The Research Of Key Technology In DTI Connectome Analysis

Human brain is a complex system with a large number of neurons and synapses, also the center of thinking, communication, sports and computering, makeing it hard to be understood. Diffusion Tensor Imaging(DTI) is an new and important non-invasive technology to describe white matter tractography of brain. Based on DTI, the brain network consisted by nodes and links is called DTI connectome. Recently, researches of DTI connectome have uncovered the structural configuration and working mechanism of brain from the perspective of network analysis, however, some key technologies in conectome analysis have not been fully covered. In this study, we focus on three key parts of DTI connectome analysis, trying to solve these technical challenges and to make contributions to the development of connectme research. The contents of this research are as follows:(1) Research on the effect of parcellation atlases on the DTI connectome analysis. After determining the factors of image acquisition, tracing algorithm and adjacency matrix type, the five types of DTI connectome of seventy-five healthy younger elderly were firstly constructed by using five parcellation atlases, including AAL, HOA, UPA32, UPA128 and UPA512. Then, this study found some results after calculating and analizing network features. Firstly, results showed that the small-worldness of all the brain networks was not limited by the selection of parcellation atlases. Secondly, results also demonstrated that there were correlations between modules of brain networks, which indicating the effect of parcellation atlases on the modules of brain networks was small. Thirdly, due to the differences of parcellation atlases, the values of network feautes in different types of brain networks were obviously different. Lastly, parcellation atlases directly impacted the sensitivity of network features to age. According to theses results, we recommend that AAL or UPA128, which has the moderate scale in the atlas, is more suitable to the DTI connectome analysis.(2) Research on the study of hub identification in DTI connectome. The type of brain network, identification parameters and indentification thresholds are the key elements in hub identification. In this study, brain networks included the anatomical networks constructed by AAL and HOA, and randomly parcellated networks constructed by UPAs. Results indicated that similar hubs were identified by different identification parameters in the same type of brain network, while significant different hubs were detected in the different types of brain networks even using the same identification parameter, which demonstrating the effect of brain network type on the identified results of hubs. Identification threshold contained 10 thresholds from 5% to 10%. Results showed that identified results derived randomly parcellated brain networks were more vulnerable to the change of thresholds than that derived from anatomical brain networks, revealing the importance of threshold in the hub identification. Here, degree, betweenness centrality, vulnerability and composite measure served as the identification parameters in this study. Our results demonstrated that indentification parameters were correlated, but it is impacted by the types of brain network. Results of hub classification also uncovered the advantages of composite measure on identifying connector hub as well as the excellence of node degree on detecting provincial hubs, implying the different roles of identification parameters played in hub classification.(3) Research on the evaluation index of small-worldness in DTI connectome. The small-worldness is one of the most critical network attributions in connectome analysis, however, no such index that could quantitatively and accurately estimate small world characteristics was proposed yet. Therefore, in this study, we proposed an index, normalized network efficiency(E), to quantitatively assess small-worldness of DTI connectome. E was vrtified through the tests, and results revealed that E was not effected by the parcellation atlases. Meanwhile, it was sensitive to the detection of randomization of real brain network and simulation network as well as cognitive parameters, elucidating the potential application of E in the study of aging and diseased brain networks, which is better than ?.