Prediction Correction Tractography through Statistical Tracking

In this work we describe a novel approach to diffusion tractography that is a notion common to a class of techniques based on diffusion MRI data aiming on tracking axonal pathways in the brain. Our approach, named Predictioncorrection Diffusion-based Tractography (PDT), is based on Extended Kalman Filtering: at each step the local fibers orientation is estimated from their orientation in the previous step. This estimate is then corrected from an estimate of the local diffusivity through a principled model of fibers orientation. PDT has been implemented using a diffusion tensor (DTI) as local diffusion model, but higher order models can be used as well. Results on both synthetic and in-vivo data are reported and discussed. PDT produces tractograms comparable to those obtained with the widely distributed tractography method provided in the FSL package [18], also in the case where crossing fibers are of relevance. From preliminary data, PDT proved superior when one fiber of low fractional anisotropy crosses a fiber with a higher fractional anisotropy, that is a critical condition for other tractography methods.