Brain Connectivity Pattern Analysis Based On DTI

Diffusion tensor imaging (DTI) is a kind of imaging method that makes use of the anisotropy of water molecule in the tissue to explore the microstructure of the tissue. It can measure the diffusion direction and velocity of water in white matter fibers, and then estimate the diffusion tensor model in the voxel and track the white matter fiber bundles. As a noninvasive method for detecting subtle changes in tissue micro-structural organization, diffusion tensor imaging has played a significant role in brain research. We introduce the machine learning approaches into the DTI researches and make classification of the fiber connectivity patterns between depressive patients and controls.Finally, we will analysis the fiber connections of significant discrimination. This paper accomplishes brain region segmentation with warp method and set up diffusion tensor models, and makes use of probabilistic tracking method for reconstruction of white matter fiber bundles. This paper proposes taking white matter fiber connectivity as the classification features, and uses machine learning approaches combining Local Linear Embedding and Support Vector Machine to classify depressive patients and controls. The classification rate is 91.7% deduced by cross-validation, then, we test the reliability of classification methods and results. Cross-validation extracts 33 features with significant differences for classification, and they mainly distributed in frontal-limbic, parietal-limbic and occipital-temporal networks. The abnormalities in the three networks are highly related to the affection, cognition, attention abnormalities in depression. This finding implies that the abnormal fiber connections in the three networks may be explain the behavior abnormalities in depression and be a biomarker in clinic.