Li Xie

Characterizing and Differentiating Brain State Dynamics via Hidden Markov Models

Functional connectivity measured from resting state fMRI (R-fMRI) data has been widely used to examine the brain’s functional activities and has been recently used to characterize and differentiate brain conditions. However, the dynamical transition patterns of the brain’s functional states have been less explored. In this work, we propose a novel computational framework to quantitatively characterize the brain state dynamics via hidden Markov models (HMMs) learned from the observations of temporally dynamic functional connectomics, denoted as functional connectome states. The framework has been applied to the R-fMRI dataset including 44 post-traumatic stress disorder (PTSD) patients and 51 normal control (NC) subjects. Experimental results show that both PTSD and NC brains were undergoing remarkable changes in resting state and mainly transiting amongst a few brain states. Interestingly, further prediction with the best-matched HMM demonstrates that PTSD would enter into, but could not disengage from, a negative mood state. Importantly, 84xa0% of PTSD patients and 86xa0% of NC subjects are successfully classified via multiple HMMs using majority voting.

Bridging low-level features and high-level semantics via fMRI brain imaging for video classification

The multimedia content analysis community has made significant effort to bridge the gap between low-level features and high-level semantics perceived by human cognitive systems such as real-world objects and concepts. In the two fields of multimedia analysis and brain imaging, both topics of low-level features and high level semantics are extensively studied. For instance, in the multimedia analysis field, many algorithms are available for multimedia feature extraction, and benchmark datasets are available such as the TRECVID. In the brain imaging field, brain regions that are responsible for vision, auditory perception, language, and working memory are well studied via functional magnetic resonance imaging (fMRI). This paper presents our initial effort in marrying these two fields in order to bridge the gaps between low-level features and high-level semantics via fMRI brain imaging. Our experimental paradigm is that we performed fMRI brain imaging when university student subjects watched the video clips selected from the TRECVID datasets. At current stage, we focus on the three concepts of sports, weather, and commercial-/advertisement specified in the TRECVID 2005. Meanwhile, the brain regions in vision, auditory, language, and working memory networks are quantitatively localized and mapped via task-based paradigm fMRI, and the fMRI responses in these regions are used to extract features as the representation of the brains comprehension of semantics. Our computational framework aims to learn the most relevant low-level feature sets that best correlate the fMRI-derived semantics based on the training videos with fMRI scans, and then the learned models are applied to larger scale test datasets without fMRI scans for category classifications. Our result shows that: 1) there are meaningful couplings between brains fMRI responses and video stimuli, suggesting the validity of linking semantics and low-level features via fMRI; 2) The computationally learned low-level feature sets from fMRI-derived semantic features can significantly improve the classification of video categories in comparison with that based on original low-level features.

Atomic dynamic functional interaction patterns for characterization of ADHD

Modeling abnormal temporal dynamics of functional interactions in psychiatric disorders has been of great interest in the neuroimaging field, and thus a variety of methods have been proposed so far. However, the temporal dynamics and disease‐related abnormalities of functional interactions within specific data‐driven discovered subnetworks have been rarely explored yet. In this work, we propose a novel computational framework composed of an effective Bayesian connectivity change point model for modeling functional brain interactions and their dynamics simultaneously and an effective variant of nonnegative matrix factorization for assessing the functional interaction abnormalities within subnetworks. This framework has been applied on the resting state fmagnetic resonance imaging (fMRI) datasets of 23 children with attention‐deficit/hyperactivity disorder (ADHD) and 45 normal control (NC) children, and has revealed two atomic functional interaction patterns (AFIPs) discovered for ADHD and another two AFIPs derived for NC. Together, these four AFIPs could be grouped into two pairs, one common pair representing the common AFIPs in ADHD and NC, and the other abnormal pair representing the abnormal AFIPs in ADHD. Interestingly, by comparing the abnormal AFIP pair, two data‐driven abnormal functional subnetworks are derived. Strikingly, by evaluating the approximation based on the four AFIPs, all of the ADHD children were successfully differentiated from NCs without any false positive. Hum Brain Mapp 35:5262–5278, 2014.

A Robot Collision Avoidance Scheme Based on the Moving Obstacle Motion Prediction

This paper proposes an integrated scheme for the robot to avoid collision based on obstaclepsilas motion prediction. The robot obtains the target bearing by using object detection technique and fuses the bearing information with the external sensors data to locate the current position of the moving target. The motion state of the moving obstacle is described using the constant velocity (CV) model, the constant accelerate (CA) model and the current statistical (CS) model. Then, a Kalman-based interacting multiple model (IMM) filter is adopted to estimate the obstacle motion trend. This scheme mainly focuses on the obstacle detection and the motion prediction. Based on the motion prediction of the obstacle, an abbreviated strategy is proposed for collision avoidance. To validate the proposed scheme, we have implemented an obstacle avoidance experiment with a Pioneer 3-AT robot and a moving obstacle. The results verify that the proposed scheme is valid and viable.

