Shape Modelling for Tract Selection Jonathan D. Clayden, Martin D. King, and Chris A. Clark Institute of Child Health, University College London, UK j.clayden@ucl.ac.uk Abstract. Probabilistic tractography provides estimates of the proba- bility of a structural connection between points or regions in a brain volume, based on information from diffusion MRI. The ability to esti- mate the uncertainty associated with reconstructed pathways is valuable, but noise in the image data leads to premature termination or erroneous trajectories in sampled streamlines. In this work we describe automated methods, based on a probabilistic model of tract shape variability be- tween individuals, which can be applied to select seed points in order to maximise consistency in tract segmentation; and to discard stream- lines which are unlikely to belong to the tract of interest. Our method is shown to ameliorate false positives and remove the widely observed falloff in connection probability with distance from the seed region due to noise, two important problems in the tractography literature. More- over, the need to apply an arbitrary threshold to connection probability maps is entirely obviated by our approach, thus removing a significant user-specified parameter from the tractography pipeline. 1 Introduction Probabilistic tractography uses diffusion MRI (dMRI) data to provide estimates of the probability of a connection existing between a seed point, or seed region, and all other points within a brain volume. When the seed region is placed within a white matter tract, areas of high probability are typically found within other sections of the same tract. The first step towards estimating these proba- bilities of connection is to derive an orientation distribution function (ODF) for each voxel in the brain, which characterises the orientations of local structure. Several alternative methods for calculating such an ODF have been described [1], some of which are based on a specific model of diffusion, while others take a model-free approach. Probabilistic streamlines are then generated by alternating between sampling from these ODFs and stepping along the sampled direction. The probability of connection between the seed region and any other voxel is then estimated as the proportion of these streamlines that visit the target voxel. Unfortunately, the probabilities of connection estimated by this Monte Carlo method are strongly affected by nuisance characteristics of the basic data, par- ticularly noise, as well as limitations of the applied diffusion model. Streamlines may be deflected away from the tract of interest or prematurely truncated due to the nearby presence of a disparate tract, or due to ambiguity in the estimated G.-Z. Yang et al. (Eds.): MICCAI 2009, Part II, LNCS 5762, pp. 150–157, 2009. c  Springer-Verlag Berlin Heidelberg 2009