Cardiac fiber tracking using particle filtering in MRI

The study of myocardial fiber structure has great fundamental significance and clinical value for explaining the causes of various cardiovascular diseases and early diagnosis. Diffusion Tensor Image (DTI) is the first non-destructive fiber reconstruction technique for brain white matter and cardiac muscle. We present the use of particle filtering for tracking cardiac fibers in DTI. The method consists in formulating the uncertainty of fiber paths using a probabilistic state model. Fiber tracking is finished using a adaptive particle filtering framework by modeling all possible fiber orientations originating from a starting point as a distribution. A probability is then assigned to each orientation according to the measured diffusion tensor. The fiber orientation is derived from weighted samples. Von Mises-Fisher distribution is used to model the posterior distribution. We evaluated our method both on artificial dataset with different level noises and real cardiac DTI dataset. The results shows that the proposed method allows for fast and efficient sampling and that starting at a seed voxel, the optical fiber orientation can be computed rapidly.