IBS Institute for Basic Science

As all research groups acquire enormous dynamic data, it is crucial to store and process these data efficiently, and to extract biologically-relevant information.

Typically, imaging data are quite noisy, and functional signal changes are tiny. Thus, effective computation of imaging data to amplify the signal is a critical issue. Therefore, this group focuses on computational problems arising in large scale, high dimensional, non-Euclidean biological images and will provide new insights for analyzing images. This group directs “the IT Core” for supporting computation infrastructure in the CNIR.


Research Interests

1. Brain image pre-processing

Neuroimaging data are so large and complex and the raw data cannot be directly used for group-level analysis due to the gradient distortions from the MRI scanner, head movements while scanning, and the differences of the imaging space and intensity scales. Pre-processing steps including gradient distortion correction, head motion correction, slice timing correction, image registration and segmentation, nuisance variable regression, and temporal filtering are required to handle these problems.



Non brain tissue removal


Slice timing correction




Image registration




Image segmentation 


2. Multi-modal analysis

Imaging modalities contain functional (fMRI, SPECT, PET) and structural (DTI, T1-weighted, T2-weighted MRI) images. Each of them has unique information to quantify the function and the structure of the brain. Combining the features extracted from multiple modalities leads to increase the reliability of the analysis. New computational tools and algorithms will be developed to integrate information from functional and structural images in a coherent manner. There is a huge lack of a coherent computational and mathematical framework for integrating various imaging modalities. Thus this group will explore new technique for the fusion of multi-modal and multi-resolution imaging data.


Multimodal image fusion


Process of combining multiple modalities generates big data. The dimensionality reduction algorithms including principal component analysis (PCA) or multidimensional scaling (MDS) are required to analyze the big data. Dimensionality reduced features can be applied to modern classifier such as support vector machine (SVM) and it shows better performance.



Dimensionality reduction algorithms




Hierarchical model of the brain connections



3. Connectivity analysis

The human brain connectome project is currently one of the great scientific interests. The project tries to elucidate the neural connections of the brain function. Connectivity analysis focuses on how activities in one brain region correlate with activities in another brain region and thus allows the observation of the whole brain as a complex, interconnected network. Graph theory based algorithms can be applied to quantify the brain network patterns.


Correlation matrix



4. Fiber tractography

DTI provides in vivo neuronal fiber information using anisotropic water diffusion. Diffusion tensor is calculated by eigenvalues and eigenvectors that represent the size and the direction of the principle axes respectively. Voxels are connected based on similarities in the maximum diffusion direction and the fiber tracts are traced.

Fiber bundles

Functional connection (left) and fiber bundles (right)




5. Brain anatomy quantification

Computational neuroanatomy deals with the computational problems arising from the quantification of the structure and the function of the human brain in imaging modalities such as fMRI and DTI. The group’s research efforts will be concentrated on the methodological development of quantifying anatomical shape variations in both normal and clinical populations using various mathematical, statistical and computational techniques.