Similarity Measure for Fiber Clustering: A Constant Time Complexity Algorithm

Recently, fiber clustering algorithms have become an important tool in neuroscience for grouping the white matter tracts into anatomically meaningful bundles. The results of clustering can be used for quantification and comparison between different brains to find out abnormalities or unusual features. One essential problem in fiber clustering is to provide a similarity measure for a pair of fibers. Although there are many methods proposed in literature, most of them suffer from high time complexity, which causes difficulty in dealing with very large fiber data sets. In this paper, we propose a new flexible fibers similarity measure, which is only based on center of mass, start and end point of each fiber to capture the shape and distance similarity between them with only O(1) time complexity. Our new method is used together with a density-based clustering algorithm to segment fibers into groups. Experiments on real data sets prove the efficiency and effectiveness of our approach in comparison with other distance-based techniques like namely Dynamic Time Warping (DTW), Mean of Closest Point (MCP) distance and Hausdorf (HDD) distance.