Construction of Individual Morphological Brain Networks with Multiple Morphometric Features

• Main Authors: Chunlan Yang, Wan Li, Feng Shi, Shuicai Wu, Qun Wang, Yingnan Nie, Xin Zhang
• Format: Article
• Language: English
• Published: Frontiers Media S.A. 2017-04-01
• Series: Frontiers in Neuroanatomy
• Subjects: individual morphological brain network; multiple morphometric features; feature vector; graph theory; reliability
• Online Access: http://journal.frontiersin.org/article/10.3389/fnana.2017.00034/full

In recent years, researchers have increased attentions to the morphological brain network, which is generally constructed by measuring the mathematical correlation across regions using a certain morphometric feature, such as regional cortical thickness and voxel intensity. However, cerebral structure can be characterized by various factors, such as regional volume, surface area, and curvature. Moreover, most of the morphological brain networks are population-based, which has limitations in the investigations of individual difference and clinical applications. Hence, we have extended previous studies by proposing a novel method for realizing the construction of an individual-based morphological brain network through a combination of multiple morphometric features. In particular, interregional connections are estimated using our newly introduced feature vectors, namely, the Pearson correlation coefficient of the concatenation of seven morphometric features. Experiments were performed on a healthy cohort of 55 subjects (24 males aged from 20 to 29 and 31 females aged from 20 to 28) each scanned twice, and reproducibility was evaluated through test–retest reliability. The robustness of morphometric features was measured firstly to select the more reproducible features to form the connectomes. Then the topological properties were analyzed and compared with previous reports of different modalities. Small-worldness was observed in all the subjects at the range of the entire network sparsity (20–40%), and configurations were comparable with previous findings at the sparsity of 23%. The spatial distributions of the hub were found to be significantly influenced by the individual variances, and the hubs obtained by averaging across subjects and sparsities showed correspondence with previous reports. The intraclass coefficient of graphic properties (clustering coefficient = 0.83, characteristic path length = 0.81, betweenness centrality = 0.78) indicates the robustness of the present method. Results demonstrate that the multiple morphometric features can be applied to form a rational reproducible individual-based morphological brain network.