sss

Construction of Brain Function Network

Visual Analytics of Brain Function Network

To clarify the role of the brain in forming a network and working as a dynamic system, it is necessary to investigate the influence of certain parts of the brain on the other parts, that is, their “causal relationships.” We are conducting research into the visual analytics of the brain network by using non-invasive brain function measurement data such as magnetoencephalography (MEG) data, and analyzing the connectivity and causality between brain regions. Our research will support humans in drawing conclusions by visualizing the calculation results in real time and analyzing the results interactively.

Figure 1: MEG system and Functional Brain Imaging

Based on the development of non-invasive brain function measurement methods, it is slowly becoming clearer “when” and “where” the human brain is active. In the field of human brain function imaging, research is being conducted to investigate the function of the cerebral cortex as a network so as to understand the functions between cortical regions. For example, the development of diffusion MRI, which applies MRI technology and makes it possible to quantitatively investigate not only anatomical images but also nerve fiber bundles, as well as the rise of analytical methods such as probabilistic tractography, have now made it possible to visualize the appearance of network connections. However, because this tractography does not provide any information on the direction of the connections (information flow), it is necessary to investigate the connection relationships between brain parts using time-series data capturing the electrical activity of the neurons such as magnetoencephalograms (MEG) and electroencephalograms (EEG), as well as the flow of information related to the connection relationships. To clarify the role of the brain in forming a network and working as a dynamic system, it is necessary to investigate the influence of certain parts of the brain on other parts, that is, their “causal relationships.”

In addition, along with the development of measurement methods, as more and more diverse data are gathered at a large scale, data analysis methods are also advancing. However, to cope with many complicated problems, the analysis methods themselves also tend to become complicated, and it is necessary to rely on the intelligence of the analyst at an early stage in the data analysis. For instance, there is a demand for the ability to make judgments, such as which data and pre-processing should be used, or which analysis method should be applied, and for judgments regarding the decisions to be taken after obtaining sufficient information by visualizing the stages of the complicated analysis process. For this purpose, a visual analysis (visual analytics) is needed, which is an analytical reasoning technique supported by an interactive visual interface. This is an analysis method that combines the knowledge of the analysts (doctors and researchers) in the areas of neuroscience and computational techniques.

Figure 2: Graph drawing of a brain network model

(upper figure) Node link graph in MNI coordinate system (lower figure) Node link graph in force-directed layout

Against this backdrop, through our research, we are constructing a visual analysis method for brain networks to support analysts in making inferences when analyzing the connectivity between brain regions by visualizing the calculation process and results in real time and interactively analyzing the results. In particular, we aim at providing comprehensive hypothesis verification by constructing a visual analysis for various types of coupling analysis and causal analysis methods such as Granger Causality based on a vector autoregressive model, dynamical causal modeling, which is a model-based analysis method, and convergent cross mapping, which was recently developed.

This research involves collaborations with various researchers in information science and medicine, and has the objectives of developing and applying visual analyses. In recent years, aiming at clinical applications such as diagnostic support, research fields focusing on the relationship between diseases and imaging indices such as brain volume information using voxel-based morphometry (VBM), diffusion anisotropy using diffusion MRI, tractography, and other tools are flourishing; however, disease-specific changes that appear in behavior and cognition may be prominently expressed as changes in the parameters of the brain network rather than as a partial form change. The goal is to be able to analyze such changes in an easy-to-understand manner. We proposed a visual analytics method and applied it to clinical data to clarify the relationships between diseases and the brain network with the aim of contributing to medical treatments, rehabilitation, and other areas.

Research Achievements

We proposed a method for visualizing neural network models handled in dynamical causal modeling, which is a type of coupling analysis between brain parts using magnetoencephalography (MEG). This network model is a directed graph with edge weighting and bidirectional coupling and has a hierarchical structure. As a graph representation, we proposed a node link graph with MNI coordinates and a force-directed node link network. In the future, we plan to develop the above into a visual analysis technique with detailed evaluations of visual expressions.

PAGE TOP

Comprehensive Visualization

Causal Relationship Visualization