement techniques have been utilized to reveal the mechanism of MDD, with rsfMRI being the predominant method. The previous functional connectivity and energy landscape studies have shown the difference in the coactivation patterns between MDD and HCs. However, these studies did not consider oscillatory temporal dynamics. Methods: In this study, the dynamic mode decomposition, a method to compute a set of coherent spatial patterns associated with the oscillation frequency and temporal decay rate, was employed to investigate the alteration of the occurrence of dynamic modes between MDD and HCs. Specifically, The BOLD signals of each subject were transformed into dynamic modes representing coherent spatial patterns and discrete-time eigenvalues to capture temporal variations using dynamic mode decomposition. All the dynamic modes were disentangled into a two-dimensional manifold using t-SNE. Density estimation and density ratio estimation were applied to the two-dimensional manifolds after the two-dimensional manifold was split based on HCs and MDD. Results: The dynamic modes that uniquely emerged in the MDD were not observed. Instead, we have found some dynamic modes that have shown increased or reduced occurrence in MDD compared with HCs. The reduced dynamic modes were associated with the visual and saliency networks while the increased dynamic modes were associated with the default mode and sensory-motor networks. Conclusion: To the best of our knowledge, this study showed initial evidence of the alteration of occurrence of the dynamic modes between MDD and HCs. To deepen understanding of how the alteration of the dynamic modes emerges from the structure, it is vital to investigate the relationship between the dynamic modes, cortical thickness, and surface areas.

Fast, accurate, and interpretable decoding of electrocorticographic signals using dynamic mode decomposition.

Dynamic mode (DM) decomposition decomposes spatiotemporal signals into basic oscillatory components (DMs). DMs can improve the accuracy of neural decoding when used with the nonlinear Grassmann kernel, compared to conventional power features. However, such kernel-based machine learning algorithms have three limitations: large computational time preventing real-time application, incompatibility with non-kernel algorithms, and low interpretability. Here, we propose a mapping function corresponding to the Grassmann kernel that explicitly transforms DMs into spatial DM (sDM) features, which can be used in any machine learning algorithm. Using electrocorticographic signals recorded during various movement and visual perception tasks, the sDM features were shown to improve the decoding accuracy and computational time compared to conventional methods. Furthermore, the components of the sDM features informative for decoding showed similar characteristics to the high-γ power of the signals, but with higher trial-to-trial reproducibility. The proposed sDM features enable fast, accurate, and interpretable neural decoding.

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.

Learning task-agnostic and interpretable subsequence-based representation of time series and its applications in fMRI analysis.

The recent success of sequential learning models, such as deep recurrent neural networks, is largely due to their superior representation-learning capability for learning the informative representation of a targeted time series. The learning of these representations is generally goal-directed, resulting in their task-specific nature, giving rise to excellent performance in completing a single downstream task but hindering between-task generalisation. Meanwhile, with increasingly intricate sequential learning models, learned representation becomes abstract to human knowledge and comprehension. Hence, we propose a unified local predictive model based on the multi-task learning paradigm to learn the task-agnostic and interpretable subsequence-based time series representation, allowing versatile use of learned representations in temporal prediction, smoothing, and classification tasks. The targeted interpretable representation could convey the spectral information of the modelled time series to the level of human comprehension. Through a proof-of-concept evaluation study, we demonstrate the empirical superiority of learned task-agnostic and interpretable representation over task-specific and conventional subsequence-based representation, such as symbolic and recurrent learning-based representation, in solving temporal prediction, smoothing, and classification tasks. These learned task-agnostic representations can also reveal the ground-truth periodicity of the modelled time series. We further propose two applications of our unified local predictive model in functional magnetic resonance imaging (fMRI) analysis to reveal the spectral characterisation of cortical areas at rest and reconstruct more smoothed temporal dynamics of cortical activations in both resting-state and task-evoked fMRI data, giving rise to robust decoding.

Virtual reality-based real-time imaging reveals abnormal cortical dynamics during behavioral transitions in a mouse model of autism.

Functional connectivity (FC) can provide insight into cortical circuit dysfunction in neuropsychiatric disorders. However, dynamic changes in FC related to locomotion with sensory feedback remain to be elucidated. To investigate FC dynamics in locomoting mice, we develop mesoscopic Ca2+ imaging with a virtual reality (VR) environment. We find rapid reorganization of cortical FC in response to changing behavioral states. By using machine learning classification, behavioral states are accurately decoded. We then use our VR-based imaging system to study cortical FC in a mouse model of autism and find that locomotion states are associated with altered FC dynamics. Furthermore, we identify FC patterns involving the motor area as the most distinguishing features of the autism mice from wild-type mice during behavioral transitions, which might correlate with motor clumsiness in individuals with autism. Our VR-based real-time imaging system provides crucial information to understand FC dynamics linked to behavioral abnormality of neuropsychiatric disorders.