Research on Feature Vector Dimensionality Reduction Method for Dynamic Functional Connectivity of ASD Children

Dynamic functional connectivity (dFC) analysis of resting-state fMRI data is useful for exploring the dynamic evolution patterns of Autism Spectrum Disorders (ASD) Children. While only few attention has been set on the individual differences in the process of dimensionality reduction of high-dimensional feature description, which is the basis of dynamic state change observation. A sample adaptive dimensionality reduction algorithm for dynamic functional connectivity features vector is proposed and applied to the state clustering study of ASD children’s dFC. In this method the distribution relationship of samples in high-dimensional space are measured firstly, then the Canopy algorithm is used to divide samples into three different groups with different weight values to identify the similarity between pairs of samples in high-dimensional space, finally a weighted low dimensional embedding algorithm is present to realize the dFC states clustering. The experiments show that compared with the traditional algorithm, both the visualization results and evaluation metrics with our method have been improved, the average DI value has been increased by 6.53%, the average DBI value has been reduced by 18.86%, and the average Silhouette Coefficient increased by 32.48%.