Design and study of flux-based features for 3D vascular tracking

Abstract

In this paper, we present and study two local features for the tracking of vascular structures on 3D angiograms. The first one, Flux, measures the inward gradient flux through circular cross-sections. The second one, MFlux, introduces a non-linear penalization of asymmetric flux contributions to reduce false positive responses.Through a series of experiments on synthetic and real cardiac CT data, we discuss the properties of these features with respect to their parameters. We compare them to a selection of published vesseldedicated features. We show that MFlux induces a particularly discriminative response landscape, which is a desirable property for tracking purposes on such large search spaces.A key characteristic of the proposed features is their simplicity of implementation and their high computational efficiency, enabling their practical use for advanced tracking strategies.

Hierarchical Discriminative Framework for Detecting Tubular Structures in 3D Images

“…These techniques use intensity values directly, or compute their first or second order derivatives at various locations of a sampling pattern [3,4,7,13]. Such approaches are typically not robust to noise, imaging artifacts, or high variability of tubular sizes and orientations.…”

“…The advantage of Haar features (apart from being efficient to compute) is that they form a basis and can approximate first and second order differential operators with only few basis elements. As such they can naturally mimic more complicated features such as Hessian or Flux [7,13]. Fig.…”

“…This produced voxel-wise classification of airways. We compared learning-based classification (using Haar, steerable [20], flux [7], and intensity Hessian features) and filter-based techniques (using flux filter [7] and Frangi filter [4]). Learning-based classification using Haar features performs the best.…”

“…Top-down approaches start at a root point and propagate the tree segmentation into distant branches, for example by region growing [2,9]. These algorithms obtain the segmentation by energy-based image filtering techniques [3,7,13] which evaluate manually tuned cost functions at certain image locations. Strong assumptions on the filter design (e.g.…”

Minimum Average-Cost Path for Real Time 3D Coronary Artery Segmentation of CT Images

“…The prior term P (X t | X t−1 ) in constrained radius variations is described in [5]. Observations Y t are the responses of a multi-scale oriented medialness feature and details can be found in [6]. …”