Distributed functional connectivity network method for identifying autism spectrum disorder

Functional connectivity network (FCN) has been becoming an increasingly effective tool for exploring spontaneous brain activity and changes in brain function for the diagnosis of the autism spectrum disorder (ASD). The method of constructing FCN based on Matrix Variate Normal Distribution theory (MVND method) is one of the most effective methods recently proposed, which integrates the dynamic and high-order properties of the brain. However, in practice, the problem of too many parameters and insufficient samples makes the quality of feature extraction still have room for further improvement. To address this issue, we propose a distributed FCN method. Firstly, through k-means clustering we classify all rs-fMRI time series into several clusters. In each cluster, the MVND method is used to construct low- and high-order FCNs. For the loss of functional connectivity information caused by cluster division, we propose to use the paired Pearson Correlation of rs-fMRI mean time series of each cluster to make up. We verify the effectiveness of the distributed FCN method in the classification experiments of ASD subjects and normal controls (NCs). The experimental results show that the distributed FCN method achieves higher classification accuracy than MVND method, and the highest recognition rate is 78.26%.