Communicability in complex brain networks

Jonathan J. Crofts, Desmond J. HighamarXiv 2008 · DOI · arXiv

Abstract

Recent advances in experimental neuroscience allow, for the first time, non-invasive studies of the white matter tracts in the human central nervous system, thus making available cutting-edge brain anatomical data describing these global connectivity patterns. This new, non-invasive, technique uses magnetic resonance imaging to construct a snap-shot of the cortical network within the living human brain. Here, we report on the initial success of a new weighted network communicability measure in distinguishing local and global differences between diseased patients and controls. This approach builds on recent advances in network science, where an underlying connectivity structure is used as a means to measure the ease with which information can flow between nodes. One advantage of our method is that it deals directly with the real-valued connectivity data, thereby avoiding the need to discretise the corresponding adjacency matrix, that is, to round weights up to 1 or down to 0, depending upon some threshold value. Experimental results indicate that the new approach is able to highlight biologically relevant features that are not immediately apparent from the raw connectivity data.

Introduction

In network science, the analysis of complex systems often relies on the "shortest path" between two nodes. While this is effective for navigation, the human brain does not always operate like a GPS; information can disseminate along multiple non-shortest routes simultaneously. Traditional methods for analyzing brain connectivity derived from MRI scans face two significant hurdles. First, the "binary trap" requires researchers to discretize real-valued connectivity strengths into simple 1s and 0s based on arbitrary thresholds, discarding vital nuance. Second, the "geodesic bias" ignores the "side streets" of the brain, focusing exclusively on the most direct pathways.

This research introduces a method to measure "communicability"—the ease with which information spreads across a network—while respecting the actual weights of connections and the contribution of every possible path between regions.

Methodology

The researchers extended the concept of communicability, originally proposed for unweighted graphs, into the realm of weighted networks using the matrix exponential. If $A$ is a matrix representing brain connections, raising it to the power of $k$ reveals the number of "walks" of length $k$ between regions. By summing these powers and penalizing longer walks with a $1/k!$ factor, a single value is derived representing the communicative ease between two nodes.

A critical component of this method is the normalization step. Because "promiscuous" nodes with many strong connections can skew results, connection weights are divided by the square root of the product of the nodes' degrees. This ensures the measure reflects the network's topology rather than just the raw strength of high-traffic hubs.

Experimental Setup

The study utilized a high-quality dataset consisting of 19 subjects: 9 stroke patients (at least six months post-stroke) and 10 age-matched healthy controls. The brain was partitioned into 56 anatomically distinct regions (48 cortical and 8 subcortical) using the Harvard-Oxford structural atlases. Connectivity was derived from diffusion MRI, which tracks the white matter tracts forming the brain's structural wiring.

Results and Discussion

The researchers employed unsupervised spectral clustering (specifically Singular Value Decomposition) to determine if the groups could be distinguished without prior labeling. The results demonstrated that communicability acts as a biological filter, reducing the influence of noise in raw MRI data and clarifying the underlying biological signal.

Figure 1: Comparison of spectral clustering results using raw, normalized, and communicability data. The communicability measure (right) provides a clear binary classification where stroke patients and controls have opposite signs.

As shown in Figure 1, the raw data failed to perfectly separate the groups, misidentifying two subjects. However, the communicability measure achieved perfect separation. In the communicability plot, stroke patients and controls ended up with opposite signs (negative vs. positive values), providing a clear, mathematically elegant binary classifier. Furthermore, the communicability approach identified 318 highly relevant connections, compared to only 266 found in the raw data, suggesting a higher sensitivity to the brain's functional architecture.

Conclusion

While the sample size of 19 subjects is small, it serves as a robust proof of principle for the communicability measure. The findings suggest that redundant or indirect pathways are crucial for understanding how the brain is affected by or recovers from injury. Future work will focus on scaling this measure to the microscale and using it to identify specific connections that change post-stroke, potentially aiding clinicians in predicting recovery outcomes and targeting rehabilitation.

Summary updated

Abstract

Recent advances in experimental neuroscience allow, for the first time, non-invasive studies of the white matter tracts in the human central nervous system, thus making available cutting-edge brain anatomical data describing these global connectivity patterns. This new, non-invasive, technique uses magnetic resonance imaging to construct a snap-shot of the cortical network within the living human brain. Here, we report on the initial success of a new weighted network communicability measure in distinguishing local and global differences between diseased patients and controls. This approach builds on recent advances in network science, where an underlying connectivity structure is used as a means to measure the ease with which information can flow between nodes. One advantage of our method is that it deals directly with the real-valued connectivity data, thereby avoiding the need to discretise the corresponding adjacency matrix, that is, to round weights up to 1 or down to 0, depending upon some threshold value. Experimental results indicate that the new approach is able to highlight biologically relevant features that are not immediately apparent from the raw connectivity data.
BibTeX
@misc{crofts2008communicability,
  author = {Jonathan J. Crofts and Desmond J. Higham},
  title = {{Communicability in complex brain networks}},
  year = {2008},
  eprint = {0811.0769},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/0811.0769},
}
Figure