A neighborhood-based probabilistic approach for fiber tracking in human cardiac DTI

Fiber tracking techniques are essential for representing and visualizing the cardiac fiber architecture information encoded in diffusion-tensor imaging (DTI) data. We propose a neighborhood-based probabilistic fiber tracking method for cardiac DTI which accounts for spatial correlation and data uncertainty. The method consists in tracking fiber paths by sampling step directions from a normalized weighted sum of local fiber orientation probability distributions. The sum is over a predefined neighborhood of the current position in the fiber path, and the weights depend on fractional anisotropy, angular change and distance to the current position. Experimental results on synthetic data and on real human cardiac data show that, compared to the streamlining approach, the proposed method is meaningfully more robust to noise and produces smoother and more consistent fibers.