Joint analysis of multi-level functional brain networks

• Published in: 2016 9th International Congress on Image and Signal Processing, BioMedical Engineering and Informatics (CISP-BMEI) pp. 1521 - 1526
• Main Authors: Huiwen Luo, Weibei Dou, Yu Pan, Yueheng Wang, Yujia Mu, Yudu Li, Xiaojie Zhang, Quan Xu, Shuyu Yan, Yuanyuan Tu
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.10.2016
• Subjects: Brain; Correlation; Correlation coefficient; Data mining; Feature extraction; Magnetic resonance imaging; Reliability
• DOI: 10.1109/CISP-BMEI.2016.7852956

Building brain networks based on functional Magnetic Resonance Imaging (fMRI) signal is one of the efficient methods to study functional connectivity of human brain. Various methods of constructing brain network will lead to different results. It is wondered which method is reliable. Therefore, it is necessary to set up a synthetical framework of brain network analysis to study the functional connectivity. A joint analysis method of multi-level functional brain networks is proposed in this paper. These networks are constructed based on different correlation matrixes of fMRI signal between voxels and between anatomical areas (regions) of brain. They are called whole brain network of voxel-based and region-based, and local network of voxel-based inside brain regions. The joint analysis implements feature combination of global and local network attributes to measure or evaluate the brain region characteristics towards reducing uncertainty. The resting-state fMRI data of 37 subjects (22 normal subjects and 15 patients with spinal cord injury (SCI)) have been used to test the proposed method. Three-level functional connectivity networks are jointly analyzed to combine the two-type significant features, the significant differences between normal and patient, and the significant correlations between network features and clinic function scores of patient. The results of the features combination are validated by the specific Brodmann area (BA) regions characterized by the similar and the complementary features, and most of them belong to the dorsolateral prefrontal cortex (DLPFC) and correspond with SCI disease. Compared with network analysis of the commonly used voxel-based whole brain network, the proposed joint analysis method can provide more central, more robust and more reliable evidences. Overall, the proposed method takes advantages of different functional networks and shows the complete discovery to us by the consistency and mutual complementation of these kinds of networks. It would be a new network analysis method of human brain.