Robust RANSAC-based blood vessel segmentation

Many vascular clinical applications require a vessel segmentation process that is able to extract both the centerline and the surface of the blood vessels. However, noise and topology issues (such as kissing vessels) prevent existing algorithm from being able to easily retrieve such a complex system as the brain vasculature. We propose here a new blood vessel tracking algorithm that 1) detects the vessel centerline; 2) provides a local radius estimate; and 3) extracts a dense set of points at the blood vessel surface. This algorithm is based on a RANSAC-based robust fitting of successive cylinders along the vessel. Our method was validated against the Multiple Hypothesis Tracking (MHT) algorithm on 10 3DRA patient data of the brain vasculature. Over 744 blood vessels of various sizes were considered for each patient. Our results demonstrated a greater ability of our algorithm to track small, tortuous and touching vessels (96% success rate), compared to MHT (65% success rate). The computed centerline precision was below 1 voxel when compared to MHT. Moreover, our results were obtained with the same set of parameters for all patients and all blood vessels, except for the seed point for each vessel, also necessary for MHT. The proposed algorithm is thereafter able to extract the full intracranial vasculature with little user interaction. 1. DESCRIPTION OF PURPOSE The segmentation of vascular structures is particularly valuable for diagnosis assistance, treatment and surgery planning. A wide range of applications may benefit from an improved blood vessel segmentation process: quantitative studies of pathologies (e.g. stenoses), automated vascular navigation, accurate blood flow computation and computer-based simulations for surgeons education. In such cases, a mere surface depiction is not enough and the extraction of the vessel medial axes is also required. Vascular segmentation has resulted in a vast literature. Many previous works extract the vessel centerline tree, and claim to be robust to the kissing vessel issue: two vessels may happen to be locally tangent, or a vessel may run along a dense structure, e.g. an aneurism or bone. Such works address this issue by using a shape prior. But the downside of this assumption is that the vessel cross-section is circular, which is not always true, especially for large vessels. Thus, a trade-off is usually made between accuracy and robustness. Our work aims at preserving this desired robustness against noise and topology issues while not compromising accuracy on extraction of the vessel surface and its centerline. This paper presents a tracking procedure which builds a vessel tree through successive robust fitting of cylinders to the image. Thereby, our dedicated tracking process: delineates the centerline of the vessel, supplies a local estimation of its radius and robustly extracts dense points on the vessel surface. The reminder of this paper is organized as follows. Our RANSAC-Based Tracking (RBT) algorithm is introduced in Section 2.1. In Section 2.2, we describe the methodology of comparison between our proposal and the Multiple Hypothesis Tracking (MHT) algorithm. Section 3 discusses the results produced by both procedures on a data set of 10 patients, and conclusions are presented in Section 4.