Online Agglomerative Hierarchical Clustering Of Neural Fiber Tracts

We consider the problem of clustering neural fiber pathways, produced from diffusion MRI data via tractography, into different bundles. Existing clustering methods often suffer from the burden of computing pairwise fiber (dis) similarities, which escalates quadratically as the number of fiber pathways increases. To address this challenge, we adopt the scenario of clustering data streams into the fiber clustering framework. Specifically, we propose to use an online hierarchical clustering method, which yields a framework similar to doing clustering while simultaneously performing tractography. We evaluate the proposed method through experiments on phantom and real diffusion MRI data. Experiments on phantom data evaluate the sensitivity of our method to initialization and show its superior performance compared with alternative methods. Experiments on real data demonstrate the accuracy in clustering selected white matter fiber tracts into anatomically consistent bundles.