Local tracing of curvilinear structures in volumetric color images: Application to the Brainbow analysis

In this study, we compare two vectorial tracing methods for 3D color images: (i) a conventional piecewise linear generalized cylinder algorithm that uses color and edge information and (ii) a principal curve tracing algorithm that uses the gradient and Hessian of a given density estimate. We tested the algorithms on synthetic and Brainbow dataset to show the effectiveness of the proposed algorithms. Results indicate that the proposed methods can successfully trace multiple axons in dense neighborhoods.

► Vectorial tracing methods can reliably be used for connectivity analysis of neuronal arbors.
► Local conditions for principal curves provides necessary conditions for the centerline detection.
► Subspace local maxima analysis can be used to trace the centerline.