A Hough transform global approach to diffusion MRI tractography

Abstract : Tractography in Diffusion-Weighted MRI provides a unique quantitative measurement of the brain's anatomical connectivity using information not available from other imaging techniques. Many tractography algorithms are based on local fiber orientation estimates, such as streamline methods, and are vulnerable to noise and partial volume effects; fiber crossing and kissing are also difficult to distinguish. This led to the development of probabilistic techniques [1] and global approaches relying on front propagation [2, 3] or simulation of the diffusion process [4]. In this work, we present a global approach based on the voting process provided by the Hough transform [5]. Our proposed tractography algorithm essentially tests all possible 3D curves in the volume, assigning a score to each of them, then selecting the curves with the highest scores, and returning them as the potential anatomical connections. We present experimental results on both artificial and real diffusion tensor images (DTI) and high-angular resolution diffusion images (HARDI).