Elsevier

NeuroImage

Volume 36, Issue 3, 1 July 2007, Pages 645-660
NeuroImage

Characterizing brain anatomical connections using diffusion weighted MRI and graph theory

https://doi.org/10.1016/j.neuroimage.2007.02.012Get rights and content

Abstract

A new methodology based on Diffusion Weighted Magnetic Resonance Imaging (DW-MRI) and Graph Theory is presented for characterizing the anatomical connections between brain gray matter areas. In a first step, brain voxels are modeled as nodes of a non-directed graph in which the weight of an arc linking two neighbor nodes is assumed to be proportional to the probability of being connected by nervous fibers. This probability is estimated by means of probabilistic tissue segmentation and intravoxel white matter orientational distribution function, obtained from anatomical MRI and DW-MRI, respectively. A new tractography algorithm for finding white matter routes is also introduced. This algorithm solves the most probable path problem between any two nodes, leading to the assessment of probabilistic brain anatomical connection maps. In a second step, for assessing anatomical connectivity between K gray matter structures, the previous graph is redefined as a K + 1 partite graph by partitioning the initial nodes set in K non-overlapped gray matter subsets and one subset clustering the remaining nodes. Three different measures are proposed for quantifying anatomical connections between any pair of gray matter subsets: Anatomical Connection Strength (ACS), Anatomical Connection Density (ACD) and Anatomical Connection Probability (ACP). This methodology was applied to both artificial and actual human data. Results show that nervous fiber pathways between some regions of interest were reconstructed correctly. Additionally, mean connectivity maps of ACS, ACD and ACP between 71 gray matter structures for five healthy subjects are presented.

Introduction

Random motion of water molecules inside the brain is influenced by the architectural properties of tissues. Water diffusion is known to be highly anisotropic in certain white matter regions, with preferential movement along the nervous fibers. A recent development of a non-invasive technique which quantifies water diffusion process, known as Diffusion Weighted Magnetic Resonance Imaging (DW-MRI), has allowed to obtain structural information about the intravoxel axon arrangement (Basser et al., 1994, LeBihan, 2003). Based on this information, fiber tractography arises as a crucial technique to attain a better in vivo anatomical characterization of the brain (Mori et al., 1999, Conturo et al., 1999, Tuch, 2002, Parker et al., 2002, Koch et al., 2002, Behrens et al., 2003). Also, quantification of the anatomical connectivity between different gray matter structures would be a significant contribution to the understanding of functional integration of the human brain (LeBihan et al., 2001, Koch et al., 2002, Ramnani et al., 2004, Sporns et al., 2005, Sotero et al., 2007).

Reconstruction of nervous fiber trajectories is an extensively treated topic. In the traditional Streamline Tractography (SLT) approach (Mori et al., 1999, Conturo et al., 1999, Basser et al., 2000), a continuous trajectory is traced tangential to the direction of the principal eigenvector of the diffusion tensor measured at each voxel using a discretization step smaller than the size of the voxel. This approach usually fails in voxels where fibers cross each other, merge, kiss or diverge, and it is very sensitive to the influence of MR signal noise (Basser and Pajevic, 2000, Lori et al., 2002). In those situations, traced path strays from the real trajectory of nervous fibers. To overcome these limitations, modified Streamline Tractography (mSLT) methods based on Diffusion Tensor Deflection (Weinstein et al., 1999, Lazar et al., 2003) and Probabilistic Monte-Carlo Method (Parker and Alexander, 2003) have been proposed. The former uses the entire diffusion tensor to deflect the propagation direction computed in the previous step. The latter uses the uncertainty of the estimated nervous fiber orientation to compute a large number of possible paths from the seed point; a quantity can be assigned to each path reflecting some connectivity relationship between seed and target points.

In recent years, many other mSLT methods have been proposed (Tuch, 2002, Tench et al., 2002, Behrens et al., 2003, Hagmann et al., 2003). Usually, they define the anatomical connection probability between seed and target voxels as the ratio between the number of shared paths and the number of generated paths.

In contrast to SLT and mSLT methods, Level Set-Based Fast Marching (FM) techniques (Parker et al., 2002, Staempfli et al., 2006) express the tractography in terms of a wave front that emanates from a source point and whose evolution is controlled by the diffusion data. FM methods have two advantages over the SLT and mSLT methods: 1, better performance in situations of branching and fiber crossing, and 2, direct estimation of the probability of white matter connectivity between two points (Jun Zhang et al., 2005).

In FM methods, front evolution speed and direction in a voxel depend on the measured diffusion tensor. Generally, all proposed FM algorithms have used only the principal eigenvector of the diffusion tensor, therefore these methods fail to reconstruct fiber pathways in those places where fibers cross, merge, kiss or diverge. For dealing with this limitation, recently Staempfli et al. (2006) proposed an advanced implementation of FM (aFM), combining the advantages of classical FM and the tensor deflection approach. The objective is to take into account the entire information contained in the diffusion tensor. As an intrinsic limitation, aFM needs an empirical threshold value to classify geometrically the diffusion tensor ellipsoid (i.e. prolate, oblate or spherical tensor) and therefore to set the corresponding speed function. Also, only four possible situations of voxel transitions are considered, which are those involving prolate and oblate cases. Thus, some combinations of more than two fibers crossing may be ignored.

