A High-Dimensional Fiber Tract Atlas
O'Donnell L, Westin C
We present a method for creation of an atlas of major white matter fiber tracts. The method employs population clustering to generate fiber bundles, and can be applied in a straightforward way to label new subjects. We perform tractography in 5 DTI datasets then apply spectral clustering to the paths, producing population clusters in the high-dimensional spectral embedding space. These clusters are labeled with anatomical information and grouped into desired bundles, creating an atlas. We label the tracts of a new subject by embedding them into the atlas space using the Nystrom method, and assigning paths to the nearest cluster.