Sequential Asynchronous Filters for Target Tracking in Wireless Sensor Networks

Asynchronous data fusion is inevitable for target tracking in asynchronous wireless sensor networks, where multiple sensors are required to locate a target collaboratively. The predicted estimates of the follow-up states based on the to-be-estimated state are first introduced to overcome the drawback that asynchronous measurements cannot be fused directly. Then, the sequential asynchronous Bayesian state estimation is deduced based on the predicted estimates. The proposed estimation process is comprised of two steps: 1) the prediction step and 2) the update step. Finally, sequential asynchronous filters based Kalman filter and particle filter are proposed. Simulations demonstrate that the proposed algorithms perform not only better than the benchmark algorithms with asynchronous measurements, but also better than the benchmark algorithm with synchronous measurements.

A distributed localization scheme based on mobility prediction for underwater wireless sensor networks

A distributed localization scheme based on mobility prediction for mobile underwater wireless sensor networks is proposed. Anchor nodes perform self-localization and mobility prediction, and serve as reference nodes to localize ordinary nodes. According to the group movement properties of underwater objects, ordinary nodes perform self-localization by utilizing the neighboring nodes locations and their predicted speed vectors. Well-localized ordinary nodes serve as reference nodes to localize neighbors further. The modified covariance algorithm is employed to estimate the linear prediction parameters of the mobility pattern. A Node-Selection Strategy is proposed to select the most suitable reference nodes to localize an ordinary node. The simulation results demonstrate the advantages of the proposed scheme.

Modeling brain functional dynamics via hidden Markov models

Functional connectivities constructed via resting state fMRI (R-fMRI) data have been widely used to study the brains functional activities and to characterize the brains states. However, the temporal dynamic transition patterns of the brains functional states have been rarely investigated before. In this paper, we present a novel algorithmic framework to cluster and label the brains functional states, and learn their hidden Markov models (HMMs). Here, the brains functional state is compactly represented by a large-scale functional connectivity matrix, called functional connectome state (FCS), and the temporal FCS sequences are derived via an overlapping sliding time window approach. The best-matched HMM learned for ADHD patients revealed a meaningful phenomenon of psychiatric conditions, that is, the tendency to enter into, and inability to disengage from, a negative mood state. Experimental results demonstrated 87% of ADHD patients and 89% of normal controls are successfully classified via multiple HMMs by using majority voting.

Atomic connectomics signatures for characterization and differentiation of mild cognitive impairment

In recent years, functional connectomics signatures have been shown to be a very valuable tool in characterizing and differentiating brain disorders from normal controls. However, if the functional connectivity alterations in a brain disease are localized within sub-networks of a connectome, then accurate identification of such disease-specific sub-networks is critical and this capability entails both fine-granularity definition of connectome nodes and effective clustering of connectome nodes into disease-specific and non-disease-specific sub-networks. In this work, we adopted the recently developed DICCCOL (dense individualized and common connectivity-based cortical landmarks) system as a fine-granularity high-resolution connectome construction method to deal with the first issue, and employed an effective variant of non-negative matrix factorization (NMF) method to pinpoint disease-specific sub-networks, which we called atomic connectomics signatures in this work. We have implemented and applied this novel framework to two mild cognitive impairment (MCI) datasets from two different research centers, and our experimental results demonstrated that the derived atomic connectomics signatures can effectively characterize and differentiate MCI patients from their normal controls. In general, our work contributed a novel computational framework for deriving descriptive and distinctive atomic connectomics signatures in brain disorders.

Assessing graph models for description of brain networks

Both structural and functional brain networks have been investigated in the literature with enthusiasm via graph-theoretical methods. However, an important issue that has not been adequately addressed before is: what is the optimal graph model for describing brain networks, both in structural and functional aspects? We address this question in the following three aspects. First, multi-resolution structural brain networks are reconstructed via cortical surface parcellation based on white matter fiber density information. Second, the global and local graph properties of the constructed networks are measured using state-of-the-art graph analysis algorithms and tools, and are further compared with five popular random graph models. Third, a functional simulation study is conducted to evaluate the synchronizability of the five models. Our results suggest that the STICKY graph model fits brain networks the best in terms of global and local graph properties, and the fastest speed of functional synchronization.

Improved Object Classification and Tracking Based on Overlapping Cameras in Video Surveillance

Object classification and tracking are important in intelligent video surveillance systems. In this paper, an approach based on multiple overlapping cameras cooperation is proposed for object classification and tracking. In the proposed surveillance system, all the cameras are connected to the central computer server through network connection. Viewpoint correspondence and data fusion from multiple overlapping cameras are utilized to improve object classification and tracking in complex occlusion scenes. This paper demonstrates the benefit gained both in tracking and classification through the communication between the two individual modules. Experimental results show that the proposed method achieves higher classification accuracy and tracking performance in comparison with single-camera method.