A Unified Tractography Framework for Comparing Diffusion Models on Clinical Scans

In this paper, we compare several parametric and non-parametric models of diffusion in a unified framework that allows simultaneous model estimation and tractography. The framework uses the Unscented Kalman Filter (UKF) to compare several variants of spherical harmonics (SH), i.e., SH with sharpening, spherical deconvolution, SH with solid angle and also several parametric models like single and two-tensor models with and without an additional “free water” component. We estimate all these models and perform tractography using the same optimizer, namely, the UKF. Comparison is done by tracing two fiber bundles whose connectivity is known from the literature on human anatomy. We trace these fiber bundles on 10 healthy subjects and compare how well each of these models perform on clinical in-vivo scans. For quantitative comparison, we propose two new measures; traceability and coverage, both of which capture how well each method performs in terms of tracing known fiber bundles. Our results show that, the two-tensor with “free-water” model performs very well for both these measures.