Development of 3D Carotid Ultrasound Segmentation, Registration and Quantification Techniques for Assessing and Monitoring of Carotid Atherosclerosis


Student thesis: Doctoral Thesis

View graph of relations


Related Research Unit(s)


Awarding Institution
Award date16 Oct 2017


Carotid atherosclerosis due to the buildup of plaque can lead to stroke, one of the leading causes of disability and death around the world. Monitoring the progression (or regression) of carotid atherosclerosis and quantifying its risk are of great importance for the diagnosis and prevention of stroke and for the evaluation of treatment options. 3D ultrasound has shown to be a reproducible and accurate noninvasive imaging tool widely used for the visualization, measurement and quantification of carotid plaque and for monitoring the progression/regression of atherosclerosis.

Total Plaque Volume (TPV) measured from 3D carotid ultrasound has been shown to be able to predict cardiovascular events and sensitive in assessing treatment effects. Conventional manual plaque segmentation is performed in previous studies to quantify TPV, but is tedious, requires long training times and is prone to observer variability. A direct 3D segmentation technique is proposed in this thesis to segment plaques from 3D carotid ultrasound images. This method incorporates modified initialization strategy and 3D level-set evolution, yielding more accurate and robust segmentation results compared to existing 2D segmentation method.

Besides the widely used metrics in current clinical trial such as Intima-Media Thickness (IMT), Total Plaque Area (TPA), Vessel Wall Volume (VWV) and Total Plaque Volume (TPV), accurate metrics or biomarkers which offer sufficient information about the local distribution of plaque changes and burdens in carotid arteries are in need. Motivated by this, a sensitive biomarker on the basis of 2D standardized vessel-wall-plus-plaque thickness change (VWT-Change) is proposed. This biomarker interprets hundreds or thousands of sample points from an information theory viewpoint. Experiment results demonstrated its capability for detecting statistically significant changes between subjects treated by Vitamin B and placebo compared with average VWT-Change for all points and partial points selected by feature selection.

The abovementioned metric requires the surfaces of outer wall and lumen in baseline and follow-up, which requires the manual segmentation of these surfaces. To avoid the issue in laborious and subjective manual segmentation, an image registration-based method is introduced to quantify the regions of local variation during time in this thesis. The proposed method makes full use of image intensity information to yield accurate image deformation results. A coarse-to-fine scheme is implemented to capture large nonlinear deformations caused by the regression under medicine treatment. In addition, an efficient dual-optimization based algorithm is then employed to compute an updated incremental deformation field at each resolution scale, which is also implemented using general-purpose programming on graphics processing units (GPGPU) to obtain high computational efficiency. Moreover, experimental results demonstrated that a significant improvement in accuracy over rigid registration was achieved by the proposed non-rigid registration method.