eduzhai > Applied Sciences > Engineering >

Flexible Bayesian Modelling for Nonlinear Image Registration

  • Save

... pages left unread,continue reading

Document pages: 13 pages

Abstract: We describe a diffeomorphic registration algorithm that allows groups ofimages to be accurately aligned to a common space, which we intend toincorporate into the SPM software. The idea is to perform inference in aprobabilistic graphical model that accounts for variability in both shape andappearance. The resulting framework is general and entirely unsupervised. Themodel is evaluated at inter-subject registration of 3D human brain scans. Here,the main modeling assumption is that individual anatomies can be generated bydeforming a latent average brain. The method is agnostic to imaging modalityand can be applied with no prior processing. We evaluate the algorithm usingfreely available, manually labelled datasets. In this validation we achievestate-of-the-art results, within reasonable runtimes, against previousstate-of-the-art widely used, inter-subject registration algorithms. On theunprocessed dataset, the increase in overlap score is over 17 . These resultsdemonstrate the benefits of using informative computational anatomy frameworksfor nonlinear registration.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×