In multiple sclerosis research, the analysis of microstructural white matter properties has become increasingly important [1]. There are currently promising techniques like diffusion weighted MRI or magnetization transfer imaging that are sensitive to white matter alterations that cannot be assessed by conventional MRI [2]. However, the definition of anatomical based regions of interest remained elusive for the white matter and thus a region specific analysis of the white matter. While the parcellation of the human cortex into functionally differentiable areas can be easily performed on the basis of cortical gyri and sulci, there are no macro-anatomical landmarks that permit such a classification for the white matter. We present an approach that permits a fiber association based definition of white matter regions of interest.

We are using tools from the FreeSurfer software collection as well as self developed software to generate reliable cortical and subcortical labels by employing a combination of high-resolution T1 and T2 weighted images. In the present case, the labels were produced by employing the FreeSurfer Destrieux atlas. By using these labels as seed points for a probabilistic fiber tracking approach we derive regions of interest that outline the major projection pathways. The probabilistic fiber tracking approach is written in Rust [3] with self-developed medical image libraries for computation and NIfTI image handling. The fibers are propagated through the voxels by choosing a random direction with a probability proportional to the length of the three eigenvectors of the corresponding diffusion tensor. We then apply morphological operations (opening, closing) to these resulting regions of interest to eliminate outliers and produce reliable white matter labels.

In less than one second, one million fibers were propagated out of every seed point and were stopped by reaching a voxel that satisfies one of the stopping criteria (FA threshold, angle) or reaching one of the end points. An example fiber tracking result is shown in Figure 1, while Figure 2 gives an example of a typical result for an automatically derived ROI. The generated region of interest permits a region specific analysis of measures estimated by DWI, magnetization transfer, or any other quantitative MRI technique.Recent studies on patients with multiple sclerosis demonstrated thalamic atrophy even in the earliest stage of the disease [4]. However, until now it remained unclear to which degree the thalamic volume loss is associated with modality specific white matter alterations. The presented approach might be a promising tool for clinical and neuroscience research to investigate modalities or system specific microstructural alterations of white matter areas in a quantitative manner.