DT-MRI White Matter Fiber Tractography with Global Constraints: An Unsupervised Learning Approach

Brain white matter fiber tracking imaging using diffusion tensor magnetic resonance imaging (DT-MRI) traces brain white matter fiber bundle and reconstruct the structures of the fibers according to the diffusion of water molecular in the white matter. In this paper, a novel fiber tracking technique based on well established unsupervised learning algorithms was proposed. For a pair of regions of interest (ROIs), a random fiber pathway that connect both ROIs are generated initially. This pathway is evaluated for fitness to the diffusion tensor field and fiber geometric with global constraints. Then another random fiber pathway was generated and compared with the former one. Training was done according to the fitness between the two fiber and weights was renewed to generate new fiber. These processes are iterated until convergence to get a deterministic tracking result and three dimensional white matter fiber structure can get from the multiple results. This method was applied to a synthetic dataset and two sets of in vivo DTI data acquired from different healthy human volunteers. The experiments demonstrate that the fiber tracking algorithm we proposed can reconstruct white matter fiber trajectories faithfully for both synthetic and in vivo DTI data and is in-susceptible to image noise and other local artifacts.