Edge detection in diffusion weighted MRI using a tangent curve similarity metric

Abstract

We present a technique to automatically characterize the geometry of important anatomical structures in diffusion weighted MRI (DWI) data. Our approach is based on the interpretation of diffusion data as a superimposition of multiple line fields that each form a continuum of space filling curves. Using a dense tractography computation, our method quantifies the spatial variations of the geometry of these curves and use the resulting measure to characterize salient structures as edges. Anatomically, these structures have a boundary-like nature and yield a clear picture of major fiber bundles. In particular, the application of our algorithm to high angular resolution imaging (HARDI) data yields a precise geometric description of subtle anatomical configurations associated with the local presence of multiple fiber orientations. We evaluate our technique and study its robustness to noise in the context of a phantom dataset and present results obtained with two diffusion weighted brain images.