Abstract
Complex network analysis (CNA), a subset of graph theory, is an emerging approach to the analysis of functional connectivity in the brain, allowing quantitative assessment of network properties such as functional segregation, integration, resilience, and centrality. Here, we show how a classification framework complements complex network analysis by providing an efficient and objective means of selecting the best network model characterizing given functional connectivity data. We describe a novel kernel-sum learning approach, block diagonal optimization (BDopt), which can be applied to CNA features to single out graph-theoretic characteristics and/or anatomical regions of interest underlying discrimination, while mitigating problems of multiple comparisons. As a proof of concept for the method’s applicability to future neurodiagnostics, we apply BDopt classification to two resting state fMRI data sets: a trait (between-subjects) classification of patients with schizophrenia vs. controls, and a state (within-subjects) classification of wake vs. sleep, demonstrating powerful discriminant accuracy for the proposed framework.
Highlights
Recent years have seen a growing interest in the analysis of functional connectivity [1] data resulting from brain mapping techniques such as fMRI
We compared the performance of several models - block diagonal optimization (BDopt), recursive composite kernels (RCK; [35]), recursive feature elimination (RFE, [34]), and standard support vector machine (SVM) classification - on both global and local Complex network analysis (CNA) features
We present a complete classification framework for conducting complex network analysis, permitting the flexibility afforded by various network measures, without the loss of power resulting from multiple comparisons
Summary
Recent years have seen a growing interest in the analysis of functional connectivity [1] data resulting from brain mapping techniques such as fMRI. Complex network analysis (CNA), a subset of graph theory that focuses on the topologically complex networks often found in nature, has proved to be a powerful approach to quantifying important features of functional connectivity. These include general network properties such as its functional segregation, integration, resilience, and centrality [2], as well as quantifying the contribution of individual brain regions to the network at large. CNA is performed on graphs, which topologically represent a matrix of functional connectivity. Graph representations are obtained after excluding negative and auto-connections and are thresholded to retain only strongest connections, sometimes to the point of binarization
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
