Complex Brain Networks: Applications and Issues
The brain can be regarded as a network: nodes represent different specialized regions, and links represent communication pathways between the regions. In his recent talk at CNS, Mario Chavez, a professor at the Centre National de la Recherche Scientifique (CNRS), gave the audience an introduction of the recent progress of applying graph analysis techniques to quantify brain dysfunctions. The rationale behind this line of study is the following: given the evidence that several brain diseases lead to aberrant reconfiguration of functional brain network, a graph analysis of the brain network might provide clinical insights into the diseases.
Mapping the brain to a network is a fundamental yet tricky process. The functional brain network is constructed following a processing pipeline as illustrated by Mario in Figure 1. It’s worth noting that the measurement or calculation method used in each step in the pipeline affects the final network structure and therefore has to be chosen with caution. The mapping of brain regions to nodes can be done using different modality-specific methods: voxel-based modalities such as fMRI define nodes in the measurement space (after image reconstruction), while sensor-based modalities such as EEG, MEG can choose to assign nodes to sensors or to reconstructed sources. After defining brain nodes, links are given by evaluating similarities between two brain nodes signals. Again there is no unique way to define the links (depending on method, the link can be directed or undirected), but they all follow the functional connectivity (FC) method: given a set of signals from brain nodes defined previously, the magnitude of statistical interaction between the signals gives the weight of the link between the nodes. As the next step, graph filtering is applied to filter out weak links which may due to noise in volume conduction process, and finally, we arrive at the functional brain network.
Properly selected topological metrics of the functional brain network can give interesting clinical insights into several brain diseases. At the large topological scale, stroke, spinal cord injuries and Alzheimer’s disease significantly alters the integration and segregation of information between brain regions which are supported by small world organization of the brain network. In other words, it seems that healthy brains are more ‘small world’ than brains with the above diseases. At the intermediate scale, studies show that the brain graph of epileptic patients suffering from absence seizures (e.g. stroke) have a more regular modular organization compared with healthy ones. At small scale, centrality, for example, is a good measure to study brain syndromes with disconnection effects. Another advantage of these topological metrics is that they provide an objective measure of the brain disease. For patients with Alzheimer’s disease, for example, doctors could get information of their recovering process from the structure of the brain network, instead of asking subjective questions such as ‘Are you talking better?’
An interesting part of the talk was that Mario kept reminding us of the unaddressed issues of the brain network approach despite its strong impact in neuroscience. One category of issues lies in the construction of the network. How to better deal with signal noise? How to tackle the confounding effect of different brain diseases? How to better represent negative links? Another set of issues concerns computation methods. Do we really need overcomplicated measures when a much easier one could do the job? Is the controllability index a true measure of the system’s controllability or is it just the result of calculation bugs in Matlab? As Mario pointed out, we should not blindly borrow the concepts of network science unless we have the knowledge of the neural phenomenon under study, and are clear about all the methodological steps of the pipeline that manipulate the input brain signals and extract the functional network properties.
Fallani, Fabrizio De Vico, et al. "Graph analysis of functional brain networks: practical issues in translational neuroscience." Phil. Trans. R. Soc. B 369.1653 (2014): 20130521.
Blog post by Manran Zhu