Surface Analysis and Template Construction for Carotid Arteries and Cerebral Lateral Ventricles


Student thesis: Doctoral Thesis

View graph of relations


  • Yimin CHEN

Related Research Unit(s)


Awarding Institution
Award date27 May 2016


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 important for the diagnosis and prevention of stroke and for the evaluation of treatment options. Magnetic resonance (MR) and ultrasound (US) imaging have been widely used to examine the 3D anatomical structure of the carotid artery.
The high resolution of MR imaging (MRI) enables high precision 3D models of the carotid artery to be reconstructed, and blood hemodynamic forces inside the lumen can be simulated by computational fluid dynamics (CFD) techniques to predict the risk of plaque vulnerability. However, CFD is computational intensive and time-consuming. To address this issue, we developed a technique to model wall shear stress (WSS) using geometric parameters that can be efficiently computed. Several parameters characterizing the 3D MR carotid lumen surfaces were proposed and correlated with WSS distribution. To facilitate visualization and inter-subject comparisons, 3D feature and WSS maps were flattened to standardized 2D maps to reveal more precise connections between geometric features and WSS.
3D ultrasound is a reproducible noninvasive imaging technology that has been used for the visualization, measurement and quantification of carotid plaque and for monitoring the progression/regression of atherosclerosis. Vessel-wall-plus-plaque thickness (VWT) was defined to be the point-by-point distance between the wall and the lumen, and the quantification of the change in the VWT (∆V W T ) between two scanning sessions allows longitudinal monitoring of spatial distribution of vessel wall and plaque progression/regression. However, since geometry of carotid arteries is highly subject-specific, geometric variability must be properly adjusted before an objective analysis of spatial distribution of plaque over a population of subjects is possible. Our group has previously described a technique to map VWT and ∆V W T into a flattened 2D standard domain to establish point-by-point correspondences between different subjects to allow for further inter-subject comparisons. However, the point correspondences were not optimized by any criterion in generating the 2D standardized map. An algorithm to optimize point correspondences based on minimization of the description length is proposed in this thesis. The results show that biomarkers that are more able to quantify effects of therapies can be obtained from the optimized standardized map.
Research related to surface analysis of the carotid artery was extended to the study of intraventricular hemorrhage (IVH), a common disease among low birth weight preterm neonates. Traditional 2D ultrasound imaging measures a selected slice of the ventricle and its application is limited to longitudinal monitoring of ventricle surface changes. 3D ultrasound has been used to quantify the volume of the ventricular system and may help clinicians to make better intervention decisions as it reveals the shape of the whole system. As manual segmentation of cerebral ventricles from 3D ultrasound images is laborious and time-consuming, automatic or semi-automatic segmentation algorithms are required. However, these algorithms must be validated before their clinical application. A framework is proposed to quantify and visualize the accuracy and variability of a semi-automatic segmentation algorithm of lateral ventricles from 3D ultrasound images. The application of this framework could be extended to analyzing local surface changes in lateral ventricles, which could potentially be correlated with deficits in the intellectual development of hydrocephalic patients. To further reduce inter-observer variability in segmentation, a fully automated segmentation approach was developed in this thesis and validated using the proposed statistical framework.