Improving Multi-Tensor Fitting with Global Information from Track Orientation Density Imaging

The number of fascicles per-voxel N is a crucial parameter for multiple-fiber reconstruction methods from diffusion magnetic resonance imaging (dMRI) data, such as the multi-tensor model (MTM). This parameter is especially important when the goal is to provide bundle-specific tissue metrics. However, for MTM, statistical selection methods, such as the F-test, the Akaike and the Bayesian information criteria, tend to overestimate the number of tensors per-voxel for typical multi-shell dMRI acquisitions. In this work, we explore the reasons for this overestimation and propose to combine track orientation density imaging (TODI) with a robust MTM fitting framework to improve the estimation of N per-voxel. We conducted experiments on both synthetic and clinical in vivo dMRI data to validate our method, and we observed that tractography seems to be a useful spatial regularizer for N. Our results demonstrate the effectiveness of incorporating TODI for a more accurate and robust estimation of the intra-voxel number of fascicles.