Globally optimized fiber tracking and hierarchical clustering -- a unified framework

Abstract

Structural connectivity between cortical regions of the human brain can be characterized noninvasively with diffusion tensor imaging (DTI)-based fiber tractography. In this paper, a novel fiber tractography technique, globally optimized fiber tracking and hierarchical fiber clustering, is presented. The proposed technique uses k-means clustering in conjunction with modified Hubert statistic to partition fiber pathways, which are evaluated with simultaneous consideration of consistency with underlying DTI data and smoothness of fiber courses in the sense of global optimality, into individual anatomically coherent fiber bundles. In each resulting bundle, fibers are sampled, perturbed and clustered iteratively to approach the optimal solution. The global optimality allows the proposed technique to resist local image artifacts and to possess inherent capabilities of handling complex fiber structures and tracking fibers between gray matter regions. The embedded hierarchical clustering allows multiple fiber bundles between a pair of seed regions to be naturally reconstructed and partitioned. The integration of globally optimized tracking and hierarchical clustering greatly benefits applications of DTI-based fiber tractography to clinical studies, particularly to studies of structure-function relations of the complex neural network of the human. Experiments with synthetic and in vivo human DTI data have demonstrated the effectiveness of the proposed technique in tracking complex fiber structures, thus proving its significant advantages over traditionally used streamline fiber tractography.