We propose a Hessian matrix based multiscale tubular structure detection (TSD) algorithm adapted to 3D B-mode vascular US images. The algorithm is designed to highlight blood vessel centerline points and yield an estimate of the cross-section radius at each centerline point. It can be combined with a simple centerline extraction scheme, yielding precise, fast and fully automatic lumen segmentation initializations. TSD algorithms designed with CTA and MRA datasets in mind, e.g. the Frangi Filter, are not capable of reliably distinguishing centerline points from other points in vascular US datasets, since some assumptions underlying these algorithms are not reasonable for US datasets. The algorithm we propose, does not have these shortcomings and performs significantly better on vascular US datasets. We propose a statistic to evaluate how well a TSD algorithm is able to distinguish centerline points from other points. Based on this statistic, we compare the Frangi Filter to various versions of our new algorithm, on 11 3D US carotid datasets.     «

We propose a Hessian matrix based multiscale tubular structure detection (TSD) algorithm adapted to 3D B-mode vascular US images. The algorithm is designed to highlight blood vessel centerline points and yield an estimate of the cross-section radius at each centerline point. It can be combined with a simple centerline extraction scheme, yielding precise, fast and fully automatic lumen segmentation initializations. TSD algorithms designed with CTA and MRA datasets in mind, e.g. the Frangi Filter,...     »