A random effects modelling approach to the crossing-fibre problem in tractography

This paper examines a Bayesian random effects modelling approach to the analysis of multiple-directions diffusion-weighted MR data, with a focus on the crossing-fibre problem. Various models were investigated including a spatial (Markov random field) model, an exchangeable model and the Besag-York-Mollie model, which includes both exchangeable and spatial random effect terms. Each of these models was built around the diffusion-weighted signal intensity mixture model outlined in Behrens et al. (Behrens, T.E.J., Johansen Berg, H., Jbabdi, S., Rushworth, M.F.S., Woolrich, M.W., 2007. Probabilistic diffusion tractography with multiple fibre orientations: what can we gain? NeuroImage 34, 144-155.). The analyses were performed using Markov chain Monte Carlo simulation. Two regions were selected for investigation, both of which include distinct, non-collinear pathways in close proximity, resulting in crossing-fibre voxels. The first region includes the corpus callosum, the corona radiata and the superior longitudinal fasciculus. The second region is within the pons. Convincing fibre angular distributions were obtained using diffusion data generated with a low b-value (1000 s mm− 2) and restricted to 20 directions with only two acquisitions per direction. The results indicate that random effects modelling provides a useful alternative to current methods documented in the MR tractography literature.