VISUALIZING DTI PARAMETERS ON BOUNDARY SURFACES OF WHITE MATTER FIBER BUNDLES Mathias Goldau 1,4 , Alexander Wiebel 2 , Mario Hlawitschka 3 , Gerik Scheuermann 1 , Marc Tittgemeyer 4 1 Universit¨ at Leipzig, Germany, E-mail: {math,scheuermann}@informatik.uni-leipzig.de. 2 Max Planck Institute for Human Cognitive and Brain Sciences, Germany, E-mail: wiebel@cbs.mpg.de. 3 University of California, Davis, USA, E-mail: hlawitschka@ucdavis.edu. 4 Max Planck Institute for Neurological Research, Germany, E-Mail: tittgemeyer@nf.mpg.de. ABSTRACT Diffusion magnetic resonance imaging is so far the only medical imaging modality that has the potential for prob- ing anatomical brain connectivity in vivo. Specifically, it provides the data basis for a set of techniques allowing for tracking of fiber bundles in the brain’s white matter. Fur- thermore, due to the micro-structural basis of the diffusion process, fiber integrity might be estimated. Typically, this is achieved by tensor-derived parameters, such as by frac- tional anisotropy (FA), which allows for a quantification of the directionality of local diffusion properties. In neuro- science, such parameterization of the diffusion tensor has greatly stimulated studies of localized brain changes, re- lated to development, aging, or various neurological and psychiatric diseases. However, thus far, there is no satisfac- tory solution for the visualization and assessment of such parameters along fiber bundles. In this paper, we present a novel technique to visual- ize changes of tensor-derived parameters along clusters of the trajectories obtained from diffusion tractography. This visualization approach consists of two steps: First, an auto- matic local aggregation of data values around the trajecto- ries for quantitative analysis and visualization on the fiber bundle boundary and second, a color-coded slice that is in- tuitively movable along the bundle for interactive explo- ration of the bundle’s parameters. KEY WORDS Medical Imaging, Surface Modeling, Fiber Bundles, diffu- sion MRI 1 Introduction Magnetic resonance imaging (MRI) produces detailed in vivo images of biological tissues allowing for new insights into structure and functionality of the brain. Specifically diffusion MRI, an MRI method based on diffusion of wa- ter molecules at discrete locations in space, enables us to study the outline and integrity of the brain’s fiber anatomy. The outline of specific fiber bundles is usually acquired us- ing diffusion tractographic methods [3, 22]. For visualiza- tion, streamlines connecting the main eigenvector from the second-order diffusion tensor [1] at each image voxel are taken as representatives for fiber tracts. However, due to the often complex topology of the brain’s fiber system, stream- line tractography can only be taken as an approximation to fiber anatomy – yet it is the only means to straightforwardly visualize fiber bundles in vivo. Besides the segmentation of fiber bundles, this is not the only information that can be inferred from diffusion MRI. As the diffusion process is clearly bound to the tis- sue microstructure, parameters reflecting microstructural properties might be inferred. This is usually done by an eigenvalue analysis of the diffusion tensor, and as such three values have been intensively studied in the past: axial diffusivity (largest eigenvalue of the diffusion tensor), ra- dial diffusivity (mean of median and minor eigenvalues), and fractional anisotropy [2]. While axial diffusivity is an indicator for axonal injury, and radial diffusivity is par- ticularly linked to demyelination [6, 17, 27, 31], fractional anisotropy is a rather general measure of the directional- ity of the diffusion process. However, it has been shown to rather sensitively indicate structural integrity and was therefore applied in numerous studies investigating struc- tural brain changes that are related to neurological and psy- chiatric diseases [14, 18], brain development [23], and ag- ing [24]. In order to relate parameters and their characteris- tics to such brain changes, neuroscientists perform inter- or intra-individual statistical comparisons. Such a compar- ison is not possible on the basis of single diffusion trac- tograms as these may vary strongly between subjects or measurements, in general far more than actual bundles do. The ultimate goal of our approach are longitudinal studies, i.e., intra-subject studies, when alteration of brain struc- ture plays a major role (e.g. due to aging). Even here, single tractograms become meaningless as they cannot be identified reliably in different data sets of the same sub- ject due to acquisition noise and their numerical calcula- tion. In cross-subject comparisons, due to the usually large morphological variability of the seeding regions, there is no reliable way to define a direct correspondence between tractograms. Conversely, large-scale structures like fiber bundles have been demonstrated to correspond reliably be- tween subjects. Hence, bundle representations will serve as basis for our visualization and analysis approach.