False Positive Detection using Filtered Tractography

Introduction: Diffusion-weighted MR imaging allows for non-invasive investigation of the neural architecture of the brain. In the past decade, several algorithms have been proposed to trace the fiber bundles using a variety of fiber model representations. The simplest and the most widely used model is the diffusion tensor model, with tracts generated by following the principal diffusion direction of the tensor. Other complex multitensor and non-parametric models have also been proposed. However, most of these methods estimate the model independently at each voxel, and tractography is done as a post-processing step. In their recent work [Malcolm et. al. 2009], the authors proposed an unscented Kalman filter based tractography algorithm. In this method, the model parameters (one or multi-tensor) are simultaneously estimated in a recursive fashion as we follow the fiber path. Thus, the inherent correlation in diffusion is taken into account in the model estimation step while tracing the fiber bundles. In this work, we show how false positive detection (fibers that don't exist anatomically, but are nevertheless traced by tractography algorithms) can be done naturally within this filtering framework.