Fine Brain Functional Parcellation Based on t-Distribution Stochastic Neighbor Embedding and Automatic Spectral Clustering

Abstract

In this paper, a new method for fine brain functional parcellation based on resting-state functional magnetic resonance imaging (rs-fMRI) data was proposed. The method combines the t-distribution stochastic neighbor embedding (t-SNE) and automatic spectral clustering (ASC) algorithms. First, correlation analyses are conducted between the time courses of the brain region to be parcellated and the whole brain. Second, t-SNE is used to extract the high-dimensional functional connectivity patterns. Last, the number of clusters is automatically determined by the ASC algorithm, and to divide the brain region of interest to generate the fine brain subregions. The results of simulated seed regions proved that the method proposed had higher accuracy than the commonly-used spectral clustering and spectral clustering with principal component analysis. Moreover, the method was successfully applied to parcellate the parahippocampal gyrus into 3 functional subregions in the left and right hemispheres. In conclusion, the algorithm combining t-SNE and ASC is an effective method for fine brain functional parcellation and construction of functional brain atlas.

Keywords

• resting-state fmri
• functional connectivity
• functional parcellation
• t-sne
• asc