White Matter Bundle Registration and Population Analysis Based on Gaussian Processes

D. Wasserman, Y. Rathi, S. Bouix, M. Kubicki, R. Kikinis, M. E. Shenton, C.F. Westin
Inf Process Med Imaging
Volume 22, Pages 320-332

Download full paper


This paper proposes a method for the registration of white matter tract bundles traced from diffusion images and its extension to atlas generation, Our framework is based on a Gaussian process representation of tract density maps. Such a representation avoids the need for point-to-point correspondences, is robust to tract interruptions and reconnections and seamlessly handles the comparison and combination of white matter tract bundles. Moreover, being a parametric model, this approach has the potential to be defined in the Gaussian processes' parameter space, without the need for resampling the fiber bundles during the registration process. We use the similarity measure of our Gaussian process framework, which is in fact an inner product, to drive a diffeomorphic registration algorithm between two sets of homologous bundles which is not biased by point-to-point correspondences or the parametrization of the tracts. We estimate a dense deformation of the underlying white matter using the bundles as anatomical landmarks and obtain a population atlas of those fiber bundles. Finally we test our results in several different bundles obtained from in-vivo data.


Wasserman D, Rathi Y, Bouix S, Kubicki M, Kikinis R, Shenton ME, Westin C. White matter bundle registration and population analysis based on gaussian processes. Inf Process Med Imaging 2011;22:320-332.


R01 MH82918, P41 RR13218, R01 MH074794, R01 MH092862, R01 MH5074, P41 RR019703, U54 EB005149

Research area

© 2013 Psychiatry Neuroimaging Laboratory | Last updated 04.15.2013