A Similarity Model and Segmentation Algorithm for White Matter Fiber Tracts

Abstract

Recently, fiber segmentation has become an emerging technique in neuroscience. Grouping fiber tracts into anatomical meaningful bundles allows to study the structure of the brain and to investigate onset and progression of neurodegenerative and mental diseases. In this paper, we propose a novel technique for fiber tracts based on shape similarity and connection similarity. For shape similarity, we propose some new techniques adapted from existing similarity measures for trajectory data. We also propose a new technique called Warped Longest Common Subsequence (WLCS) for which we additionally developed a lower-bounding distance function to speed up the segmentation process. Our segmentation is based on an outlier-robust density-based clustering algorithm. Extensive experiments on diffusion tensor images demonstrate the efficiency and effectiveness of our technique.