White Matter Fiber Tract Segmentation
Using Nonnegative Matrix Factorization

Xuwei Liang(1), Jie Wang(2), Zhenmin Lin(1), Jun Zhang(1)
(1)Laboratory for Computational Medical Imaging & Data Analysis
Department of Computer Science
University of Kentucky
773 Anderson Tower
Lexington, KY 40506-0046, USA

(2)Department of Computer Science
Minnesota State University at Mankato
Mankato, MN 56001

Abstract

Accurate and efficient white matter fiber tract segmentation is an important step in clinical and anatomical studies that use diffusion tensor magnetic resonance imaging (DTI) tractography techniques. In this work, we present a novel technique to group white matter fiber tracts reconstructed from DTI into bundles using Nonnegative Matrix Factorization (NMF) of the frequency-tract matrix. A fiber tract is quantified by Fourier descriptors in terms of frequencies. Fourier descriptors derived from the shape signature, the central angle dot product, are used to construct the nonnegative frequency-tract matrix which is analogous to the term-document matrix in the document clustering context. In the NMF derived feature space, each basis vector captures the base shape of a particular fiber tract bundle. Each fiber tract is represented as an additive combination of the base shapes. The cluster label of each fiber tract is easily determined by finding the basis vector with which a fiber tract has the largest projection value. Preliminary experimental results with real DTI data show that this method efficiently groups tracts into plausible bundles. This indicates that NMF may be used in fiber tract segmentation with appropriate fiber tract encodings.


Key words: Diffusion tensor imaging, nonnegative matrix factorization, clustering, white matter fiber, Fourier descriptors

Mathematics Subject Classification:


Download the PDF file liang5.pdf.
Technical Report CMIDA-HiPSCCS 012-08, Department of Computer Science, University of Kentucky, Lexington, KY, 2008.


The authors acknowledge the support of funding agencies and the collaborators. The research work of J. Zhang was supported in part by the US National Science Foundation under grant CCF-0527967 and CCF-0727600, in part by the National Institutes of Health under grant 1R01HL086644-01, in part by the Kentucky Science and Engineering Foundation under grant KSEF-148-502-06-186, and in part by the Alzheimer's Association under grant NIRG-06-25460.