Segmentation And Elasticity Analysis Of Atherosclerotic Plaques In Carotid Contrast-enhanced Ultrasound Images

Carotid atherosclerosis is a type of cardiovascular and cerebrovascular disease. Clinically, theproperties of carotid atherosclerosis are mainly assessed by the artery stenosis, the intima-mediathickness, and the morphology, blood flow and elasticity of atherosclerotic plaques. Thevulnerability of plaques is a critical property of carotid atherosclerosis, and high vulnerability is amajor threat to human health. Studies have shown that the vulnerability of plaques is related withthe mechanical properties inside plaques, as well as the external stress outside plaques. A plaquewith high vulnerability, called a vulnerable plaque, often contains a large lipid core and a thinfibrous cap, while a plaque with low vulnerability, called a non-vulnerable plaque, usuallyincludes much fibre. If a plaque consists of a larger lipid core and a thinner fibrous cap, it will bemore likely to rupture, resulting in thrombosis. Meanwhile, the intraplaque neovascularization isalso associated with the vulnerability of plaques, which encourages the ongoing process ofatherosclerosis, resulting in plaque rupture and local thrombosis, and hence causing acutecardiovascular events. Therefore, the elasticity and neovascularization in plaques can be used toassess their vulnerability. Contrast-enhanced ultrasound (CEUS) introduces microbubbles ascontrast agents into the conventional ultrasound and it uses harmonic imaging, which can clearlydisplay the distribution of intraplaque neovascularization. Thus, CEUS provides a new technologyfor assessing the vulnerability of plaques. However, CEUS cannot evaluate the elasticity ofplaques, which restricts its application to deep investigation of plaques.CEUS is a real-time imaging modality, which dynamicly displays the morphology ofatherosclerotic plaques and the distribution of intraplaque neovascularization as a video. Theperiodic pulses lead to the deformation and strain of plaques in carotid artery. If the strain maps ofplaques can be estimated from the video sequnces of CEUS, then the elasticity of plaques can beextracted from the strain maps. So CEUS will become a more useful imaging modality thatprovides both blood flow and elasticity imaging. The technology of medical image processingmakes it possible to estimate strain and extract elastic features from plaques in the CEUSsequences. In this paper, we firstly locate plaques by using an image segmentation method, andthen compared two frames in a CEUS sequence to estimate the plaque strain and extract elasticfeatures. The work presented in this paper is mainly conducted from the following three aspects.Firstly, an improved method of multi-scale fuzzy clustering (MsFCM) is proposed forsegmentation of CEUS images. It uses the speckle reducing anisotropic diffusion filter to suppressspeckle noise in an image and build a series of multiscale images. The fuzzy clustering is carriedout in mutiscale images and the global cluster centers at each scale are found with the particleswarm optimization. The experimental results demonstrated the robustness and accuracy of theproposed method, which outperformed the traditional fuzzy clustering methods by26.6%and5.4%, in terms of the Pratt’s figure of merit and segmentation accuracy, respectively.Secondly, another method for segmentation of CEUS images is proposed by integrating theabovementioned improved MsFCM method with the B-spline directional gradient vector flowmodel. It overcomes two problems in CEUS image segmentation, i.e., the initialization of contoursand the interference of spurious edges near targets, and makes accurate detection for both the carotid lumen and plaque. This method uses the improved MsFCM to perform a coarsesegmentation and then adopts the B-spline directional gradient vector flow model for refinementof the segmentation. The experimental results showed that for segmentation of carotid lumen, theproposed method was superior to the traditional MsFCM method and the Chan-Vese level setmethod, increasing the root mean square error by19.8%and the false positive rate value by23.2%.For the segmentation of carotid plaques, the proposed method gets a specificity as high as97.8%,a true positive rate of83.2%, and a root mean square error of4.654pixels, demonstrating its highaccuracy.Thirdly, we used a free form deformation (FFD) model to extract elastic features of plaques.It firstly obtains the displacement filed from a pair of systolic and diastolic CEUS images usingthe FFD with B-spline method, then extracts elastic features from the displacement filed andfinally investigates the relationship between the elastic features and the manual grades ofintraplaque neovascularization. We extracted46elastic features from51samples of30patients,where17features existed significant difference (t-test, P <0.01) between two manual grades, i.e.,the high and low grades of plaques. The experimental results demonstrate that the elasticity ofplaques is associated with the intraplaque neovascularization, and the elastic features candistinguish the degree of neovascularization. This finding may be useful for further exploring therelationships between the intraplaque neovascularization, the mechanical properties and thevulnerability of plaques, thus leading to better recognization and treatment of vulnerable plaques.