Nonparameteric vessel detection with applications to segmentation,
               registration and visualization of vascular images

                              Xenios Papademetris
           Depts of Diagnostic Radiology and Biomedical Engineering,
                                Yale University

Abstract
--------

I will describe our nonparametric method* for the detection of vascular
structures. The major weakness of most existing methods for vessel
detection/segmentation is the assumption that at each voxel there exists no
more than one cylindrical structure -- an assumption that is violated at vessel
branching points. Our proposed method uses a polar parameterization of the
local intensity to get around this problem and achieves improved performance
over standard methods. I will present results on both synthetic 2D images and
3D MRA animal vascular images. In addition, I will discuss more recent
extensions of this work for multiscale vessel detection, non-rigid registration
of vascular images, and vascular visualization.

* Qian et al, "A nonparametric vessel detection method for complex
vascular structures," Medical Image Analysis, 13(1):49--61, 2009