Visualization of uncertainty in fiber tracking based on diffusion tensor imaging

Diffusion tensor imaging (DTI) is an imaging technique based on magnetic resonance that describes, in each point of the tissue, the distribution of diffusing water molecules. The distribution is mathematically modelled using a second-order tensor. In fibrous tissues the diffusion tensor will have an elongated, ellipsoid shape whose main axis is assumed to be aligned with the underlying fiber structure. Fiber tractography traces paths through the tensor field by following each tensor's main direction thereby resulting in a three-dimensional reconstruction of the fibers. This is particularly interesting for the exploration and visualization of neuronal connections in brain white matter and has great potential for applications in neuroscience and neurosurgery. DTI and fiber tractography are unique in that they provide insight into white matter structures in vivo and non-invasively. However, despite these capabilities the application of DTI and fiber tractography in clinical practice remains limited. The image acquisition and post-processing pipeline is complex and consists of many stages. At each stage errors and uncertainties are introduced due to image noise, magnetic distortions, partial volume effects, scanner settings, diffusion model assumptions and user parameters. These uncertainties are propagated through the pipeline and possibly enhanced in subsequent stages thereby leading to potentially unreliable results in the final tractography output. To the user the processing pipeline behaves like a black box whose internal details remain hidden and whose quality of output cannot be reliably assessed. Contrary to standard CT and MR images it is not possible to look at the "raw" diffusion-weighted images. Without further processing the images are practically meaningless. This means the user either has to accept (and trust) the processing output or refrain from using fiber tracking all together. In this thesis we assume that the user has certain reservations about the quality of the tractography output. Unfortunately, there is no gold standard against which the output of tractography can be validated. Consequently, we cannot make definitive statements about the "true" certainty or uncertainty of fiber reconstructions. We can, however, discuss tractography output in terms of stability and reproducibility. The output of tractography algorithms can be subject to large variations. In this thesis we present a number of visualization strategies that make these variations visible to the user and allow a better assessment of the reliability of fiber reconstructions obtained from any given tractography algorithm.