Effects of brain network construction on individual differences of topological properties

Abstract

Advances in diffusion MRI techniques have made it possible to virtually reconstruct white matter tracts, which further allows the modeling of the human brain as a complex network in vivo . Graph theoretical approaches have been applied to characterize topological architectures of whole-brain white matter networks. By using these techniques, many important topological properties for the whole-brain white matter network of human brain have been revealed in many previous works, while the construction method for white matter networks differs across studies. At present, a few studies have assessed the effects of network construction methods on specific white matter network properties and substantial influences of the construction methods were observed. However, the previous research mainly focuses on the effects of the constructed methods on the same subjects for their network properties. Up to now, how the topological properties are affected by construction methods for the individual differences is still unknown, even it is the key for the investigation of white matter networks in a cross-sectional manner, including group comparison, correlational analysis, etc. Therefore, it is necessary but still insufficient to make a systematic evaluation for the network construction methods on the individual differences. In the present work, the influence of different network construction methods on the network properties is desired to be assessed individual differences. Thirty-two methods were adopted to construct the whole-brain white matter networks of 80 healthy adult subjects, which includes two node definitions (low-resolution and high-resolution), four deterministic tractography algorithms (Fiber Assignment by Continuous Tracking, Tensorline Tracking, Interpolated streamline method and 2nd-order Runge-Kutta method) and four weighted edge definitions (binary, FA weighted, fiber-length weighted, and fiber-number weighted). For these white matter networks, individual differences were analyzed by graph theory and hierarchical clustering in three topological parameters including the global efficiency, the local efficiency and the nodal efficiency of the network. The results can be summarized that, firstly, the nodal efficiency with the same nodal definition exhibits convergent individual differences, which indicated that the individual differences of network nodal properties are divergent across network resolutions. Secondly, the individual differences of global and local efficiency of white matter networks constructed by different deterministic fiber tracking algorithms do not differ significantly. However, the constructed networks based on the same deterministic fiber tracking algorithm exhibited the same individual differences with the different node efficiency, which indicated that the effect of fiber tracking algorithm should be emphasized in the comparison and analysis of nodal properties in different studies. Finally, the method clustering analysis for the network global efficiency and local efficiency indicates that the networks with both low-resolution and high-resolution will compose to be the same module with the definition of the same edges, which indicated the high convergent of individual differences for the network efficiencies with different resolutions. It can be concluded that the difference of network construction methods depends on the network properties when it is used to reflexing the individual difference. These findings may provide valuable implications for understanding the intersubject variability of white matter networks and comparing results across different studies.