Carotid plaque 3D compound imaging and echo-morphology analysis: a Bayesian approach

This paper describes a method for volume reconstruction of the carotid plaque and presents a novel local characterization of its echo-morphology. The data is composed by a series of nearly parallel ultrasound images (3D Compound Imaging) and the acquisition is performed using traditional non-invasive ultrasound equipment available in most medical facilities, without need of a spatial locator device. The reconstruction algorithm uses the observed pixels inside the plaque, which were obtained in a pre-segmentation stage performed under medical guidance [1]. The paper proposes a Bayesian algorithm which estimates the underlying volume inside the plaque, by filtering and interpolating the data in order to remove speckle noise and fill non-observed regions, respectively. This volume is further used in plaque echo-morphology analysis. The observation model is based on the Rayleigh distribution, commonly used to model speckle noise in ultrasound images. A prior model based on the edge preserving Total Variation Gibbs distribution is also used to fill the gaps on non-evenly spaced observations. An energy function is derived from these models and an iterative algorithm computes its minimizer. The estimated function, defined in a given volume of interest, is used in global and local plaque characterization, namely to estimate its average levels of stenosis, echo-morphology and to identify vulnerable foci inside the plaque. The goal is to make atherosclerosis diagnosis more accurate and complete than using traditional 2D ultrasound analysis.