Synopsis
Brain networks have a modular structure in function and this modular organization changes dynamically with time. However, the available brain network modularization methods are limited to detect functional modules for static brain network.So the aim of this sttudy is to propose a consensus algorithm to evaluate the modularization of functional brain network of healthy individuals and patients with Alzheimer's disease.The measures for the modularization of functional brain networks included normalized mutual information (NMI), average participation coefficient and classification accuracy.To evaluate the results from the consensus algorithm, machine learning based classification was performed on a new AD data sets.
Purpose
To proposed a consensus modularization approach to detect functional modules for dynamic brain network and evaluated the algorithm by machine learning based classification. Methods
A sample set of 50 adults, including 25 normal controls and 25 AD subjects was drawn from the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset. For each subject, resting state fMRI images were obtained. The raw fMRI data were preprocessed with SPM8 and DPARSF software package. Then the whole brain was partitioned into 90 regions using AAL atlas to obtain the brain static network. For each subject, we constructed a series of time-dependent brain subnetworks from resting-state functional MRI by sliding window, and then adopted a consensus algorithm to extract dynamic information and integrate multiple partitions corresponding to dynamic subnetworks into an optimized modularization partition 1,2. Based on the approach proposed, we analyzed the modularization of functional brain network of healthy individuals and patients with Alzheimer's disease (AD). To evaluate the results from the consensus algorithm, machine learning based classification was performed on a new AD data sets.
Results
At subject level, evaluations in normalized mutual information (NMI) suggested that approach proposed can get a consensus modular structure of brain network which contains dynamic variations of modular organization over time. On the other hand, analysis of network properties and simulations of module attack demonstrated that consensus modularization results are more in accordance with the mechanism of functional segregation and integration of the brain. In addition, AD classification results based on features from modularization also verified that the consensus modularization result did describe the key characteristics of modular organizations in AD. In group analysis, the results indicated that the consensus algorithm can extract more consistent modular information over individuals with differences and get a concordant and rational group-level modular pattern structure.
Conclusion
The consensus modularization approach we proposed could capture dynamic
brain network information and can be taken as an effective dynamic modular
method. Consensus modularization approach
could be used to detecting dynamic community structure of brain network in tracking
Alzheimer's disease-related functional reorganization.
Acknowledgements
No acknowledgement found.References
1, Newman
ME. Modularity and community structure in networks. Proc Natl Acad Sci U S A
2006;103(23):8577-8582.
2, Lancichinetti
A, Fortunato S. Consensus clustering in complex networks. Sci Rep 2012;2:336.
