关键词: 半监督聚类, 脑区分割, 功能磁共振图像, 功能连接

Abstract:

Resting state functional magnetic resonance imaging (fMRI) data have been increasingly used for identifying functional subunits of the human brain by clustering image voxels using algorithms such as k-means and spectral clustering.However,such unsupervised clustering methods are sensitive to imaging noise,thus generating unstable brain parcellation results.In this study,we present a prior information guided,semi-supervised clustering method to identify reliable boundaries between functional subunits for achieving robust brain parcellation.The proposed method has been validated for parcellation of Broca area (Brodmann areas 44 and 45) in the right hemisphere of the human brain based on resting state fMRI data.Experimental results demonstrate that our method not only identifies more reliable boundaries between functional subunits,but also achieves higher reproducibility across subjects.

Key words: semi-supervised clustering;brain parcellation;fMRI;functional connectivity