Gray matter parcellation constrained full brain fiber bundling with diffusion tensor imaging

Authors

  • Xu Qing,

    1. Vanderbilt University Institute of Imaging Science, Vanderbilt University, Nashville, Tennessee 37232-2310 and Department of Electrical Engineering and Computer Science, Vanderbilt University, Nashville, Tennessee 37235-1826
    Search for more papers by this author
  • Anderson Adam W.,

    1. Vanderbilt University Institute of Imaging Science, Vanderbilt University, Nashville, Tennessee 37232-2310 and Department of Biomedical Engineering, Vanderbilt University, Nashville, Tennessee 37235-1631
    Search for more papers by this author
  • Gore John C.,

    1. Vanderbilt University Institute of Imaging Science, Vanderbilt University, Nashville, Tennessee 37232-2310 and Department of Biomedical Engineering, Vanderbilt University, Nashville, Tennessee 37235-1631
    Search for more papers by this author
  • Ding Zhaohua

    1. Vanderbilt University Institute of Imaging Science, Vanderbilt University, Nashville, Tennessee 37232-2310; Department of Electrical Engineering and Computer Science, Vanderbilt University, Nashville, Tennessee 37235-1826; and Department of Biomedical Engineering, Vanderbilt University, Nashville, Tennessee 37235-1631
    Search for more papers by this author

Abstract

Purpose:

Studying white matter fibers from diffusion tensor imaging (DTI) often requires them to be grouped into bundles that correspond to coherent anatomic structures, particularly bundles that connect cortical/subcortical basic units. However, traditional fiber clustering algorithms usually generate bundles with poor anatomic correspondence as they do not incorporate brain anatomic information into the clustering process. On the other hand, image registration-based bundling methods segment fiber bundles by referring to a coregistered atlas or template with prelabeled anatomic information, but these approaches suffer from the uncertainties introduced from misregistration and fiber tracking errors and thus the resulting bundles usually have poor coherence. In this work, a bundling algorithm is proposed to overcome the above issues.

Methods:

The proposed algorithm combines clustering- and registration-based approaches so that the bundle coherence and the consistency with brain anatomy are simultaneously achieved. Moreover, based on this framework, a groupwise fiber bundling method is further proposed to leverage a group of DTI data for reducing the effect of the uncertainties in a single DTI data set and improving cross-subject bundle consistency.

Results:

Using the Montreal Neurological Institute template, the proposed methods are applied to building a full brain bundle network that connects cortical/subcortical basic units. Based on several proposed metrics, the resulting bundles show promising bundle coherence and anatomic consistency as well as improved cross-subject consistency for the groupwise bundling.

Conclusions:

A fiber bundling algorithm has been proposed in this paper to cluster a set of whole brain fibers into coherent bundles that are consistent to the brain anatomy.

Ancillary