Reducing Model Uncertainty in Crossing Fiber Tractography

Diffusion MRI (dMRI) tractography permits the non-invasive reconstruction of major white matter tracts, and is therefore widely used in neurosurgical planning and in neuroscience. However, it is affected by various sources of uncertainty. In this work, we consider the model uncertainty that arises in crossing fiber tractography, from having to select between alternative mathematical models for the estimation of multiple fiber orientations in a given voxel. This type of model uncertainty is a source of instability in dMRI tractography that has not received much attention so far. We develop a mathematical framework to quantify it, based on computing posterior probabilities of competing models, given the local dMRI data. Moreover, we explore a novel strategy for crossing fiber tractography, which computes tracking directions from a consensus of multiple mathematical models, each one contributing with a weight that is proportional to its probability. Experiments on different white matter tracts in multiple subjects indicate that reducing model uncertainty in this way increases the accuracy of crossing fiber tractography. CCS Concepts • Applied computing → Life and medical sciences; • Mathematics of computing → Probabilistic algorithms; • Humancentered computing → Visualization techniques;