Automatic Fast and Reliable Recognition of a Small Brain White Matter Bundle

Abstract

Large diffusion MRI datasets provide the statistical power necessary to model complex effects in the white matter. They also motivate the need for fully automatic algorithms for finding white matter bundles. One popular algorithm, Automated Fiber Quantification (AFQ), has been shown to be reliable for analyzing a suite of large bundles. Here, we demonstrate that this approach can be extended to a relatively small white matter bundle, the optic tract. We develop an automated method that finds a portion of this bundle automatically. We compare the automatically found optic tract to optic tracts previously found using an algorithm that requires expert intervention, and find high degree of overlap. While previous methods work well in high-quality data, we demonstrate that the novel method proposed here generalizes broadly in subjects from three different datasets with differing data quality and a broad age range. Finally, we describe how this approach could be easily extended to other small bundles.