The impact of sampling density upon cortical network analysis: regions or points

Area of Science:

• Neuroimaging
• Network Science
• Computational Neuroscience

Background:

• Network analysis of brain imaging data requires defining nodes and edges.
• The choice of representation significantly impacts network analysis outcomes.
• Magnetic resonance imaging (MRI)-based cortical thickness networks present unique challenges in representation.

Purpose of the Study:

• To compare network analysis results between two representations of MRI-based cortical thickness networks: vertex-level and regional-level.
• To investigate the influence of representation detail on network properties.
• To assess the robustness of network parameters to correlation thresholds at different resolutions.

Main Methods:

• Constructed cortical thickness networks using nodes representing either individual cortical surface mesh vertices or defined cortical brain regions.
• Compared standard network analysis measures (connectivity, clustering, centrality) between vertex-level and regional-level representations.
• Evaluated the impact of varying correlation thresholds on network parameters at both resolution levels.

Main Results:

• Basic network measures showed expected behavior with increased detail.
• Vertex-level networks exhibited higher overall node connectivity and lower clustering compared to regional-level networks.
• Node centralities differed between the two representations, with vertex-level parameters demonstrating greater robustness to correlation threshold selection.

Conclusions:

• While qualitative network properties are often consistent, vertex-level resolution uncovers details not apparent in regional-level networks.
• The increased detail from vertex-level analysis may offer advantages for specific applications.
• The methodology can be extended to study sampling density effects in other brain imaging networks, such as resting-state functional MRI.