Analysis of fiber similarity metric for fiber tract clustering in white matter of human brain

Abstract

MR diffusion tensor imaging fiber tracking (DTI-FT) could lead to false propagation because of image noise and low resolution of DTI. A variety of fiber clustering (FC) approaches have been used to refine the tract configuration. The measure of fiber similarity metric is vital in the FC algorithms. In this paper, two most common used fiber similarity metrics were compared. We firstly reconstructed one fiber tract; and then the fiber tract was mapped to a point by kernel principal component analysis (KPCA) with the two fiber similarity metrics, respectively; finally, the point cloud of fiber tract was clustered by hierarchical clustering which could could distinguish false fibers from true fibers. For evaluating our results, the corticospinal tract (CST) of a healthy human data in vivo was chosen. Our experiment showed that two factors fiber similarity metric could obviously decrease wrong propagated fibers than one factor fiber similarity metric. In conclusion, the selection of fiber similarity metric is vital in FC algorithms and it could refine the fiber tract configuration.