A scalable approach to streamline tractography clustering

Introduction Diffusion tractography has become an important method for assessing white matter (WM) structure. Depending on the application, the large number of streamlines typically produced by the tracking algorithm may make the dataset difficult to handle. As there will be many streamlines corresponding to each anatomical WM tract, it is often advantageous to group them into clusters in which all streamlines correspond to the same anatomical tract. In group studies, identifying tracts across subjects may be an additional objective. A common problem with clustering methods is their ability to scale, especially if data from multiple subjects is analysed. We present an approach based on repeated clustering of subsets that is conceptually transparent and that scales well to large tractography datasets and subject groups.