Adaptive intensity models for probabilistic tracking of 3D vasculature

The segmentation of vascular structures in 3D medical images is of great importance for many clinical applications, ranging from the detection and measurement of vascular disease to providing information for surgical intervention. Accurate and robust vascular segmentation is made difficult by variations in the vessel's contrast enhancement and its surrounding background, both within the same patient and across the patient population. This paper introduces a technique that first creates a patient-specific vessel intensity model and then adaptively varies this model during tracking as a function of vessel radius. The intensity model is used for estimating the likelihood of the observed intensity distribution within a sequential Monte Carlo tracking framework. We apply the proposed method to coronary artery segmentation from Computed Tomography Angiography 3D volumes and present results demonstrating the improved segmentation results achieved using the approach.