Tractography-embedded white matter stream clustering

While automated segmentation of white matter fibers is essential for understanding the human brain connectome, fast unsupervised clustering of these fibers emanating from a manually specified region of interest (ROI) into tracts is more desired in a clinical environment. In this work, we propose a tractography-embedded white matter stream clustering method to apply fiber tracking and clustering in a simultaneous manner. Integrated into a filtered tractography scheme, our method continuously checks for a drift in the fiber trajectories, which in turn controls the timing of the clustering. This affinity propagation-based clustering only involves a small portion of fibers and exemplars are selected to label the rest of the fibers. The proposed method is found to be five times faster than a traditional clustering framework, yet still achieves high accuracy on phantom and real data.