Fiber segmentation using constrained clustering

Abstract

In this work we introduce the novel concept of applying constraints into the fiber segmentation problem within a clustering based framework. The segmentation process is guided in an interactive manner. It allows the definition of relationships between individual and sets of fibers. These relationships are realized as pairwise linkage constraints to perform a constrained clustering. Furthermore, they can be refined iteratively, making the process of segmenting tracts quicker and more intuitive. The current implementation is based on a constrained threshold based clustering algorithm using the mean closest point distance as measure to estimate the similarity between fibers. The feasibility and the advantage of constrained clustering are demonstrated via segmentation of a set of specific tracts such as the cortico-spinal tracts and corpus collosum. © 2010 Springer-Verlag.