A Comparative Study of the Effect of Weighted or Binary Functional Brain Networks in fMRI Data Analysis
Purpose: Graph theory is a widely used and reliable tool to quantify brain connectivity. Brain functional connectivity is modeled as graph edges employing correlation coefficients. The correlation coefficients can be used as the weight that shows the power of connectivity between two nodes or can be binarized to show the existence of a connection regardless of its strength. To binarize the brain graph two approaches, namely fixed threshold and fixed density are often used.
Materials and Methods: This paper aims to investigate the difference between weighted or binarized graphs in brain functional connectivity analysis. To achieve this goal, the brain connectivity matrices are generated employing the functional Magnetic Resonance Imaging (fMRI) data of Alzheimer's Disease (AD). After preprocessing the data, weighted and binarized connectivity matrices are constructed using a fixed threshold and fixed density techniques. Graph global features are extracted and a non-parametric statistical test is performed to analyze the performance of the methods.
Results: Results show that all three methods are powerful in distinguishing the healthy group from AD subjects. The P-Values of the weighted graph is close to the fixed threshold method.
Conclusion: Also, it is worthwhile mentioning that the fixed threshold method is robust in changing the threshold while the fixed density method is very sensitive. On the other hand, graph global measures such as clustering coefficient and transitivity, regardless of the method, show significant differences between the control and AD groups. Furthermore, the P-Values of modularity measure are very varied according to the method and the selected threshold.
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.