On the other hand, although probability of connection between seed and target voxels has been previously used (Tuch, 2002, Parker et al., 2002, Koch et al., 2002, Parker et al., 2003, Behrens et al., 2003, Staempfli et al., 2006), the generalization of this concept to characterize anatomical connections between different brain gray matter structures is not straightforward. An initial approach (Iturria-Medina et al., 2005) was proposed to quantify the anatomical connection strength (ACS) between two gray matter structures using geometrical information from probabilistic fiber paths. ACS was considered proportional to the total area comprised by the fiber connector volume over the surfaces of the two connected structures. This was evaluated by counting the number of superficial voxels involved in the connection, where each voxel is weighted according to the validity of the paths that connect it with the second structure. A connectivity matrix estimated using the aforementioned approach was employed to couple several brain areas in a realistic neural mass model for the EEG generation, obtaining physiologically plausible results (Sotero et al., 2007).

In addition, recently Hagmann et al. (2006) proposed a technique based on graph theory to study the connectivity between small cortical areas. Nodes of a graph correspond to small cubic regions of interest (ROI) covering the brain gray matter. Fiber tractography is performed by initiating fibers over the whole brain and arc weight between any two ROIs is assigned according to the connection density between them. An unweighted version of this graph was constructed in order to analyze its small world and hierarchical properties.

In this work, our interest lies in the development of a DW-MRI-based methodology, capable of characterizing directly anatomical connections between brain gray matter structures, which can be defined according to cytoarchitectonic, histological or other sort of anatomical and functional information. In order to accomplish this, the graph framework is employed to introduce a new anatomical connectivity model. Firstly, each voxel of the cerebral volume is assumed to be a node of a non-directed weighted graph. In this case, the weight of an arc is considered to be proportional to the probability of the existence of a nervous fiber connecting its corresponding nodes. Probabilistic tissue segmentation and intravoxel white matter orientational distribution function (ODF) are combined to compute the arc weight. Secondly, an iterative algorithm is used to solve the most probable path problem between any two nodes in the graph, which we will indistinctly refer to as the most reliable connection route between these nodes. This approach allows to asses probabilistic anatomical connectivity maps between brain voxels. Finally, in order to assessing anatomical connectivity between K gray matter structures, the graph is partitioned in the corresponding K non-overlapped subsets and one subset containing the remaining nodes. This allowed for the definition of three different anatomical connectivity measures between any pair of gray matter structures: Anatomical Connection Strength (ACS), Anatomical Connection Density (ACD) and Anatomical Connection Probability (ACP).

Section snippets

Methods

This section will be devoted to present some basic elements of graph theory, as well as the principal steps of the proposed methodology: 1, definition of a Brain Graph, 2, introduction of an iterative fiber tracking algorithm and quantification of node–node connectivity and 3, definition of anatomical connectivity measures between gray matter areas. Details on experimental data to be used and its preprocessing will also appear.

Artificial data

The performance of the proposed methodology was explored using artificial DW-MRI data. Firstly, anatomical connections between different regions of interest (ROIs), defined for the straight crossing, curve crossing and Feuille trajectory were estimated (see Fig. 7). Each ROI consists of seven contiguous voxels representing a line perpendicular to the fiber tract being analyzed. Left column (Figs. 7a, c and e) shows the reconstructed connection routes (most probable paths) between ROI1 and the

Summary and discussions

In this work, we developed a diffusion imaging methodology capable to characterize anatomical connections between different brain gray matter structures. It consists of three basic steps: 1, definition of a Brain Graph model in which each voxel is considered as a node of a non-directed weighted graph; 2, the use of an iterative algorithm based on analysis of the voxels neighborhood to find the route of maximum probability between two nodes and the subsequent definition of the anatomical

Acknowledgments

Authors would like to thank Nelson Rubal Lorenzo, Agustín Lage-Castellanos and Erick C. Jones for their useful comments on the anatomical, statistical and algorithmic aspects of this work, respectively. Also, we thank to Roberto C. Sotero and Thomas Koenig, for the final revision of the manuscript.

References (51)

  • D.C. Alexander et al.

    Spatial transformations of diffusion tensor magnetic resonance images

    IEEE Trans. Med. Imag.

    (2001)
  • J. Ashburner et al.

    Nonlinear spatial normalization using basis functions

    Hum. Brain Mapp.

    (1999)
  • P.J. Basser et al.

    Statistical artifacts in diffusion tensor MRI (DT-MRI) caused by background noise

    Magn. Reson. Med.

    (2000)
  • P.J. Basser et al.

    In vivo fiber tractography using DT-MRI data

    Magn. Reson. Med.

    (2000)
  • T.E. Behrens et al.

    Non-invasive mapping of connections between human thalamus and cortex using diffusion imaging

    Nat. Neurosci.

    (2003)
  • D.L. Collins et al.

    Automatic 3D intersubject registration of MR volumetric data in standardized Talairach space

    J. Comput. Assist. Tomogr.

    (1994)
  • T.E. Conturo et al.

    Tracking neuronal fiber pathways in the living human brain

    Proc. Natl. Acad. Sci. U. S. A.

    (1999)
  • E.W. Dijkstra

    A note on two problems in connexion with graphs

    Numer. Math.

    (1959)
  • B. Efron

    Large-scale simultaneous hypothesis testing: the choice of a null hypothesis

    J. Am. Stat. Assoc.

    (2004)
  • Efron, B., 2005. Local false discovery rates....
  • Evans, A.C., Collins, D.L., 1993. A 305-member MRI-based stereotactic atlas for CBF activation studies. Proceedings of...
  • A.C. Evans et al.

    Three-dimensional correlative imaging: applications in human brain mapping

  • I. Gómez-Padrón et al.

    Anatomía II. Folleto complementario

    (1985)
  • M. Gondran et al.

    Graphs and Algorithms

    (1984)
  • P. Hagmann et al.

    Imaging the brain neuronal network with diffusion MRI: a way to understand its global architecture

    Proc. ISMRM

    (2006)
  • Cited by (306)

    View all citing articles on Scopus
    View full text